[{"user_id":"2DF688A6-F248-11E8-B48F-1D18A9856A87","publication_status":"published","date_updated":"2026-06-03T05:53:30Z","article_processing_charge":"No","has_accepted_license":"1","alternative_title":["PMLR"],"department":[{"_id":"DaAl"}],"status":"public","OA_type":"diamond","tmp":{"legal_code_url":"https://creativecommons.org/licenses/by/4.0/legalcode","name":"Creative Commons Attribution 4.0 International Public License (CC-BY 4.0)","short":"CC BY (4.0)","image":"/images/cc_by.png"},"date_published":"2026-04-06T00:00:00Z","citation":{"ieee":"E. Schultheis and D.-A. Alistarh, “LLMQ: Efficient lower-precision LLM training for consumer GPUs,” in <i>2nd Conference on Parsimony and Learning</i>, Stanford, CA, United States, 2026, vol. 328, pp. 265–284.","short":"E. Schultheis, D.-A. Alistarh, in:, 2nd Conference on Parsimony and Learning, ML Research Press, 2026, pp. 265–284.","ama":"Schultheis E, Alistarh D-A. LLMQ: Efficient lower-precision LLM training for consumer GPUs. In: <i>2nd Conference on Parsimony and Learning</i>. Vol 328. ML Research Press; 2026:265-284.","ista":"Schultheis E, Alistarh D-A. 2026. LLMQ: Efficient lower-precision LLM training for consumer GPUs. 2nd Conference on Parsimony and Learning. CPAL: Conference on Parsimony and Learning, PMLR, vol. 328, 265–284.","chicago":"Schultheis, Erik, and Dan-Adrian Alistarh. “LLMQ: Efficient Lower-Precision LLM Training for Consumer GPUs.” In <i>2nd Conference on Parsimony and Learning</i>, 328:265–84. ML Research Press, 2026.","mla":"Schultheis, Erik, and Dan-Adrian Alistarh. “LLMQ: Efficient Lower-Precision LLM Training for Consumer GPUs.” <i>2nd Conference on Parsimony and Learning</i>, vol. 328, ML Research Press, 2026, pp. 265–84.","apa":"Schultheis, E., &#38; Alistarh, D.-A. (2026). LLMQ: Efficient lower-precision LLM training for consumer GPUs. In <i>2nd Conference on Parsimony and Learning</i> (Vol. 328, pp. 265–284). Stanford, CA, United States: ML Research Press."},"ddc":["000"],"oa_version":"Published Version","type":"conference","month":"04","abstract":[{"lang":"eng","text":"We present LLMQ, an end-to-end CUDA/C++ implementation for medium-sized language-model training, e.g. 3B to 32B parameters, on affordable, commodity GPUs. These devices are characterized by low memory availability and slow communication compared to datacentre-grade GPUs. Consequently, we showcase a range of optimizations that target these bottlenecks, including activation checkpointing, offloading, and copy-engine based collectives. LLMQ is able to train or fine-tune a 7B model on a single 16GB mid-range gaming card, or a 32B model on a workstation equipped with 4 RTX 4090s. This is achieved while executing a standard 8-bit training pipeline, without additional algorithmic approximations, and maintaining FLOP utilization of around 50%. The efficiency of LLMQ rivals that of production-scale systems on much more expensive cloud-grade GPUs."}],"related_material":{"link":[{"relation":"software","url":"https://github.com/IST-DASLab/llmq"}]},"acknowledgement":"We would like to thank contacts at NVIDIA (Vartika Singh, Nina Carrejo, Kyla Wilkes, and Tijmen\r\nBlankevoort), HP (Curtis Burkhalter), and Datacrunch/Verda (Paul Chang and Antonio\r\nDominguez) for hardware support that was essential to this project. ES was supported in part\r\nby ERC Proof-of-Concept grant FastML.","day":"06","volume":328,"publication":"2nd Conference on Parsimony and Learning","conference":{"name":"CPAL: Conference on Parsimony and Learning","location":"Stanford, CA, United States","start_date":"2025-03-24","end_date":"2025-03-27"},"oa":1,"intvolume":"       328","year":"2026","OA_place":"publisher","corr_author":"1","language":[{"iso":"eng"}],"file_date_updated":"2026-06-03T05:51:19Z","_id":"21932","page":"265-284","date_created":"2026-05-31T22:02:13Z","title":"LLMQ: Efficient lower-precision LLM training for consumer GPUs","quality_controlled":"1","author":[{"id":"2786b299-e6b0-11f0-91da-9243fe3ef96b","last_name":"Schultheis","full_name":"Schultheis, Erik","first_name":"Erik"},{"full_name":"Alistarh, Dan-Adrian","last_name":"Alistarh","id":"4A899BFC-F248-11E8-B48F-1D18A9856A87","orcid":"0000-0003-3650-940X","first_name":"Dan-Adrian"}],"publication_identifier":{"eissn":["2640-3498"]},"publisher":"ML Research Press","file":[{"file_name":"2026_CPAL_Schultheis.pdf","date_created":"2026-06-03T05:51:19Z","date_updated":"2026-06-03T05:51:19Z","content_type":"application/pdf","file_size":2099944,"access_level":"open_access","file_id":"21942","relation":"main_file","creator":"dernst","checksum":"72f9a87c70f1e2105ef64050ee5017e5","success":1}],"scopus_import":"1"},{"day":"17","publication":"Medical Imaging with Deep Learning","volume":315,"conference":{"end_date":"2026-07-10","start_date":"2026-07-08","location":"Taipei, Taiwan","name":"MIDL: Medical Imaging with Deep Learning"},"oa":1,"intvolume":"       315","year":"2026","OA_place":"publisher","corr_author":"1","language":[{"iso":"eng"}],"_id":"21949","page":"1969-1982","date_created":"2026-06-07T22:01:36Z","title":"T1-PILOT: Physics-informed learned optimized trajectories for T1 mapping acceleration","quality_controlled":"1","author":[{"full_name":"Shor, Tamir","last_name":"Shor","first_name":"Tamir"},{"first_name":"Moti","full_name":"Freiman, Moti","last_name":"Freiman"},{"full_name":"Baskin, Chaim","last_name":"Baskin","first_name":"Chaim"},{"orcid":"0000-0001-9699-8730","first_name":"Alexander","last_name":"Bronstein","full_name":"Bronstein, Alexander","id":"58f3726e-7cba-11ef-ad8b-e6e8cb3904e6"}],"publication_identifier":{"eissn":["2640-3498"]},"publisher":"ML Research Press","scopus_import":"1","keyword":["Cardiac T1 Mapping","Trajectory Optimization and Reconstruction","PhysicsInformed Deep-Learning"],"user_id":"2DF688A6-F248-11E8-B48F-1D18A9856A87","publication_status":"accepted","date_updated":"2026-06-08T08:05:24Z","article_processing_charge":"No","has_accepted_license":"1","department":[{"_id":"AlBr"}],"alternative_title":["PMLR"],"status":"public","main_file_link":[{"open_access":"1","url":"https://openreview.net/forum?id=nZaPtHbd6N#discussion"}],"OA_type":"gold","date_published":"2026-03-17T00:00:00Z","tmp":{"legal_code_url":"https://creativecommons.org/licenses/by/4.0/legalcode","name":"Creative Commons Attribution 4.0 International Public License (CC-BY 4.0)","short":"CC BY (4.0)","image":"/images/cc_by.png"},"ddc":["000"],"citation":{"ista":"Shor T, Freiman M, Baskin C, Bronstein AM. T1-PILOT: Physics-informed learned optimized trajectories for T1 mapping acceleration. Medical Imaging with Deep Learning. MIDL: Medical Imaging with Deep Learning, PMLR, vol. 315, 1969–1982.","ama":"Shor T, Freiman M, Baskin C, Bronstein AM. T1-PILOT: Physics-informed learned optimized trajectories for T1 mapping acceleration. In: <i>Medical Imaging with Deep Learning</i>. Vol 315. ML Research Press; :1969-1982.","chicago":"Shor, Tamir, Moti Freiman, Chaim Baskin, and Alex M. Bronstein. “T1-PILOT: Physics-Informed Learned Optimized Trajectories for T1 Mapping Acceleration.” In <i>Medical Imaging with Deep Learning</i>, 315:1969–82. ML Research Press, n.d.","short":"T. Shor, M. Freiman, C. Baskin, A.M. Bronstein, in:, Medical Imaging with Deep Learning, ML Research Press, n.d., pp. 1969–1982.","ieee":"T. Shor, M. Freiman, C. Baskin, and A. M. Bronstein, “T1-PILOT: Physics-informed learned optimized trajectories for T1 mapping acceleration,” in <i>Medical Imaging with Deep Learning</i>, Taipei, Taiwan, vol. 315, pp. 1969–1982.","apa":"Shor, T., Freiman, M., Baskin, C., &#38; Bronstein, A. M. (n.d.). T1-PILOT: Physics-informed learned optimized trajectories for T1 mapping acceleration. In <i>Medical Imaging with Deep Learning</i> (Vol. 315, pp. 1969–1982). Taipei, Taiwan: ML Research Press.","mla":"Shor, Tamir, et al. “T1-PILOT: Physics-Informed Learned Optimized Trajectories for T1 Mapping Acceleration.” <i>Medical Imaging with Deep Learning</i>, vol. 315, ML Research Press, pp. 1969–82."},"oa_version":"Published Version","type":"conference","month":"03","abstract":[{"lang":"eng","text":"Cardiac T1 mapping provides critical quantitative insights into myocardial tissue composition, enabling the assessment of pathologies such as fibrosis, inflammation, and edema.\r\nHowever, the inherently dynamic nature of the heart imposes strict limits on acquisition\r\ntimes, making high-resolution T1 mapping a persistent challenge. Compressed sensing (CS)\r\napproaches have reduced scan durations by undersampling k-space and reconstructing images from partial data, and recent studies show that jointly optimizing the undersampling\r\npatterns with the reconstruction network can substantially improve performance. Still,\r\nmost current T1 mapping pipelines rely on static, hand-crafted masks that do not exploit\r\nthe full acceleration and accuracy potential. Furthermore, most existing methods do not\r\nlevarage the physical T1 decay model in optimization. In this work, we introduce T1-\r\nPILOT: an end-to-end method that explicitly incorporates the T1 signal relaxation model\r\ninto the sampling–reconstruction framework to guide the learning of non-Cartesian trajectories, cross-frame alignment, and T1 decay estimation. Through extensive experiments\r\non the CMRxRecon dataset, T1-PILOT significantly outperforms several baseline strategies (including learned single-mask and fixed radial or golden-angle sampling schemes),\r\nachieving higher T1 map fidelity at greater acceleration factors. In particular, we observe consistent gains in PSNR and VIF relative to existing methods, along with marked\r\nimprovements in delineating finer myocardial structures. Our results highlight that optimizing sampling trajectories in tandem with the physical relaxation model leads to both\r\nenhanced quantitative accuracy and reduced acquisition times. Code for reproducing all\r\nexperiments and results is available at https://github.com/tamirshor7/T1-PILOT"}],"related_material":{"link":[{"relation":"software","url":"https://github.com/tamirshor7/T1-PILOT"}]}},{"publication_identifier":{"eissn":["2640-3498"]},"publisher":"ML Research Press","file":[{"creator":"dernst","relation":"main_file","access_level":"open_access","file_id":"20828","checksum":"1d744fbaeb199b08e8b6f48bc0dd047e","success":1,"file_size":908379,"date_created":"2025-12-16T12:32:40Z","file_name":"2025_ICML_Sieberling.pdf","date_updated":"2025-12-16T12:32:40Z","content_type":"application/pdf"}],"scopus_import":"1","title":"EvoPress: Accurate dynamic model compression via evolutionary search","quality_controlled":"1","author":[{"full_name":"Sieberling, Oliver","last_name":"Sieberling","first_name":"Oliver"},{"last_name":"Kuznedelev","full_name":"Kuznedelev, Denis","first_name":"Denis"},{"last_name":"Kurtic","full_name":"Kurtic, Eldar","id":"47beb3a5-07b5-11eb-9b87-b108ec578218","first_name":"Eldar"},{"id":"4A899BFC-F248-11E8-B48F-1D18A9856A87","last_name":"Alistarh","full_name":"Alistarh, Dan-Adrian","first_name":"Dan-Adrian","orcid":"0000-0003-3650-940X"}],"file_date_updated":"2025-12-16T12:32:40Z","_id":"20820","date_created":"2025-12-14T23:02:05Z","page":"55556-55590","year":"2025","corr_author":"1","OA_place":"publisher","language":[{"iso":"eng"}],"conference":{"name":"ICML: International Conference on Machine Learning","location":"Vancouver, Canada","start_date":"2025-07-13","end_date":"2025-07-19"},"oa":1,"intvolume":"       267","volume":267,"publication":"42nd International Conference on Machine Learning","day":"01","arxiv":1,"month":"05","abstract":[{"lang":"eng","text":"The high computational costs of large language models (LLMs) have led to a flurry of research on LLM compression, via methods such as quantization, sparsification, or structured pruning. A new frontier in this area is given by dynamic, non-uniform compression methods, which adjust the compression levels (e.g., sparsity) per-block or even per-layer in order to minimize accuracy loss, while guaranteeing a global compression threshold. Yet, current methods rely on estimating the \"importance\" of a given layer, implicitly assuming that layers contribute independently to the overall compression error. We begin from the motivating observation that this independence assumption does not generally hold for LLM compression: pruning a model further may even significantly recover performance. To address this, we propose EvoPress, a novel evolutionary framework for dynamic LLM compression. By formulating dynamic compression as a general optimization problem, EvoPress identifies optimal compression profiles in a highly efficient manner, and generalizes across diverse models and compression techniques. Via EvoPress, we achieve state-of-the-art performance for dynamic compression of Llama, Mistral, and Phi models, setting new benchmarks for structural pruning (block/layer dropping), unstructured sparsity, and quantization with dynamic bitwidths."}],"oa_version":"Published Version","type":"conference","external_id":{"arxiv":["2410.14649"]},"tmp":{"legal_code_url":"https://creativecommons.org/licenses/by/4.0/legalcode","name":"Creative Commons Attribution 4.0 International Public License (CC-BY 4.0)","short":"CC BY (4.0)","image":"/images/cc_by.png"},"date_published":"2025-05-01T00:00:00Z","citation":{"ama":"Sieberling O, Kuznedelev D, Kurtic E, Alistarh D-A. EvoPress: Accurate dynamic model compression via evolutionary search. In: <i>42nd International Conference on Machine Learning</i>. Vol 267. ML Research Press; 2025:55556-55590.","chicago":"Sieberling, Oliver, Denis Kuznedelev, Eldar Kurtic, and Dan-Adrian Alistarh. “EvoPress: Accurate Dynamic Model Compression via Evolutionary Search.” In <i>42nd International Conference on Machine Learning</i>, 267:55556–90. ML Research Press, 2025.","ista":"Sieberling O, Kuznedelev D, Kurtic E, Alistarh D-A. 2025. EvoPress: Accurate dynamic model compression via evolutionary search. 42nd International Conference on Machine Learning. ICML: International Conference on Machine Learning, PMLR, vol. 267, 55556–55590.","ieee":"O. Sieberling, D. Kuznedelev, E. Kurtic, and D.-A. Alistarh, “EvoPress: Accurate dynamic model compression via evolutionary search,” in <i>42nd International Conference on Machine Learning</i>, Vancouver, Canada, 2025, vol. 267, pp. 55556–55590.","short":"O. Sieberling, D. Kuznedelev, E. Kurtic, D.-A. Alistarh, in:, 42nd International Conference on Machine Learning, ML Research Press, 2025, pp. 55556–55590.","mla":"Sieberling, Oliver, et al. “EvoPress: Accurate Dynamic Model Compression via Evolutionary Search.” <i>42nd International Conference on Machine Learning</i>, vol. 267, ML Research Press, 2025, pp. 55556–90.","apa":"Sieberling, O., Kuznedelev, D., Kurtic, E., &#38; Alistarh, D.-A. (2025). EvoPress: Accurate dynamic model compression via evolutionary search. In <i>42nd International Conference on Machine Learning</i> (Vol. 267, pp. 55556–55590). Vancouver, Canada: ML Research Press."},"ddc":["000"],"status":"public","OA_type":"gold","article_processing_charge":"No","has_accepted_license":"1","alternative_title":["PMLR"],"department":[{"_id":"DaAl"}],"user_id":"2DF688A6-F248-11E8-B48F-1D18A9856A87","publication_status":"published","date_updated":"2025-12-16T12:34:32Z"},{"abstract":[{"lang":"eng","text":"Modern deep neural networks exhibit heterogeneity across numerous layers of various types such as residuals, multi-head attention, etc., due to varying structures (dimensions, activation functions, etc.), distinct representation characteristics, which impact predictions. We develop a general layer-wise quantization framework with tight variance and code-length bounds, adapting to the heterogeneities over the course of training. We then apply a new layer-wise quantization technique within distributed variational inequalities (VIs), proposing a novel Quantized Optimistic Dual Averaging (QODA) algorithm with adaptive learning rates, which achieves competitive convergence rates for monotone VIs. We empirically show that QODA achieves up to a 150% speedup over the baselines in end-to-end training time for training Wasserstein GAN on 12+GPUs."}],"month":"05","type":"conference","external_id":{"arxiv":["2505.14371"]},"oa_version":"Published Version","citation":{"ama":"Nguyen AD, Markov I, Wu FZ, et al. Layer-wise quantization for quantized optimistic dual averaging. In: <i>42nd International Conference on Machine Learning</i>. Vol 267. ML Research Press; 2025:46026-46072.","chicago":"Nguyen, Anh Duc, Ilia Markov, Frank Zhengqing Wu, Ali Ramezani-Kebrya, Kimon Antonakopoulos, Dan-Adrian Alistarh, and Volkan Cevher. “Layer-Wise Quantization for Quantized Optimistic Dual Averaging.” In <i>42nd International Conference on Machine Learning</i>, 267:46026–72. ML Research Press, 2025.","ista":"Nguyen AD, Markov I, Wu FZ, Ramezani-Kebrya A, Antonakopoulos K, Alistarh D-A, Cevher V. 2025. Layer-wise quantization for quantized optimistic dual averaging. 42nd International Conference on Machine Learning. ICML: International Conference on Machine Learning, PMLR, vol. 267, 46026–46072.","ieee":"A. D. Nguyen <i>et al.</i>, “Layer-wise quantization for quantized optimistic dual averaging,” in <i>42nd International Conference on Machine Learning</i>, Vancouver, Canada, 2025, vol. 267, pp. 46026–46072.","short":"A.D. Nguyen, I. Markov, F.Z. Wu, A. Ramezani-Kebrya, K. Antonakopoulos, D.-A. Alistarh, V. Cevher, in:, 42nd International Conference on Machine Learning, ML Research Press, 2025, pp. 46026–46072.","apa":"Nguyen, A. D., Markov, I., Wu, F. Z., Ramezani-Kebrya, A., Antonakopoulos, K., Alistarh, D.-A., &#38; Cevher, V. (2025). Layer-wise quantization for quantized optimistic dual averaging. In <i>42nd International Conference on Machine Learning</i> (Vol. 267, pp. 46026–46072). Vancouver, Canada: ML Research Press.","mla":"Nguyen, Anh Duc, et al. “Layer-Wise Quantization for Quantized Optimistic Dual Averaging.” <i>42nd International Conference on Machine Learning</i>, vol. 267, ML Research Press, 2025, pp. 46026–72."},"ddc":["000"],"date_published":"2025-05-01T00:00:00Z","tmp":{"legal_code_url":"https://creativecommons.org/licenses/by/4.0/legalcode","name":"Creative Commons Attribution 4.0 International Public License (CC-BY 4.0)","short":"CC BY (4.0)","image":"/images/cc_by.png"},"OA_type":"gold","project":[{"grant_number":"101158077","name":"FastML: Efficient and Cost-Effective Distributed Machine Learning","_id":"8e35c14b-16d5-11f0-9cad-a3fc35339161"}],"status":"public","alternative_title":["PMLR"],"department":[{"_id":"DaAl"}],"has_accepted_license":"1","article_processing_charge":"No","date_updated":"2025-12-16T12:46:54Z","publication_status":"published","user_id":"2DF688A6-F248-11E8-B48F-1D18A9856A87","scopus_import":"1","file":[{"creator":"dernst","relation":"main_file","file_id":"20830","access_level":"open_access","success":1,"checksum":"a7edf0e4304171a3e035842b3aab1704","file_size":756213,"content_type":"application/pdf","file_name":"2025_ICML_Nguyen.pdf","date_created":"2025-12-16T12:45:41Z","date_updated":"2025-12-16T12:45:41Z"}],"publisher":"ML Research Press","publication_identifier":{"eissn":["2640-3498"]},"author":[{"first_name":"Anh Duc","last_name":"Nguyen","full_name":"Nguyen, Anh Duc"},{"first_name":"Ilia","full_name":"Markov, Ilia","id":"D0CF4148-C985-11E9-8066-0BDEE5697425","last_name":"Markov"},{"last_name":"Wu","full_name":"Wu, Frank Zhengqing","first_name":"Frank Zhengqing"},{"first_name":"Ali","full_name":"Ramezani-Kebrya, Ali","last_name":"Ramezani-Kebrya"},{"first_name":"Kimon","full_name":"Antonakopoulos, Kimon","last_name":"Antonakopoulos"},{"orcid":"0000-0003-3650-940X","first_name":"Dan-Adrian","last_name":"Alistarh","full_name":"Alistarh, Dan-Adrian","id":"4A899BFC-F248-11E8-B48F-1D18A9856A87"},{"first_name":"Volkan","last_name":"Cevher","full_name":"Cevher, Volkan"}],"quality_controlled":"1","title":"Layer-wise quantization for quantized optimistic dual averaging","_id":"20821","date_created":"2025-12-14T23:02:06Z","page":"46026-46072","file_date_updated":"2025-12-16T12:45:41Z","language":[{"iso":"eng"}],"OA_place":"publisher","year":"2025","intvolume":"       267","oa":1,"conference":{"name":"ICML: International Conference on Machine Learning","location":"Vancouver, Canada","start_date":"2025-07-13","end_date":"2025-07-19"},"volume":267,"publication":"42nd International Conference on Machine Learning","day":"01","arxiv":1,"acknowledgement":"This work was supported by Hasler Foundation Program: Hasler Responsible AI (project number 21043). The research was also sponsored by the Army Research Office and was accomplished under Grant Number W911NF-24-1-0048. This work was further funded by the Swiss National Science Foundation (SNSF) under grant number 200021_205011. We also acknowledge project A11 of the Swiss National Supercomputing Centre (CSCS) for providing computing resources. Dan Alistarh and Ilia Markov were supported in part through the ERC Proofof-Concept grant FastML (Grant Agreement 101158077). Ali Ramezani-Kebrya was supported by the Research Council of Norway through FRIPRO Grant under project number 356103, its Centres of Excellence scheme, Integreat - Norwegian Centre for knowledge-driven machine learning under\r\nproject number 332645 - and its Centre for Research-based Innovation funding scheme (Visual Intelligence under grant no. 309439)."},{"ddc":["000"],"citation":{"chicago":"Montagna, Francesco, Philipp Faller, Patrik Blöbaum, Elke Kirschbaum, and Francesco Locatello. “Score Matching through the Roof: Linear, Nonlinear, and Latent Variables Causal Discovery.” In <i>Proceedings of the Fourth Conference on Causal Learning and Reasoning</i>, 275:552–605. ML Research Press, 2025.","ama":"Montagna F, Faller P, Blöbaum P, Kirschbaum E, Locatello F. Score matching through the roof: Linear, nonlinear, and latent variables causal discovery. In: <i>Proceedings of the Fourth Conference on Causal Learning and Reasoning</i>. Vol 275. ML Research Press; 2025:552-605.","ista":"Montagna F, Faller P, Blöbaum P, Kirschbaum E, Locatello F. 2025. Score matching through the roof: Linear, nonlinear, and latent variables causal discovery. Proceedings of the Fourth Conference on Causal Learning and Reasoning. CLeaR: Conference on Causal Learning and Reasoning, PMLR, vol. 275, 552–605.","short":"F. Montagna, P. Faller, P. Blöbaum, E. Kirschbaum, F. Locatello, in:, Proceedings of the Fourth Conference on Causal Learning and Reasoning, ML Research Press, 2025, pp. 552–605.","ieee":"F. Montagna, P. Faller, P. Blöbaum, E. Kirschbaum, and F. Locatello, “Score matching through the roof: Linear, nonlinear, and latent variables causal discovery,” in <i>Proceedings of the Fourth Conference on Causal Learning and Reasoning</i>, Lausanne, Switzerland, 2025, vol. 275, pp. 552–605.","mla":"Montagna, Francesco, et al. “Score Matching through the Roof: Linear, Nonlinear, and Latent Variables Causal Discovery.” <i>Proceedings of the Fourth Conference on Causal Learning and Reasoning</i>, vol. 275, ML Research Press, 2025, pp. 552–605.","apa":"Montagna, F., Faller, P., Blöbaum, P., Kirschbaum, E., &#38; Locatello, F. (2025). Score matching through the roof: Linear, nonlinear, and latent variables causal discovery. In <i>Proceedings of the Fourth Conference on Causal Learning and Reasoning</i> (Vol. 275, pp. 552–605). Lausanne, Switzerland: ML Research Press."},"date_published":"2025-05-01T00:00:00Z","tmp":{"legal_code_url":"https://creativecommons.org/licenses/by/4.0/legalcode","name":"Creative Commons Attribution 4.0 International Public License (CC-BY 4.0)","short":"CC BY (4.0)","image":"/images/cc_by.png"},"OA_type":"gold","status":"public","main_file_link":[{"open_access":"1","url":"https://proceedings.mlr.press/v275/montagna25a.html"}],"abstract":[{"text":"Causal discovery from observational data holds great promise, but existing methods rely on strong assumptions about the underlying causal structure, often requiring full observability of all relevant variables. We tackle these challenges by leveraging the score function ∇logp(X)\r\n of observed variables for causal discovery and propose the following contributions. First, we generalize the existing results of identifiability with the score to additive noise models with minimal requirements on the causal mechanisms. Second, we establish conditions for inferring causal relations from the score even in the presence of hidden variables; this result is two-faced: we demonstrate the score’s potential as an alternative to conditional independence tests to infer the equivalence class of causal graphs with hidden variables, and we provide the necessary conditions for identifying direct causes in latent variable models. Building on these insights, we propose a flexible algorithm for causal discovery across linear, nonlinear, and latent variable models, which we empirically validate.","lang":"eng"}],"month":"05","external_id":{"arxiv":["2407.18755"]},"type":"conference","oa_version":"Published Version","publication_status":"published","date_updated":"2026-02-10T11:54:02Z","user_id":"2DF688A6-F248-11E8-B48F-1D18A9856A87","alternative_title":["PMLR"],"department":[{"_id":"FrLo"}],"article_processing_charge":"No","has_accepted_license":"1","file_date_updated":"2026-01-29T14:17:48Z","_id":"21066","date_created":"2026-01-29T14:19:09Z","page":"552-605","file":[{"content_type":"application/pdf","date_updated":"2026-01-29T14:17:48Z","file_name":"montagna25a.pdf","date_created":"2026-01-29T14:17:48Z","file_size":1739334,"file_id":"21067","access_level":"open_access","creator":"flocatel","relation":"main_file","success":1,"checksum":"f2bc44b2320667d4049b3518b1f2fe5d"}],"publication_identifier":{"eissn":["2640-3498"]},"publisher":"ML Research Press","author":[{"first_name":"Francesco","last_name":"Montagna","full_name":"Montagna, Francesco"},{"full_name":"Faller, Philipp","last_name":"Faller","first_name":"Philipp"},{"first_name":"Patrik","last_name":"Blöbaum","full_name":"Blöbaum, Patrik"},{"full_name":"Kirschbaum, Elke","last_name":"Kirschbaum","first_name":"Elke"},{"last_name":"Locatello","id":"26cfd52f-2483-11ee-8040-88983bcc06d4","full_name":"Locatello, Francesco","orcid":"0000-0002-4850-0683","first_name":"Francesco"}],"title":"Score matching through the roof: Linear, nonlinear, and latent variables causal discovery","quality_controlled":"1","volume":275,"publication":"Proceedings of the Fourth Conference on Causal Learning and Reasoning","acknowledgement":"Philipp M. Faller was supported by a doctoral scholarship of the Studienstiftung des deutschen\r\nVolkes (German Academic Scholarship Foundation). This work has been supported by AFOSR,\r\ngrant n. FA8655-20-1-7035. FM is supported by Programma Operativo Nazionale ricerca e innovazione 2014-2020. We thank Atalanti A. Mastakouri, Kun Zhang and Haoyue Dai for the insightful discussions.","day":"01","arxiv":1,"OA_place":"publisher","corr_author":"1","language":[{"iso":"eng"}],"year":"2025","intvolume":"       275","conference":{"name":"CLeaR: Conference on Causal Learning and Reasoning","location":"Lausanne, Switzerland","start_date":"2025-05-07","end_date":"2025-05-09"},"oa":1},{"month":"11","abstract":[{"lang":"eng","text":"Test-time training (TTT) methods explicitly update the weights of a model to adapt to the specific test instance, and they have found success in a variety of settings, including most recently language modeling and reasoning. To demystify this success, we investigate a gradient-based TTT algorithm for in-context learning, where we train a transformer model on the in-context demonstrations provided in the test prompt. Specifically, we provide a comprehensive theoretical characterization of linear transformers when the update rule is a single gradient step. Our theory (i) delineates the role of alignment between pretraining distribution and target task, (ii) demystifies how TTT can alleviate distribution shift, and (iii) quantifies the sample complexity of TTT including how it can significantly reduce the eventual sample size required for in-context learning. As our empirical contribution, we study the benefits of TTT for TabPFN, a tabular foundation model. In line with our theory, we demonstrate that TTT significantly reduces the required sample size for tabular classification (3 to 5 times fewer) unlocking substantial inference efficiency with a negligible training cost."}],"oa_version":"Published Version","type":"conference","external_id":{"pmid":["41321376"]},"tmp":{"legal_code_url":"https://creativecommons.org/licenses/by/4.0/legalcode","name":"Creative Commons Attribution 4.0 International Public License (CC-BY 4.0)","short":"CC BY (4.0)","image":"/images/cc_by.png"},"date_published":"2025-11-30T00:00:00Z","citation":{"ieee":"H. A. Gozeten, M. E. Ildiz, X. Zhang, M. Soltanolkotabi, M. Mondelli, and S. Oymak, “Test-time training provably improves transformers as in-context learners,” in <i>Proceedings of the 42nd International Conference on Machine Learning</i>, Vancouver, Canada, 2025, vol. 267, pp. 20266–20295.","short":"H.A. Gozeten, M.E. Ildiz, X. Zhang, M. Soltanolkotabi, M. Mondelli, S. Oymak, in:, Proceedings of the 42nd International Conference on Machine Learning, ML Research Press, 2025, pp. 20266–20295.","ista":"Gozeten HA, Ildiz ME, Zhang X, Soltanolkotabi M, Mondelli M, Oymak S. 2025. Test-time training provably improves transformers as in-context learners. Proceedings of the 42nd International Conference on Machine Learning. ICML: International Conference on Machine Learning, PMLR, vol. 267, 20266–20295.","ama":"Gozeten HA, Ildiz ME, Zhang X, Soltanolkotabi M, Mondelli M, Oymak S. Test-time training provably improves transformers as in-context learners. In: <i>Proceedings of the 42nd International Conference on Machine Learning</i>. Vol 267. ML Research Press; 2025:20266-20295.","chicago":"Gozeten, Halil Alperen, Muhammed Emrullah Ildiz, Xuechen Zhang, Mahdi Soltanolkotabi, Marco Mondelli, and Samet Oymak. “Test-Time Training Provably Improves Transformers as in-Context Learners.” In <i>Proceedings of the 42nd International Conference on Machine Learning</i>, 267:20266–95. ML Research Press, 2025.","mla":"Gozeten, Halil Alperen, et al. “Test-Time Training Provably Improves Transformers as in-Context Learners.” <i>Proceedings of the 42nd International Conference on Machine Learning</i>, vol. 267, ML Research Press, 2025, pp. 20266–95.","apa":"Gozeten, H. A., Ildiz, M. E., Zhang, X., Soltanolkotabi, M., Mondelli, M., &#38; Oymak, S. (2025). Test-time training provably improves transformers as in-context learners. In <i>Proceedings of the 42nd International Conference on Machine Learning</i> (Vol. 267, pp. 20266–20295). Vancouver, Canada: ML Research Press."},"ddc":["000"],"status":"public","project":[{"_id":"911e6d1f-16d5-11f0-9cad-c5c68c6a1cdf","grant_number":"101161364","name":"Inference in High Dimensions: Light-speed Algorithms and Information Limits"}],"OA_type":"gold","pmid":1,"article_processing_charge":"No","has_accepted_license":"1","department":[{"_id":"MaMo"}],"alternative_title":["PMLR"],"user_id":"2DF688A6-F248-11E8-B48F-1D18A9856A87","publication_status":"published","date_updated":"2026-02-19T08:18:24Z","publication_identifier":{"eissn":["2640-3498"]},"publisher":"ML Research Press","file":[{"checksum":"f774f8619a0d72f3975d9cb23942a1e9","success":1,"access_level":"open_access","file_id":"21336","creator":"dernst","relation":"main_file","date_created":"2026-02-19T08:15:48Z","file_name":"2025_ICML_Gozeten.pdf","date_updated":"2026-02-19T08:15:48Z","content_type":"application/pdf","file_size":471176}],"title":"Test-time training provably improves transformers as in-context learners","quality_controlled":"1","author":[{"first_name":"Halil Alperen","full_name":"Gozeten, Halil Alperen","last_name":"Gozeten"},{"first_name":"Muhammed Emrullah","last_name":"Ildiz","full_name":"Ildiz, Muhammed Emrullah"},{"first_name":"Xuechen","full_name":"Zhang, Xuechen","last_name":"Zhang"},{"first_name":"Mahdi","full_name":"Soltanolkotabi, Mahdi","last_name":"Soltanolkotabi"},{"last_name":"Mondelli","id":"27EB676C-8706-11E9-9510-7717E6697425","full_name":"Mondelli, Marco","first_name":"Marco","orcid":"0000-0002-3242-7020"},{"last_name":"Oymak","full_name":"Oymak, Samet","first_name":"Samet"}],"file_date_updated":"2026-02-19T08:15:48Z","_id":"21325","page":"20266-20295","date_created":"2026-02-18T12:00:44Z","year":"2025","OA_place":"publisher","language":[{"iso":"eng"}],"conference":{"location":"Vancouver, Canada","name":"ICML: International Conference on Machine Learning","end_date":"2025-07-19","start_date":"2025-07-13"},"oa":1,"intvolume":"       267","volume":267,"publication":"Proceedings of the 42nd International Conference on Machine Learning","acknowledgement":"H.A.G., M.E.I., X.Z., and S.O. were supported in part by the NSF grants CCF2046816, CCF-2403075, CCF-2008020, and the Office of Naval Research grant N000142412289.\r\nM. M. is funded by the European Union (ERC, INF2 , project number 101161364). Views and opinions expressed are, however, those of the author(s) only and do not necessarily\r\nreflect those of the European Union or the European Research Council Executive Agency. Neither the European Union nor the granting authority can be held responsible for them. M.S. is supported by the Packard Fellowship in Science and Engineering, a Sloan Research Fellowship in Mathematics, an NSF-CAREER under award #1846369, DARPA FastNICS program, and NSF-CIF awards #1813877 and #2008443, and NIH DP2LM014564-01. The authors also\r\nacknowledge further support from Open Philanthropy, OpenAI, Amazon Research, Google Research, and Microsoft Research.","day":"30"},{"alternative_title":["PMLR"],"department":[{"_id":"PaSc"},{"_id":"AlBr"},{"_id":"GradSch"}],"article_processing_charge":"No","has_accepted_license":"1","publication_status":"published","date_updated":"2026-02-19T08:56:43Z","user_id":"2DF688A6-F248-11E8-B48F-1D18A9856A87","external_id":{"arxiv":["2502.09372"]},"type":"conference","oa_version":"Published Version","abstract":[{"text":"Proteins exist as a dynamic ensemble of multiple conformations, and these motions are often crucial for their functions. However, current structure prediction methods predominantly yield a single conformation, overlooking the conformational heterogeneity revealed by diverse experimental modalities. Here, we present a framework for building experiment-grounded protein structure generative models that infer conformational ensembles consistent with measured experimental data. The key idea is to treat stateof-the-art protein structure predictors (e.g., AlphaFold3) as sequence-conditioned structural priors, and cast ensemble modeling as posterior inference of protein structures given experimental measurements. Through extensive real-data experiments, we demonstrate the generality of our method to incorporate a variety of experimental measurements. In particular, our framework uncovers previously unmodeled conformational heterogeneity from crystallographic densities, and generates high-accuracy NMR ensembles orders of magnitude faster than the status quo. Notably, we demonstrate that our ensembles outperform AlphaFold3 (Abramson et al., 2024) and sometimes better fit experimental data than publicly deposited structures to the Protein Data Bank (PDB, Burley et al. (2017)). We believe that this approach will unlock building predictive models that fully embrace experimentally observed conformational diversity.","lang":"eng"}],"month":"07","OA_type":"gold","project":[{"name":"AlloSpace. The emergence and mechanisms of allostery","grant_number":"I05812","_id":"eb9c82eb-77a9-11ec-83b8-aadd536561cf"},{"grant_number":"I06223","name":"Structure and mechanism of the mitochondrial MIM insertase","_id":"bdb9578d-d553-11ed-ba76-ed5d39fce6f0"}],"status":"public","ddc":["000","540"],"citation":{"ista":"Maddipatla SA, Sellam NE, Bojan MI, Vedula S, Schanda P, Marx A, Bronstein AM. 2025. Inverse problems with experiment-guided AlphaFold. Proceedings of the 42nd International Conference on Machine Learning. ICML: International Conference on Machine Learning, PMLR, vol. 267, 42366–42393.","chicago":"Maddipatla, Sai A, Nadav E Sellam, Meital I Bojan, Sanketh Vedula, Paul Schanda, Ailie Marx, and Alex M. Bronstein. “Inverse Problems with Experiment-Guided AlphaFold.” In <i>Proceedings of the 42nd International Conference on Machine Learning</i>, 267:42366–93. ML Research Press, 2025.","ama":"Maddipatla SA, Sellam NE, Bojan MI, et al. Inverse problems with experiment-guided AlphaFold. In: <i>Proceedings of the 42nd International Conference on Machine Learning</i>. Vol 267. ML Research Press; 2025:42366-42393.","short":"S.A. Maddipatla, N.E. Sellam, M.I. Bojan, S. Vedula, P. Schanda, A. Marx, A.M. Bronstein, in:, Proceedings of the 42nd International Conference on Machine Learning, ML Research Press, 2025, pp. 42366–42393.","ieee":"S. A. Maddipatla <i>et al.</i>, “Inverse problems with experiment-guided AlphaFold,” in <i>Proceedings of the 42nd International Conference on Machine Learning</i>, Vancouver, Canada, 2025, vol. 267, pp. 42366–42393.","apa":"Maddipatla, S. A., Sellam, N. E., Bojan, M. I., Vedula, S., Schanda, P., Marx, A., &#38; Bronstein, A. M. (2025). Inverse problems with experiment-guided AlphaFold. In <i>Proceedings of the 42nd International Conference on Machine Learning</i> (Vol. 267, pp. 42366–42393). Vancouver, Canada: ML Research Press.","mla":"Maddipatla, Sai A., et al. “Inverse Problems with Experiment-Guided AlphaFold.” <i>Proceedings of the 42nd International Conference on Machine Learning</i>, vol. 267, ML Research Press, 2025, pp. 42366–93."},"tmp":{"legal_code_url":"https://creativecommons.org/licenses/by/4.0/legalcode","name":"Creative Commons Attribution 4.0 International Public License (CC-BY 4.0)","short":"CC BY (4.0)","image":"/images/cc_by.png"},"date_published":"2025-07-30T00:00:00Z","intvolume":"       267","conference":{"start_date":"2025-07-13","end_date":"2025-07-19","location":"Vancouver, Canada","name":"ICML: International Conference on Machine Learning"},"oa":1,"OA_place":"publisher","corr_author":"1","language":[{"iso":"eng"}],"year":"2025","acknowledged_ssus":[{"_id":"ScienComp"}],"acknowledgement":"This work was supported by the Israeli Science Foundation (ISF) grant number 1834/24. We acknowledge support from the Austrian Science Fund (FWF, grant numbers I5812-B and I6223) and the financial support of the Helmsley Fellowships Program for Sustainability and Health. This research uses resources of the Institute of Science and Technology Austria’s scientific computing cluster. ","day":"30","arxiv":1,"volume":267,"publication":"Proceedings of the 42nd International Conference on Machine Learning","author":[{"first_name":"Sai A","full_name":"Maddipatla, Sai A","last_name":"Maddipatla","id":"e957f5e5-91c9-11f0-a95f-e090f66ecb4d"},{"first_name":"Nadav E","full_name":"Sellam, Nadav E","id":"ef280fe0-91c9-11f0-a95f-8dea3f5bc513","last_name":"Sellam"},{"full_name":"Bojan, Meital I","id":"11d88cf5-91ca-11f0-a95f-edf9f08f47b7","last_name":"Bojan","first_name":"Meital I"},{"id":"94f2fe44-70fa-11f0-b76b-92922c09452b","full_name":"Vedula, Sanketh","last_name":"Vedula","first_name":"Sanketh"},{"first_name":"Paul","orcid":"0000-0002-9350-7606","id":"7B541462-FAF6-11E9-A490-E8DFE5697425","last_name":"Schanda","full_name":"Schanda, Paul"},{"first_name":"Ailie","full_name":"Marx, Ailie","last_name":"Marx"},{"first_name":"Alexander","orcid":"0000-0001-9699-8730","full_name":"Bronstein, Alexander","id":"58f3726e-7cba-11ef-ad8b-e6e8cb3904e6","last_name":"Bronstein"}],"title":"Inverse problems with experiment-guided AlphaFold","quality_controlled":"1","file":[{"file_size":1924177,"content_type":"application/pdf","file_name":"2025_ICML_Maddipatla.pdf","date_created":"2026-02-19T08:56:10Z","date_updated":"2026-02-19T08:56:10Z","creator":"dernst","relation":"main_file","file_id":"21338","access_level":"open_access","success":1,"checksum":"f33230a6d59b7978d4cd72795e4e9059"}],"publication_identifier":{"eissn":["2640-3498"]},"publisher":"ML Research Press","file_date_updated":"2026-02-19T08:56:10Z","page":"42366 - 42393","_id":"21327","date_created":"2026-02-18T12:11:17Z"},{"language":[{"iso":"eng"}],"corr_author":"1","OA_place":"publisher","year":"2025","intvolume":"       291","oa":1,"conference":{"location":"Lyon, France","name":"COLT: Conference on Learning Theory","start_date":"2025-06-30","end_date":"2025-07-04"},"volume":291,"publication":"Proceedings of 38th Conference on Learning Theory","day":"01","arxiv":1,"acknowledgement":"This work was done when Y. Z. was at the Institute of Science and Technology Austria. Y. Z. and\r\nM. M. are funded by the European Union (ERC, INF2, project number 101161364). Views and\r\nopinions expressed are however those of the author(s) only and do not necessarily reflect those of the European Union or the European Research Council Executive Agency. Neither the European Union nor the granting authority can be held responsible for them. The authors would like to acknowledge (in alphabetical order) discussions with Yatin Dandi, Leonardo Defilippis and Bruno Loureiro concerning their parallel work (Defilippis et al., 2025).","scopus_import":"1","file":[{"content_type":"application/pdf","file_name":"2025_LearningTheory_Kovacevic.pdf","date_created":"2026-02-19T09:03:43Z","date_updated":"2026-02-19T09:03:43Z","file_size":844611,"file_id":"21339","access_level":"open_access","creator":"dernst","relation":"main_file","success":1,"checksum":"19aa70ab4f57fb9067b6ebb99a5fd6f0"}],"publisher":"ML Research Press","publication_identifier":{"eissn":["2640-3498"]},"author":[{"id":"d0258e7b-50b8-11ef-ad56-8b9f537b6b1b","last_name":"Kovačević","full_name":"Kovačević, Filip","first_name":"Filip"},{"first_name":"Zhang","last_name":"Yihan","full_name":"Yihan, Zhang"},{"orcid":"0000-0002-3242-7020","first_name":"Marco","id":"27EB676C-8706-11E9-9510-7717E6697425","last_name":"Mondelli","full_name":"Mondelli, Marco"}],"quality_controlled":"1","title":"Spectral estimators for multi-index models: Precise asymptotics and optimal weak recovery","_id":"21328","date_created":"2026-02-18T12:12:47Z","page":"3354-3404","file_date_updated":"2026-02-19T09:03:43Z","alternative_title":["PMLR"],"department":[{"_id":"MaMo"}],"has_accepted_license":"1","article_processing_charge":"No","date_updated":"2026-02-19T09:03:53Z","publication_status":"published","user_id":"2DF688A6-F248-11E8-B48F-1D18A9856A87","abstract":[{"lang":"eng","text":"Multi-index models provide a popular framework to investigate the learnability of functions with low-dimensional structure and, also due to their connections with neural networks, they have been object of recent intensive study. In this paper, we focus on recovering the subspace spanned by the signals via spectral estimators – a family of methods routinely used in practice, often as a warm-start for iterative algorithms. Our main technical contribution is a precise asymptotic characterization of the performance of spectral methods, when sample size and input dimension grow proportionally and the dimension p of the space to recover is fixed. Specifically, we locate the top-p eigenvalues of the spectral matrix and establish the overlaps between the corresponding eigenvectors (which give the spectral estimators) and a basis of the signal subspace. Our analysis unveils a phase transition phenomenon in which, as the sample complexity grows, eigenvalues escape from the bulk of the spectrum and, when that happens, eigenvectors recover directions of the desired subspace. The precise characterization we put forward enables the optimization of the data preprocessing, thus allowing to identify the spectral estimator that requires the minimal sample size for weak recovery."}],"month":"07","external_id":{"arxiv":["2502.01583"]},"type":"conference","oa_version":"Published Version","citation":{"short":"F. Kovačević, Z. Yihan, M. Mondelli, in:, Proceedings of 38th Conference on Learning Theory, ML Research Press, 2025, pp. 3354–3404.","ieee":"F. Kovačević, Z. Yihan, and M. Mondelli, “Spectral estimators for multi-index models: Precise asymptotics and optimal weak recovery,” in <i>Proceedings of 38th Conference on Learning Theory</i>, Lyon, France, 2025, vol. 291, pp. 3354–3404.","chicago":"Kovačević, Filip, Zhang Yihan, and Marco Mondelli. “Spectral Estimators for Multi-Index Models: Precise Asymptotics and Optimal Weak Recovery.” In <i>Proceedings of 38th Conference on Learning Theory</i>, 291:3354–3404. ML Research Press, 2025.","ama":"Kovačević F, Yihan Z, Mondelli M. Spectral estimators for multi-index models: Precise asymptotics and optimal weak recovery. In: <i>Proceedings of 38th Conference on Learning Theory</i>. Vol 291. ML Research Press; 2025:3354-3404.","ista":"Kovačević F, Yihan Z, Mondelli M. 2025. Spectral estimators for multi-index models: Precise asymptotics and optimal weak recovery. Proceedings of 38th Conference on Learning Theory. COLT: Conference on Learning Theory, PMLR, vol. 291, 3354–3404.","apa":"Kovačević, F., Yihan, Z., &#38; Mondelli, M. (2025). Spectral estimators for multi-index models: Precise asymptotics and optimal weak recovery. In <i>Proceedings of 38th Conference on Learning Theory</i> (Vol. 291, pp. 3354–3404). Lyon, France: ML Research Press.","mla":"Kovačević, Filip, et al. “Spectral Estimators for Multi-Index Models: Precise Asymptotics and Optimal Weak Recovery.” <i>Proceedings of 38th Conference on Learning Theory</i>, vol. 291, ML Research Press, 2025, pp. 3354–404."},"ddc":["000"],"tmp":{"legal_code_url":"https://creativecommons.org/licenses/by/4.0/legalcode","name":"Creative Commons Attribution 4.0 International Public License (CC-BY 4.0)","short":"CC BY (4.0)","image":"/images/cc_by.png"},"date_published":"2025-07-01T00:00:00Z","OA_type":"gold","status":"public","project":[{"_id":"911e6d1f-16d5-11f0-9cad-c5c68c6a1cdf","name":"Inference in High Dimensions: Light-speed Algorithms and Information Limits","grant_number":"101161364"}]},{"article_processing_charge":"No","has_accepted_license":"1","department":[{"_id":"ToHe"},{"_id":"ChLa"}],"alternative_title":["PMLR"],"ec_funded":1,"user_id":"2DF688A6-F248-11E8-B48F-1D18A9856A87","publication_status":"published","date_updated":"2025-09-03T10:37:59Z","oa_version":"Published Version","type":"conference","external_id":{"arxiv":["2412.16564"]},"month":"06","abstract":[{"text":"We study the problem of predictive runtime monitoring of black-box dynamical systems with quantitative safety properties. The black-box setting stipulates that the exact semantics of the dynamical system and the controller are unknown, and that we are only able to observe the state of the controlled (aka, closed-loop) system at finitely many time points. We present a novel framework for predicting future states of the system based on the states observed in the past. The numbers of past states and of predicted future states are parameters provided by the user. Our method is based on a combination of Taylor’s expansion and the backward difference operator for numerical differentiation. We also derive an upper bound on the prediction error under the assumption that the system dynamics and the controller are smooth. The predicted states are then used to predict safety violations ahead in time. Our experiments demonstrate practical applicability of our method for complex black-box systems, showing that it is computationally lightweight and yet significantly more accurate than the state-of-the-art predictive safety monitoring techniques.","lang":"eng"}],"status":"public","project":[{"call_identifier":"H2020","_id":"62781420-2b32-11ec-9570-8d9b63373d4d","name":"Vigilant Algorithmic Monitoring of Software","grant_number":"101020093"}],"OA_type":"gold","date_published":"2025-06-01T00:00:00Z","citation":{"ieee":"T. A. Henzinger, F. Kresse, K. Mallik, E. Yu, and D. Zikelic, “Predictive monitoring of black-box dynamical systems,” in <i>7th Annual Learning for Dynamics &#38; Control Conference</i>, Ann Arbor, MI, United States, 2025, vol. 283, pp. 804–816.","short":"T.A. Henzinger, F. Kresse, K. Mallik, E. Yu, D. Zikelic, in:, 7th Annual Learning for Dynamics &#38; Control Conference, ML Research Press, 2025, pp. 804–816.","ama":"Henzinger TA, Kresse F, Mallik K, Yu E, Zikelic D. Predictive monitoring of black-box dynamical systems. In: <i>7th Annual Learning for Dynamics &#38; Control Conference</i>. Vol 283. ML Research Press; 2025:804-816.","chicago":"Henzinger, Thomas A, Fabian Kresse, Kaushik Mallik, Emily Yu, and Dorde Zikelic. “Predictive Monitoring of Black-Box Dynamical Systems.” In <i>7th Annual Learning for Dynamics &#38; Control Conference</i>, 283:804–16. ML Research Press, 2025.","ista":"Henzinger TA, Kresse F, Mallik K, Yu E, Zikelic D. 2025. Predictive monitoring of black-box dynamical systems. 7th Annual Learning for Dynamics &#38; Control Conference. L4DC: Learning for Dynamics &#38; Control, PMLR, vol. 283, 804–816.","mla":"Henzinger, Thomas A., et al. “Predictive Monitoring of Black-Box Dynamical Systems.” <i>7th Annual Learning for Dynamics &#38; Control Conference</i>, vol. 283, ML Research Press, 2025, pp. 804–16.","apa":"Henzinger, T. A., Kresse, F., Mallik, K., Yu, E., &#38; Zikelic, D. (2025). Predictive monitoring of black-box dynamical systems. In <i>7th Annual Learning for Dynamics &#38; Control Conference</i> (Vol. 283, pp. 804–816). Ann Arbor, MI, United States: ML Research Press."},"ddc":["000"],"conference":{"start_date":"2025-06-04","end_date":"2025-06-06","name":"L4DC: Learning for Dynamics & Control","location":"Ann Arbor, MI, United States"},"oa":1,"intvolume":"       283","year":"2025","OA_place":"publisher","corr_author":"1","language":[{"iso":"eng"}],"acknowledgement":"This work was supported in part by the ERC project ERC-2020-AdG 101020093.\r\n","day":"01","arxiv":1,"publication":"7th Annual Learning for Dynamics & Control Conference","volume":283,"title":"Predictive monitoring of black-box dynamical systems","quality_controlled":"1","author":[{"last_name":"Henzinger","full_name":"Henzinger, Thomas A","id":"40876CD8-F248-11E8-B48F-1D18A9856A87","orcid":"0000-0002-2985-7724","first_name":"Thomas A"},{"first_name":"Fabian","last_name":"Kresse","id":"faff3c84-23f6-11ef-9085-e5187b51c604","full_name":"Kresse, Fabian"},{"orcid":"0000-0001-9864-7475","first_name":"Kaushik","id":"0834ff3c-6d72-11ec-94e0-b5b0a4fb8598","full_name":"Mallik, Kaushik","last_name":"Mallik"},{"id":"20aa2ae8-f2f1-11ed-bbfa-8205053f1342","last_name":"Yu","full_name":"Yu, Zhengqi","first_name":"Zhengqi"},{"orcid":"0000-0002-4681-1699","first_name":"Dorde","last_name":"Zikelic","id":"294AA7A6-F248-11E8-B48F-1D18A9856A87","full_name":"Zikelic, Dorde"}],"publication_identifier":{"eissn":["2640-3498"]},"publisher":"ML Research Press","file":[{"creator":"dernst","relation":"main_file","file_id":"20283","access_level":"open_access","success":1,"checksum":"d5236e561560635f5ae1d17de4903033","file_size":489639,"content_type":"application/pdf","date_created":"2025-09-03T10:32:12Z","date_updated":"2025-09-03T10:32:12Z","file_name":"2025_L4DC_HenzingerT.pdf"}],"scopus_import":"1","file_date_updated":"2025-09-03T10:32:12Z","_id":"20256","page":"804-816","date_created":"2025-08-31T22:01:32Z"},{"alternative_title":["PMLR"],"department":[{"_id":"ChLa"}],"article_processing_charge":"No","has_accepted_license":"1","publication_status":"published","date_updated":"2025-09-09T08:28:41Z","user_id":"2DF688A6-F248-11E8-B48F-1D18A9856A87","external_id":{"arxiv":["2402.04840"]},"type":"conference","oa_version":"Published Version","abstract":[{"text":"In this paper, we study the problem of estimating the unknown mean θ of a unit variance Gaussian distribution in a locally differentially private (LDP) way. In the high-privacy regime (ϵ≤1\r\n), we identify an optimal privacy mechanism that minimizes the variance of the estimator asymptotically. Our main technical contribution is the maximization of the Fisher-Information of the sanitized data with respect to the local privacy mechanism Q. We find that the exact solution Qθ,ϵ of this maximization is the sign mechanism that applies randomized response to the sign of Xi−θ, where X1,…,Xn are the confidential iid original samples. However, since this optimal local mechanism depends on the unknown mean θ, we employ a two-stage LDP parameter estimation procedure which requires splitting agents into two groups. The first n1 observations are used to consistently but not necessarily efficiently estimate the parameter θ by θn1~\r\n. Then this estimate is updated by applying the sign mechanism with θ~n1 instead of θ\r\n to the remaining n−n1 observations, to obtain an LDP and efficient estimator of the unknown mean.","lang":"eng"}],"month":"05","OA_type":"diamond","status":"public","citation":{"short":"N. Kalinin, L. Steinberger, in:, Proceedings of the 28th International Conference on Artificial Intelligence and Statistics, ML Research Press, 2025, pp. 118–126.","ieee":"N. Kalinin and L. Steinberger, “Efficient estimation of a Gaussian mean with local differential privacy,” in <i>Proceedings of the 28th International Conference on Artificial Intelligence and Statistics</i>, Mai Khao, Thailand, 2025, vol. 258, pp. 118–126.","ista":"Kalinin N, Steinberger L. 2025. Efficient estimation of a Gaussian mean with local differential privacy. Proceedings of the 28th International Conference on Artificial Intelligence and Statistics. AISTATS: Conference on Artificial Intelligence and Statistics, PMLR, vol. 258, 118–126.","chicago":"Kalinin, Nikita, and Lukas Steinberger. “Efficient Estimation of a Gaussian Mean with Local Differential Privacy.” In <i>Proceedings of the 28th International Conference on Artificial Intelligence and Statistics</i>, 258:118–26. ML Research Press, 2025.","ama":"Kalinin N, Steinberger L. Efficient estimation of a Gaussian mean with local differential privacy. In: <i>Proceedings of the 28th International Conference on Artificial Intelligence and Statistics</i>. Vol 258. ML Research Press; 2025:118-126.","mla":"Kalinin, Nikita, and Lukas Steinberger. “Efficient Estimation of a Gaussian Mean with Local Differential Privacy.” <i>Proceedings of the 28th International Conference on Artificial Intelligence and Statistics</i>, vol. 258, ML Research Press, 2025, pp. 118–26.","apa":"Kalinin, N., &#38; Steinberger, L. (2025). Efficient estimation of a Gaussian mean with local differential privacy. In <i>Proceedings of the 28th International Conference on Artificial Intelligence and Statistics</i> (Vol. 258, pp. 118–126). Mai Khao, Thailand: ML Research Press."},"ddc":["000"],"date_published":"2025-05-01T00:00:00Z","tmp":{"legal_code_url":"https://creativecommons.org/licenses/by/4.0/legalcode","name":"Creative Commons Attribution 4.0 International Public License (CC-BY 4.0)","short":"CC BY (4.0)","image":"/images/cc_by.png"},"intvolume":"       258","conference":{"end_date":"2025-05-05","start_date":"2025-05-03","name":"AISTATS: Conference on Artificial Intelligence and Statistics","location":"Mai Khao, Thailand"},"oa":1,"OA_place":"publisher","corr_author":"1","language":[{"iso":"eng"}],"year":"2025","acknowledgement":"We would like to express our gratitude to Christoph Lampert for his valuable insights and fruitful discussions that significantly contributed to the development of this paper.\r\nWe also thank Salil Vadhan for his constructive feedback on an earlier version of this draft.\r\nThe second author gratefully acknowledges support by the Austrian Science Fund (FWF): I 5484-N, as part of the Research Unit 5381 of the German Research Foundation.","arxiv":1,"day":"01","volume":258,"publication":"Proceedings of the 28th International Conference on Artificial Intelligence and Statistics","author":[{"first_name":"Nikita","last_name":"Kalinin","full_name":"Kalinin, Nikita","id":"4b14526e-14d2-11ed-ba64-c14c9553d137"},{"last_name":"Steinberger","full_name":"Steinberger, Lukas","first_name":"Lukas"}],"title":"Efficient estimation of a Gaussian mean with local differential privacy","quality_controlled":"1","file":[{"success":1,"checksum":"3dcd59988ca974b98662ba09a516e616","creator":"dernst","relation":"main_file","file_id":"20316","access_level":"open_access","file_size":395864,"content_type":"application/pdf","date_created":"2025-09-09T08:26:44Z","date_updated":"2025-09-09T08:26:44Z","file_name":"2025_AISTATS_Kalinin.pdf"}],"scopus_import":"1","publication_identifier":{"eissn":["2640-3498"]},"publisher":"ML Research Press","file_date_updated":"2025-09-09T08:26:44Z","page":"118-126","_id":"20298","date_created":"2025-09-07T22:01:34Z"},{"date_updated":"2025-09-09T07:00:34Z","publication_status":"published","user_id":"2DF688A6-F248-11E8-B48F-1D18A9856A87","department":[{"_id":"MaMo"}],"alternative_title":["PMLR"],"article_processing_charge":"No","citation":{"ieee":"T. Wegel, F. Kovačević, A. Ţifrea, and F. Yang, “Learning Pareto manifolds in high dimensions: How can regularization help?,” in <i>The 28th International Conference on Artificial Intelligence and Statistics</i>, Mai Khao, Thailand, 2025, vol. 258, pp. 4591–4599.","short":"T. Wegel, F. Kovačević, A. Ţifrea, F. Yang, in:, The 28th International Conference on Artificial Intelligence and Statistics, ML Research Press, 2025, pp. 4591–4599.","chicago":"Wegel, Tobias, Filip Kovačević, Alexandru Ţifrea, and Fanny Yang. “Learning Pareto Manifolds in High Dimensions: How Can Regularization Help?” In <i>The 28th International Conference on Artificial Intelligence and Statistics</i>, 258:4591–99. ML Research Press, 2025.","ama":"Wegel T, Kovačević F, Ţifrea A, Yang F. Learning Pareto manifolds in high dimensions: How can regularization help? In: <i>The 28th International Conference on Artificial Intelligence and Statistics</i>. Vol 258. ML Research Press; 2025:4591-4599.","ista":"Wegel T, Kovačević F, Ţifrea A, Yang F. 2025. Learning Pareto manifolds in high dimensions: How can regularization help? The 28th International Conference on Artificial Intelligence and Statistics. AISTATS: Conference on Artificial Intelligence and Statistics, PMLR, vol. 258, 4591–4599.","mla":"Wegel, Tobias, et al. “Learning Pareto Manifolds in High Dimensions: How Can Regularization Help?” <i>The 28th International Conference on Artificial Intelligence and Statistics</i>, vol. 258, ML Research Press, 2025, pp. 4591–99.","apa":"Wegel, T., Kovačević, F., Ţifrea, A., &#38; Yang, F. (2025). Learning Pareto manifolds in high dimensions: How can regularization help? In <i>The 28th International Conference on Artificial Intelligence and Statistics</i> (Vol. 258, pp. 4591–4599). Mai Khao, Thailand: ML Research Press."},"date_published":"2025-05-01T00:00:00Z","OA_type":"green","main_file_link":[{"url":"https://doi.org/10.48550/arXiv.2503.08849","open_access":"1"}],"status":"public","abstract":[{"lang":"eng","text":"Simultaneously addressing multiple objectives is becoming increasingly important in modern machine learning. At the same time, data is often high-dimensional and costly to label. For a single objective such as prediction risk, conventional regularization techniques are known to improve generalization when the data exhibits low-dimensional structure like sparsity. However, it is largely unexplored how to leverage this structure in the context of multi-objective learning (MOL) with multiple competing objectives. In this work, we discuss how the application of vanilla regularization approaches can fail, and propose a two-stage MOL framework that can successfully leverage low-dimensional structure. We demonstrate its effectiveness experimentally for multi-distribution learning and fairness-risk trade-offs."}],"month":"05","type":"conference","external_id":{"arxiv":["2503.08849"]},"oa_version":"Preprint","volume":258,"publication":"The 28th International Conference on Artificial Intelligence and Statistics","day":"01","arxiv":1,"acknowledgement":"We thank Junhyung Park for valuable feedback on the manuscript. AT was supported by a PhD fellowship from the Swiss Data Science Center. TW was supported by the SNF Grant 204439. This work was done in part while TW and FY were visiting the Simons Institute for the Theory of\r\nComputing.","language":[{"iso":"eng"}],"OA_place":"repository","year":"2025","intvolume":"       258","oa":1,"conference":{"location":"Mai Khao, Thailand","name":"AISTATS: Conference on Artificial Intelligence and Statistics","end_date":"2025-05-05","start_date":"2025-05-03"},"page":"4591-4599","_id":"20300","date_created":"2025-09-07T22:01:35Z","scopus_import":"1","publisher":"ML Research Press","publication_identifier":{"eissn":["2640-3498"]},"author":[{"first_name":"Tobias","full_name":"Wegel, Tobias","last_name":"Wegel"},{"first_name":"Filip","last_name":"Kovačević","full_name":"Kovačević, Filip","id":"d0258e7b-50b8-11ef-ad56-8b9f537b6b1b"},{"last_name":"Ţifrea","full_name":"Ţifrea, Alexandru","first_name":"Alexandru"},{"last_name":"Yang","full_name":"Yang, Fanny","first_name":"Fanny"}],"quality_controlled":"1","title":"Learning Pareto manifolds in high dimensions: How can regularization help?"},{"oa":1,"conference":{"start_date":"2025-05-03","end_date":"2025-05-05","location":"Mai Khao, Thailand","name":"AISTATS: Conference on Artificial Intelligence and Statistics"},"intvolume":"       258","year":"2025","language":[{"iso":"eng"}],"OA_place":"repository","day":"01","arxiv":1,"acknowledgement":"MH: This project has received funding from the European Research Council (ERC) under the European Union’s Horizon 2020 research and innovation programme (MoDynStruct, No. 101019564) and the Austrian Science Fund (FWF) grant DOI 10.55776/Z422, grant DOI 10.55776/I5982, and grant DOI 10.55776/P33775 with additional funding from the netidee SCIENCE Stiftung, 2020–2024. TAS: This work was supported by a research grant (VIL51463)\r\nfrom VILLUM FONDEN.","publication":"The 28th International Conference on Artificial Intelligence and Statistics","volume":258,"quality_controlled":"1","title":"Differentially private continual release of histograms and related queries","author":[{"first_name":"Monika H","orcid":"0000-0002-5008-6530","full_name":"Henzinger, Monika H","last_name":"Henzinger","id":"540c9bbd-f2de-11ec-812d-d04a5be85630"},{"last_name":"Sricharan","full_name":"Sricharan, A. R.","first_name":"A. R."},{"full_name":"Steiner, Teresa Anna","last_name":"Steiner","first_name":"Teresa Anna"}],"publisher":"ML Research Press","publication_identifier":{"eissn":["2640-3498"]},"scopus_import":"1","_id":"20301","page":"1990-1998","date_created":"2025-09-07T22:01:35Z","article_processing_charge":"No","alternative_title":["PMLR"],"department":[{"_id":"MoHe"}],"ec_funded":1,"user_id":"2DF688A6-F248-11E8-B48F-1D18A9856A87","date_updated":"2025-09-09T07:09:22Z","publication_status":"published","oa_version":"Preprint","external_id":{"arxiv":["2302.11341"]},"type":"conference","month":"05","abstract":[{"lang":"eng","text":"We study privately releasing column sums of a d-dimensional table with entries from a universe χ undergoing T row updates, called histogram under continual release. Our mechanisms give better additive ℓ∞-error than existing mechanisms for a large class of queries and input streams. Our first contribution is an output-sensitive mechanism in the insertions-only model (χ = {0, 1}) for maintaining (i) the histogram or (ii) queries that do not require maintaining the entire histogram, such as the maximum or minimum column sum, the median, or any quantiles. The mechanism has an additive error of O(d log2 (dq∗) + log T) whp, where q∗ is the maximum output value over all time steps on this dataset. The mechanism does not require q∗ as input. This breaks the Ω(d log T) bound of prior work when q∗ ≪ T. Our second contribution is a mechanism for the turnstile model that admits negative entry updates (χ = {−1, 0, 1}). This mechanism has an additive error of O(d log2(dK) + log T) whp, where K is the number of times two consecutive data rows differ, and the mechanism does not require K as input. This is useful when monitoring inputs that only vary under unusual circumstances. For d = 1 this gives the first\r\nprivate mechanism with error O(log2 K + log T) for continual counting in the turnstile model, improving on the O(log2 n + log T) error bound by Dwork et al. (2015), where n is the number of ones in the stream, as well as allowing negative entries, while Dwork et al. (2015) can only handle nonnegative entries (χ = {0, 1}). "}],"main_file_link":[{"open_access":"1","url":"https://doi.org/10.48550/arXiv.2302.11341"}],"status":"public","project":[{"call_identifier":"H2020","_id":"bd9ca328-d553-11ed-ba76-dc4f890cfe62","grant_number":"101019564","name":"The design and evaluation of modern fully dynamic data structures"},{"_id":"34def286-11ca-11ed-8bc3-da5948e1613c","grant_number":"Z00422","name":"Efficient algorithms"},{"name":"Static and Dynamic Hierarchical Graph Decompositions","grant_number":"I05982","_id":"bda196b2-d553-11ed-ba76-8e8ee6c21103"},{"grant_number":"P33775","name":"Fast Algorithms for a Reactive Network Layer","_id":"bd9e3a2e-d553-11ed-ba76-8aa684ce17fe"}],"OA_type":"green","date_published":"2025-05-01T00:00:00Z","citation":{"ieee":"M. Henzinger, A. R. Sricharan, and T. A. Steiner, “Differentially private continual release of histograms and related queries,” in <i>The 28th International Conference on Artificial Intelligence and Statistics</i>, Mai Khao, Thailand, 2025, vol. 258, pp. 1990–1998.","short":"M. Henzinger, A.R. Sricharan, T.A. Steiner, in:, The 28th International Conference on Artificial Intelligence and Statistics, ML Research Press, 2025, pp. 1990–1998.","chicago":"Henzinger, Monika, A. R. Sricharan, and Teresa Anna Steiner. “Differentially Private Continual Release of Histograms and Related Queries.” In <i>The 28th International Conference on Artificial Intelligence and Statistics</i>, 258:1990–98. ML Research Press, 2025.","ista":"Henzinger M, Sricharan AR, Steiner TA. 2025. Differentially private continual release of histograms and related queries. The 28th International Conference on Artificial Intelligence and Statistics. AISTATS: Conference on Artificial Intelligence and Statistics, PMLR, vol. 258, 1990–1998.","ama":"Henzinger M, Sricharan AR, Steiner TA. Differentially private continual release of histograms and related queries. In: <i>The 28th International Conference on Artificial Intelligence and Statistics</i>. Vol 258. ML Research Press; 2025:1990-1998.","mla":"Henzinger, Monika, et al. “Differentially Private Continual Release of Histograms and Related Queries.” <i>The 28th International Conference on Artificial Intelligence and Statistics</i>, vol. 258, ML Research Press, 2025, pp. 1990–98.","apa":"Henzinger, M., Sricharan, A. R., &#38; Steiner, T. A. (2025). Differentially private continual release of histograms and related queries. In <i>The 28th International Conference on Artificial Intelligence and Statistics</i> (Vol. 258, pp. 1990–1998). Mai Khao, Thailand: ML Research Press."}},{"article_processing_charge":"No","alternative_title":["PMLR"],"department":[{"_id":"FrLo"}],"user_id":"2DF688A6-F248-11E8-B48F-1D18A9856A87","date_updated":"2025-09-09T07:47:13Z","publication_status":"published","oa_version":"Preprint","type":"conference","external_id":{"arxiv":["2410.12300"]},"month":"05","abstract":[{"text":"Observational genome-wide association studies are now widely used for causal inference in genetic epidemiology. To maintain privacy, such data is often only publicly available as summary statistics, and often studies for the endogenous covariates and the outcome are available separately. This has necessitated methods tailored to two-sample summary statistics. Current state-of-the-art methods modify linear instrumental variable (IV) regression---with genetic variants as instruments---to account for unmeasured confounding. However, since the endogenous covariates can be high dimensional, standard IV assumptions are generally insufficient to identify all causal effects simultaneously. We ensure identifiability by assuming the causal effects are sparse and propose a sparse causal effect two-sample IV estimator, spaceTSIV, adapting the spaceIV estimator by Pfister and Peters (2022) for two-sample summary statistics. We provide two methods, based on L0- and L1-penalization, respectively. We prove identifiability of the sparse causal effects in the two-sample setting and consistency of spaceTSIV. The performance of spaceTSIV is compared with existing two-sample IV methods in simulations. Finally, we showcase our methods using real proteomic and gene-expression data for drug-target discovery.","lang":"eng"}],"main_file_link":[{"url":"https://doi.org/10.48550/arXiv.2410.12300","open_access":"1"}],"status":"public","OA_type":"green","date_published":"2025-05-01T00:00:00Z","citation":{"mla":"Huang, Shimeng, et al. “Sparse Causal Effect Estimation Using Two-Sample Summary Statistics in the Presence of Unmeasured Confounding.” <i>The 28th International Conference on Artificial Intelligence and Statistics</i>, vol. 258, ML Research Press, 2025, pp. 3394–402.","apa":"Huang, S., Pfister, N., &#38; Bowden, J. (2025). Sparse causal effect estimation using two-sample summary statistics in the presence of unmeasured confounding. In <i>The 28th International Conference on Artificial Intelligence and Statistics</i> (Vol. 258, pp. 3394–3402). Mai Khao, Thailand: ML Research Press.","ista":"Huang S, Pfister N, Bowden J. 2025. Sparse causal effect estimation using two-sample summary statistics in the presence of unmeasured confounding. The 28th International Conference on Artificial Intelligence and Statistics. AISTATS: Conference on Artificial Intelligence and Statistics, PMLR, vol. 258, 3394–3402.","ama":"Huang S, Pfister N, Bowden J. Sparse causal effect estimation using two-sample summary statistics in the presence of unmeasured confounding. In: <i>The 28th International Conference on Artificial Intelligence and Statistics</i>. Vol 258. ML Research Press; 2025:3394-3402.","chicago":"Huang, Shimeng, Niklas Pfister, and Jack Bowden. “Sparse Causal Effect Estimation Using Two-Sample Summary Statistics in the Presence of Unmeasured Confounding.” In <i>The 28th International Conference on Artificial Intelligence and Statistics</i>, 258:3394–3402. ML Research Press, 2025.","ieee":"S. Huang, N. Pfister, and J. Bowden, “Sparse causal effect estimation using two-sample summary statistics in the presence of unmeasured confounding,” in <i>The 28th International Conference on Artificial Intelligence and Statistics</i>, Mai Khao, Thailand, 2025, vol. 258, pp. 3394–3402.","short":"S. Huang, N. Pfister, J. Bowden, in:, The 28th International Conference on Artificial Intelligence and Statistics, ML Research Press, 2025, pp. 3394–3402."},"oa":1,"conference":{"location":"Mai Khao, Thailand","name":"AISTATS: Conference on Artificial Intelligence and Statistics","end_date":"2025-05-05","start_date":"2025-05-03"},"intvolume":"       258","year":"2025","language":[{"iso":"eng"}],"OA_place":"repository","day":"01","arxiv":1,"acknowledgement":"The authors would like to thank Stephen Burgess and Ashish Patel for helpful discussions at\r\nthe start of this research project, and Anton Rask Lundborg for helpful discussions on the\r\nuniform asymptotic results. This work was partially completed during SH’s research visit at\r\nNovo Nordisk. The authors would like to thank Jesper Ferkinghoff-Borg, Kang Li and Lewis\r\nMarsh for facilitating this visit and for discussing necessary concepts and tools in statistical\r\ngenetics at an early stage. SH and NP are supported by a research grant (0069071) from Novo\r\nNordisk Fonden. JB is funded at the University of Exeter by research grant MR/X011372/1.","publication":"The 28th International Conference on Artificial Intelligence and Statistics","volume":258,"quality_controlled":"1","title":"Sparse causal effect estimation using two-sample summary statistics in the presence of unmeasured confounding","author":[{"last_name":"Huang","full_name":"Huang, Shimeng","id":"989c2a06-fb4e-11ef-a992-ab766442255b","orcid":"0000-0001-6919-821X","first_name":"Shimeng"},{"last_name":"Pfister","full_name":"Pfister, Niklas","first_name":"Niklas"},{"last_name":"Bowden","full_name":"Bowden, Jack","first_name":"Jack"}],"publisher":"ML Research Press","publication_identifier":{"eissn":["2640-3498"]},"scopus_import":"1","page":"3394-3402","_id":"20303","date_created":"2025-09-07T22:01:35Z"},{"license":"https://creativecommons.org/licenses/by-sa/4.0/","article_processing_charge":"No","has_accepted_license":"1","alternative_title":["PMLR"],"department":[{"_id":"FrLo"}],"user_id":"2DF688A6-F248-11E8-B48F-1D18A9856A87","publication_status":"published","date_updated":"2025-11-10T08:33:11Z","month":"01","abstract":[{"lang":"eng","text":"Filtering-based probabilistic numerical solvers for ordinary differential equations (ODEs), also known as ODE filters, have been established as efficient methods for quantifying numerical uncertainty in the solution of ODEs. In practical applications, however, the underlying dynamical system often contains uncertain parameters, requiring the propagation of this model uncertainty to the ODE solution. In this paper, we demonstrate that ODE filters, despite their probabilistic nature, do not automatically solve this uncertainty propagation problem. To address this limitation, we present a novel approach that combines ODE filters with numerical quadrature to properly marginalize over uncertain parameters, while accounting for both parameter uncertainty and numerical solver uncertainty. Experiments across multiple dynamical systems demonstrate that the resulting uncertainty estimates closely match reference solutions. Notably, we show\r\nhow the numerical uncertainty from the ODE solver can help prevent overconfidence in the propagated uncertainty estimates, especially when using larger step sizes. Our results illustrate that probabilistic numerical methods can effectively quantify both numerical and parametric uncertainty in dynamical systems. "}],"oa_version":"Preprint","type":"conference","external_id":{"arxiv":["2503.04684"]},"tmp":{"image":"/images/cc_by_sa.png","legal_code_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","name":"Creative Commons Attribution-ShareAlike 4.0 International Public License (CC BY-SA 4.0)","short":"CC BY-SA (4.0)"},"date_published":"2025-01-01T00:00:00Z","ddc":["000"],"citation":{"ista":"Yao D, Tronarp F, Bosch N. 2025. Propagating model uncertainty through filtering-based probabilistic numerical ODE solvers. Proceedings of the 1st International Conference on Probabilistic Numerics. ProbNum: Conference on Probabilistic Numerics, PMLR, vol. 271.","ama":"Yao D, Tronarp F, Bosch N. Propagating model uncertainty through filtering-based probabilistic numerical ODE solvers. In: <i>Proceedings of the 1st International Conference on Probabilistic Numerics</i>. Vol 271. ML Research Press; 2025.","chicago":"Yao, Dingling, Filip Tronarp, and Nathanael Bosch. “Propagating Model Uncertainty through Filtering-Based Probabilistic Numerical ODE Solvers.” In <i>Proceedings of the 1st International Conference on Probabilistic Numerics</i>, Vol. 271. ML Research Press, 2025.","short":"D. Yao, F. Tronarp, N. Bosch, in:, Proceedings of the 1st International Conference on Probabilistic Numerics, ML Research Press, 2025.","ieee":"D. Yao, F. Tronarp, and N. Bosch, “Propagating model uncertainty through filtering-based probabilistic numerical ODE solvers,” in <i>Proceedings of the 1st International Conference on Probabilistic Numerics</i>, Sophia Antipolis, France, 2025, vol. 271.","mla":"Yao, Dingling, et al. “Propagating Model Uncertainty through Filtering-Based Probabilistic Numerical ODE Solvers.” <i>Proceedings of the 1st International Conference on Probabilistic Numerics</i>, vol. 271, ML Research Press, 2025.","apa":"Yao, D., Tronarp, F., &#38; Bosch, N. (2025). Propagating model uncertainty through filtering-based probabilistic numerical ODE solvers. In <i>Proceedings of the 1st International Conference on Probabilistic Numerics</i> (Vol. 271). Sophia Antipolis, France: ML Research Press."},"status":"public","main_file_link":[{"url":"https://openreview.net/forum?id=sgPCP9jOlS","open_access":"1"}],"OA_type":"green","year":"2025","OA_place":"repository","language":[{"iso":"eng"}],"conference":{"location":"Sophia Antipolis, France","name":"ProbNum: Conference on Probabilistic Numerics","start_date":"2025-09-01","end_date":"2025-09-03"},"oa":1,"intvolume":"       271","publication":"Proceedings of the 1st International Conference on Probabilistic Numerics","volume":271,"acknowledgement":"NB gratefully acknowledge co-funding by the European Union (ERC, ANUBIS, 101123955. Views and opinions expressed are however those of the author(s) only and do not necessarily reflect those of the European Union or the European Research Council. Neither the European Union nor the granting authority can be held responsible for them). NB thanks the International\r\nMax Planck Research School for Intelligent Systems (IMPRS-IS) for their support.","arxiv":1,"day":"01","publication_identifier":{"eissn":["2640-3498"]},"publisher":"ML Research Press","scopus_import":"1","title":"Propagating model uncertainty through filtering-based probabilistic numerical ODE solvers","quality_controlled":"1","author":[{"id":"d3e02e50-48a8-11ee-8f62-c108061797fa","full_name":"Yao, Dingling","last_name":"Yao","first_name":"Dingling"},{"first_name":"Filip","full_name":"Tronarp, Filip","last_name":"Tronarp"},{"first_name":"Nathanael","full_name":"Bosch, Nathanael","last_name":"Bosch"}],"_id":"20592","date_created":"2025-11-02T23:01:35Z"},{"title":"Lower bound on Howard policy iteration for deterministic Markov Decision Processes","quality_controlled":"1","author":[{"last_name":"Asadi","full_name":"Asadi, Ali","id":"02d96aae-000e-11ec-b801-cadd0a5eefbb","first_name":"Ali"},{"first_name":"Krishnendu","orcid":"0000-0002-4561-241X","full_name":"Chatterjee, Krishnendu","last_name":"Chatterjee","id":"2E5DCA20-F248-11E8-B48F-1D18A9856A87"},{"last_name":"De Raaij","full_name":"De Raaij, Jakob","first_name":"Jakob"}],"publication_identifier":{"eissn":["2640-3498"]},"publisher":"ML Research Press","file":[{"success":1,"checksum":"4180c81bb6ed3b4f5c7a8e48d06520c6","file_id":"20313","access_level":"open_access","creator":"dernst","relation":"main_file","content_type":"application/pdf","file_name":"2025_UAI_Asadi.pdf","date_updated":"2025-09-09T06:27:59Z","date_created":"2025-09-09T06:27:59Z","file_size":317097}],"scopus_import":"1","file_date_updated":"2025-09-09T06:27:59Z","_id":"20299","date_created":"2025-09-07T22:01:34Z","page":"223-232","conference":{"start_date":"2025-07-21","end_date":"2025-07-25","location":"Rio de Janeiro, Brazil","name":"UAI: Conference on Uncertainty in Artificial Intelligence"},"oa":1,"intvolume":"       286","year":"2025","OA_place":"publisher","corr_author":"1","language":[{"iso":"eng"}],"acknowledgement":"This research was partially supported by the ERC CoG 863818 (ForM-SMArt) grant and Austrian Science Fund (FWF) 10.55776/COE12.\r\n","arxiv":1,"day":"01","volume":286,"publication":"The 41st Conference on Uncertainty in Artificial Intelligence","oa_version":"Published Version","type":"conference","external_id":{"arxiv":["2506.12254"]},"month":"01","abstract":[{"lang":"eng","text":"Deterministic Markov Decision Processes (DMDPs) are a mathematical framework for decision-making where the outcomes and future possible actions are deterministically determined by the current action taken. DMDPs can be viewed as a finite directed weighted graph, where in each step, the controller chooses an outgoing edge. An objective is a measurable function on runs (or infinite trajectories) of the DMDP, and the value for an objective is the maximal cumulative reward (or weight) that the controller can guarantee. We consider the classical mean-payoff (aka limit-average) objective, which is a basic and fundamental objective.\r\n\r\nHoward's policy iteration algorithm is a popular method for solving DMDPs with mean-payoff objectives. Although Howard's algorithm performs well in practice, as experimental studies suggested, the best known upper bound is exponential and the current known lower bound is as follows: For the input size I, the algorithm requires (math formular) iterations, where (math formular) hides the poly-logarithmic factors, i.e., the current lower bound on iterations is sub-linear with respect to the input size. Our main result is an improved lower bound for this fundamental algorithm where we show that for the input size I, the algorithm requires (math formular) iterations."}],"project":[{"_id":"0599E47C-7A3F-11EA-A408-12923DDC885E","call_identifier":"H2020","grant_number":"863818","name":"Formal Methods for Stochastic Models: Algorithms and Applications"},{"grant_number":"COE12","name":"Bilateral Artificial Intelligence (Chatterjee)","_id":"4029cfc7-b034-11f1-9e55-88ab2ff3b6ee"}],"status":"public","OA_type":"diamond","tmp":{"legal_code_url":"https://creativecommons.org/licenses/by/4.0/legalcode","name":"Creative Commons Attribution 4.0 International Public License (CC-BY 4.0)","short":"CC BY (4.0)","image":"/images/cc_by.png"},"date_published":"2025-01-01T00:00:00Z","citation":{"short":"A. Asadi, K. Chatterjee, J. De Raaij, in:, The 41st Conference on Uncertainty in Artificial Intelligence, ML Research Press, 2025, pp. 223–232.","ieee":"A. Asadi, K. Chatterjee, and J. De Raaij, “Lower bound on Howard policy iteration for deterministic Markov Decision Processes,” in <i>The 41st Conference on Uncertainty in Artificial Intelligence</i>, Rio de Janeiro, Brazil, 2025, vol. 286, pp. 223–232.","ama":"Asadi A, Chatterjee K, De Raaij J. Lower bound on Howard policy iteration for deterministic Markov Decision Processes. In: <i>The 41st Conference on Uncertainty in Artificial Intelligence</i>. Vol 286. ML Research Press; 2025:223-232.","ista":"Asadi A, Chatterjee K, De Raaij J. 2025. Lower bound on Howard policy iteration for deterministic Markov Decision Processes. The 41st Conference on Uncertainty in Artificial Intelligence. UAI: Conference on Uncertainty in Artificial Intelligence, PMLR, vol. 286, 223–232.","chicago":"Asadi, Ali, Krishnendu Chatterjee, and Jakob De Raaij. “Lower Bound on Howard Policy Iteration for Deterministic Markov Decision Processes.” In <i>The 41st Conference on Uncertainty in Artificial Intelligence</i>, 286:223–32. ML Research Press, 2025.","mla":"Asadi, Ali, et al. “Lower Bound on Howard Policy Iteration for Deterministic Markov Decision Processes.” <i>The 41st Conference on Uncertainty in Artificial Intelligence</i>, vol. 286, ML Research Press, 2025, pp. 223–32.","apa":"Asadi, A., Chatterjee, K., &#38; De Raaij, J. (2025). Lower bound on Howard policy iteration for deterministic Markov Decision Processes. In <i>The 41st Conference on Uncertainty in Artificial Intelligence</i> (Vol. 286, pp. 223–232). Rio de Janeiro, Brazil: ML Research Press."},"ddc":["000"],"article_processing_charge":"No","has_accepted_license":"1","alternative_title":["PMLR"],"department":[{"_id":"KrCh"},{"_id":"GradSch"}],"ec_funded":1,"user_id":"2DF688A6-F248-11E8-B48F-1D18A9856A87","publication_status":"published","date_updated":"2026-09-16T07:01:05Z"},{"date_published":"2025-07-01T00:00:00Z","tmp":{"legal_code_url":"https://creativecommons.org/licenses/by/4.0/legalcode","name":"Creative Commons Attribution 4.0 International Public License (CC-BY 4.0)","short":"CC BY (4.0)","image":"/images/cc_by.png"},"citation":{"mla":"Asadi, Ali, et al. “Limit-Sure Reachability for Small Memory Policies in POMDPs Is NP-Complete.” <i>The 41st Conference on Uncertainty in Artificial Intelligence</i>, vol. 286, ML Research Press, 2025, pp. 238–47.","apa":"Asadi, A., Chatterjee, K., Saona Urmeneta, R. J., &#38; Shafiee, A. (2025). Limit-sure reachability for small memory policies in POMDPs is NP-complete. In <i>The 41st Conference on Uncertainty in Artificial Intelligence</i> (Vol. 286, pp. 238–247). Rio de Janeiro, Brazil: ML Research Press.","ieee":"A. Asadi, K. Chatterjee, R. J. Saona Urmeneta, and A. Shafiee, “Limit-sure reachability for small memory policies in POMDPs is NP-complete,” in <i>The 41st Conference on Uncertainty in Artificial Intelligence</i>, Rio de Janeiro, Brazil, 2025, vol. 286, pp. 238–247.","short":"A. Asadi, K. Chatterjee, R.J. Saona Urmeneta, A. Shafiee, in:, The 41st Conference on Uncertainty in Artificial Intelligence, ML Research Press, 2025, pp. 238–247.","chicago":"Asadi, Ali, Krishnendu Chatterjee, Raimundo J Saona Urmeneta, and Ali Shafiee. “Limit-Sure Reachability for Small Memory Policies in POMDPs Is NP-Complete.” In <i>The 41st Conference on Uncertainty in Artificial Intelligence</i>, 286:238–47. ML Research Press, 2025.","ama":"Asadi A, Chatterjee K, Saona Urmeneta RJ, Shafiee A. Limit-sure reachability for small memory policies in POMDPs is NP-complete. In: <i>The 41st Conference on Uncertainty in Artificial Intelligence</i>. Vol 286. ML Research Press; 2025:238-247.","ista":"Asadi A, Chatterjee K, Saona Urmeneta RJ, Shafiee A. 2025. Limit-sure reachability for small memory policies in POMDPs is NP-complete. The 41st Conference on Uncertainty in Artificial Intelligence. UAI: Conference on Uncertainty in Artificial Intelligence, PMLR, vol. 286, 238–247."},"ddc":["000"],"status":"public","project":[{"name":"Formal Methods for Stochastic Models: Algorithms and Applications","grant_number":"863818","_id":"0599E47C-7A3F-11EA-A408-12923DDC885E","call_identifier":"H2020"},{"grant_number":"COE12","name":"Bilateral Artificial Intelligence (Chatterjee)","_id":"4029cfc7-b034-11f1-9e55-88ab2ff3b6ee"}],"OA_type":"diamond","month":"07","abstract":[{"lang":"eng","text":"A standard model that arises in several applications in sequential decision-making is partially observable Markov decision processes (POMDPs) where a decision-making agent interacts with an uncertain environment. A basic objective in POMDPs is the reachability objective, where given a target set of states, the goal is to eventually arrive at one of them.\r\n\r\nThe limit-sure problem asks whether reachability can be ensured with probability arbitrarily close to 1. In general, the limit-sure reachability problem for POMDPs is undecidable. However, in many practical cases, the most relevant question is the existence of policies with a small amount of memory. In this work, we study the limit-sure reachability problem for POMDPs with a fixed amount of memory. We establish that the computational complexity of the problem is NP-complete."}],"oa_version":"Published Version","type":"conference","external_id":{"arxiv":["2412.00941"]},"ec_funded":1,"user_id":"2DF688A6-F248-11E8-B48F-1D18A9856A87","publication_status":"published","date_updated":"2026-09-16T07:03:02Z","article_processing_charge":"No","has_accepted_license":"1","department":[{"_id":"KrCh"},{"_id":"GradSch"}],"alternative_title":["PMLR"],"file_date_updated":"2025-09-09T08:19:41Z","_id":"20297","date_created":"2025-09-07T22:01:34Z","page":"238-247","publication_identifier":{"eissn":["2640-3498"]},"publisher":"ML Research Press","file":[{"date_updated":"2025-09-09T08:19:41Z","file_name":"2025_UAI_AsadiAli.pdf","date_created":"2025-09-09T08:19:41Z","content_type":"application/pdf","file_size":307458,"checksum":"1a37ebe7ba73ab6985765bf0d17a0acc","success":1,"access_level":"open_access","file_id":"20315","creator":"dernst","relation":"main_file"}],"scopus_import":"1","title":"Limit-sure reachability for small memory policies in POMDPs is NP-complete","quality_controlled":"1","author":[{"full_name":"Asadi, Ali","id":"02d96aae-000e-11ec-b801-cadd0a5eefbb","last_name":"Asadi","first_name":"Ali"},{"full_name":"Chatterjee, Krishnendu","id":"2E5DCA20-F248-11E8-B48F-1D18A9856A87","last_name":"Chatterjee","first_name":"Krishnendu","orcid":"0000-0002-4561-241X"},{"full_name":"Saona Urmeneta, Raimundo J","id":"BD1DF4C4-D767-11E9-B658-BC13E6697425","last_name":"Saona Urmeneta","orcid":"0000-0001-5103-038X","first_name":"Raimundo J"},{"first_name":"Ali","id":"2783031a-7378-11f0-b2d0-f17f1db2ebad","full_name":"Shafiee, Ali","last_name":"Shafiee"}],"publication":"The 41st Conference on Uncertainty in Artificial Intelligence","volume":286,"acknowledgement":"This research was partially supported by Austrian Science Fund (FWF) 10.55776/COE12, the support of the French Agence Nationale de la Recherche (ANR) under reference ANR-21-CE40-0020 (CONVERGENCE project), and the ERC CoG 863818 (ForM-SMArt) grant.","day":"01","arxiv":1,"year":"2025","OA_place":"publisher","corr_author":"1","language":[{"iso":"eng"}],"conference":{"name":"UAI: Conference on Uncertainty in Artificial Intelligence","location":"Rio de Janeiro, Brazil","end_date":"2025-07-25","start_date":"2025-07-21"},"oa":1,"intvolume":"       286"},{"file":[{"relation":"main_file","creator":"dernst","file_id":"20314","access_level":"open_access","success":1,"checksum":"90a32defed34787e771a5c1623b6b0d2","file_size":295466,"content_type":"application/pdf","date_updated":"2025-09-09T08:10:13Z","file_name":"2025_NeuS_Kresse.pdf","date_created":"2025-09-09T08:10:13Z"}],"scopus_import":"1","publication_identifier":{"eissn":["2640-3498"]},"publisher":"ML Research Press","author":[{"first_name":"Fabian","full_name":"Kresse, Fabian","last_name":"Kresse","id":"faff3c84-23f6-11ef-9085-e5187b51c604"},{"id":"20aa2ae8-f2f1-11ed-bbfa-8205053f1342","last_name":"Yu","full_name":"Yu, Zhengqi","orcid":"0000-0002-4993-773X","first_name":"Zhengqi"},{"first_name":"Christoph","orcid":"0000-0001-8622-7887","full_name":"Lampert, Christoph","last_name":"Lampert","id":"40C20FD2-F248-11E8-B48F-1D18A9856A87"},{"full_name":"Henzinger, Thomas A","last_name":"Henzinger","id":"40876CD8-F248-11E8-B48F-1D18A9856A87","first_name":"Thomas A","orcid":"0000-0002-2985-7724"}],"title":"Logic gate neural networks are good for verification","quality_controlled":"1","file_date_updated":"2025-09-09T08:10:13Z","date_created":"2025-09-07T22:01:34Z","_id":"20296","OA_place":"publisher","corr_author":"1","language":[{"iso":"eng"}],"year":"2025","acknowledged_ssus":[{"_id":"ScienComp"}],"intvolume":"       288","conference":{"location":"Philadephia, PA, United States","name":"NeuS: International Conferenceon Neuro-Symbolic Systems","end_date":"2025-05-30","start_date":"2025-05-28"},"oa":1,"volume":288,"publication":"2nd International Conferenceon Neuro-Symbolic Systems","acknowledgement":"This work is supported in part by the ERC grant under Grant No. ERC-2020-AdG 101020093 and\r\nthe Austrian Science Fund (FWF) [10.55776/COE12]. This research was supported by the Scientific\r\nService Units (SSU) of ISTA through resources provided by Scientific Computing (SciComp).","arxiv":1,"day":"01","abstract":[{"lang":"eng","text":"Learning-based systems are increasingly deployed across various domains, yet the complexity of traditional neural networks poses significant challenges for formal verification. Unlike conventional neural networks, learned Logic Gate Networks (LGNs) replace multiplications with Boolean logic gates, yielding a sparse, netlist-like architecture that is inherently more amenable to symbolic verification, while still delivering promising performance. In this paper, we introduce a SAT encoding for verifying global robustness and fairness in LGNs. We evaluate our method on five benchmark datasets, including a newly constructed 5-class variant, and find that LGNs are both verification-friendly and maintain strong predictive performance."}],"month":"06","type":"conference","external_id":{"arxiv":["2505.19932"]},"oa_version":"Published Version","citation":{"apa":"Kresse, F., Yu, E., Lampert, C., &#38; Henzinger, T. A. (2025). Logic gate neural networks are good for verification. In <i>2nd International Conferenceon Neuro-Symbolic Systems</i> (Vol. 288). Philadephia, PA, United States: ML Research Press.","mla":"Kresse, Fabian, et al. “Logic Gate Neural Networks Are Good for Verification.” <i>2nd International Conferenceon Neuro-Symbolic Systems</i>, vol. 288, 26, ML Research Press, 2025.","chicago":"Kresse, Fabian, Emily Yu, Christoph Lampert, and Thomas A Henzinger. “Logic Gate Neural Networks Are Good for Verification.” In <i>2nd International Conferenceon Neuro-Symbolic Systems</i>, Vol. 288. ML Research Press, 2025.","ista":"Kresse F, Yu E, Lampert C, Henzinger TA. 2025. Logic gate neural networks are good for verification. 2nd International Conferenceon Neuro-Symbolic Systems. NeuS: International Conferenceon Neuro-Symbolic Systems, PMLR, vol. 288, 26.","ama":"Kresse F, Yu E, Lampert C, Henzinger TA. Logic gate neural networks are good for verification. In: <i>2nd International Conferenceon Neuro-Symbolic Systems</i>. Vol 288. ML Research Press; 2025.","ieee":"F. Kresse, E. Yu, C. Lampert, and T. A. Henzinger, “Logic gate neural networks are good for verification,” in <i>2nd International Conferenceon Neuro-Symbolic Systems</i>, Philadephia, PA, United States, 2025, vol. 288.","short":"F. Kresse, E. Yu, C. Lampert, T.A. Henzinger, in:, 2nd International Conferenceon Neuro-Symbolic Systems, ML Research Press, 2025."},"ddc":["000"],"date_published":"2025-06-01T00:00:00Z","OA_type":"diamond","project":[{"grant_number":"101020093","name":"Vigilant Algorithmic Monitoring of Software","call_identifier":"H2020","_id":"62781420-2b32-11ec-9570-8d9b63373d4d"},{"_id":"d8f03aaa-b035-11f1-8588-d5147fa879e0","grant_number":"COE12","name":"Bilateral Artificial Intelligence (Lampert)"}],"status":"public","article_number":"26","alternative_title":["PMLR"],"department":[{"_id":"ChLa"},{"_id":"ToHe"}],"article_processing_charge":"No","has_accepted_license":"1","publication_status":"published","date_updated":"2026-09-16T07:02:18Z","ec_funded":1,"user_id":"2DF688A6-F248-11E8-B48F-1D18A9856A87"},{"article_processing_charge":"No","department":[{"_id":"KrCh"}],"alternative_title":["PMLR"],"user_id":"2DF688A6-F248-11E8-B48F-1D18A9856A87","ec_funded":1,"publication_status":"published","date_updated":"2026-09-16T07:01:32Z","month":"05","abstract":[{"text":"LocalSGD and SCAFFOLD are widely used methods in distributed stochastic optimization, with numerous applications in machine learning, large-scale data processing, and federated learning. However, rigorously establishing their theoretical advantages over simpler methods, such as minibatch SGD (MbSGD), has proven challenging, as existing analyses often rely on strong assumptions, unrealistic premises, or overly restrictive scenarios.\r\n\r\nIn this work, we revisit the convergence properties of LocalSGD and SCAFFOLD under a variety of existing or weaker conditions, including gradient similarity, Hessian similarity, weak convexity, and Lipschitz continuity of the Hessian. Our analysis shows that (i) LocalSGD achieves faster convergence compared to MbSGD for weakly convex functions without requiring stronger gradient similarity assumptions; (ii) LocalSGD benefits significantly from higher-order similarity and smoothness; and (iii) SCAFFOLD demonstrates faster convergence than MbSGD for a broader class of non-quadratic functions. These theoretical insights provide a clearer understanding of the conditions under which LocalSGD and SCAFFOLD outperform MbSGD.","lang":"eng"}],"oa_version":"Preprint","type":"conference","external_id":{"arxiv":["2501.04443"]},"date_published":"2025-05-01T00:00:00Z","citation":{"ieee":"R. Luo, S. U. Stich, S. Horváth, and M. Takáč, “Revisiting LocalSGD and SCAFFOLD: Improved rates and missing analysis,” in <i>The 28th International Conference on Artificial Intelligence and Statistics</i>, Mai Khao, Thailand, 2025, vol. 258, pp. 2539–2547.","short":"R. Luo, S.U. Stich, S. Horváth, M. Takáč, in:, The 28th International Conference on Artificial Intelligence and Statistics, ML Research Press, 2025, pp. 2539–2547.","ama":"Luo R, Stich SU, Horváth S, Takáč M. Revisiting LocalSGD and SCAFFOLD: Improved rates and missing analysis. In: <i>The 28th International Conference on Artificial Intelligence and Statistics</i>. Vol 258. ML Research Press; 2025:2539-2547.","chicago":"Luo, Ruichen, Sebastian U. Stich, Samuel Horváth, and Martin Takáč. “Revisiting LocalSGD and SCAFFOLD: Improved Rates and Missing Analysis.” In <i>The 28th International Conference on Artificial Intelligence and Statistics</i>, 258:2539–47. ML Research Press, 2025.","ista":"Luo R, Stich SU, Horváth S, Takáč M. 2025. Revisiting LocalSGD and SCAFFOLD: Improved rates and missing analysis. The 28th International Conference on Artificial Intelligence and Statistics. AISTATS: Conference on Artificial Intelligence and Statistics, PMLR, vol. 258, 2539–2547.","apa":"Luo, R., Stich, S. U., Horváth, S., &#38; Takáč, M. (2025). Revisiting LocalSGD and SCAFFOLD: Improved rates and missing analysis. In <i>The 28th International Conference on Artificial Intelligence and Statistics</i> (Vol. 258, pp. 2539–2547). Mai Khao, Thailand: ML Research Press.","mla":"Luo, Ruichen, et al. “Revisiting LocalSGD and SCAFFOLD: Improved Rates and Missing Analysis.” <i>The 28th International Conference on Artificial Intelligence and Statistics</i>, vol. 258, ML Research Press, 2025, pp. 2539–47."},"status":"public","project":[{"name":"Formal Methods for Stochastic Models: Algorithms and Applications","grant_number":"863818","call_identifier":"H2020","_id":"0599E47C-7A3F-11EA-A408-12923DDC885E"},{"grant_number":"COE12","name":"Bilateral Artificial Intelligence (Chatterjee)","_id":"4029cfc7-b034-11f1-9e55-88ab2ff3b6ee"}],"main_file_link":[{"url":"https://doi.org/10.48550/arXiv.2501.04443","open_access":"1"}],"OA_type":"green","year":"2025","OA_place":"repository","language":[{"iso":"eng"}],"conference":{"start_date":"2025-05-03","end_date":"2025-05-05","location":"Mai Khao, Thailand","name":"AISTATS: Conference on Artificial Intelligence and Statistics"},"oa":1,"intvolume":"       258","volume":258,"publication":"The 28th International Conference on Artificial Intelligence and Statistics","acknowledgement":"The authors thank for the helpful discussions with Eduard Gorbunov, Kumar Kshitij Patel, Anton\r\nRodomanov, and Ali Zindari during the preparation of this work. This work was partially done during the first author’s stays at CISPA and at MBZUAI. The first author also acknowledges ERC CoG 863818 (ForM-SMArt) and Austrian Science Fund (FWF) 10.55776/COE12.","day":"01","arxiv":1,"publication_identifier":{"eissn":["2640-3498"]},"publisher":"ML Research Press","scopus_import":"1","title":"Revisiting LocalSGD and SCAFFOLD: Improved rates and missing analysis","quality_controlled":"1","author":[{"first_name":"Ruichen","id":"b391db08-1ffe-11ee-8b67-d18ddcfb5a14","full_name":"Luo, Ruichen","last_name":"Luo"},{"last_name":"Stich","full_name":"Stich, Sebastian U.","first_name":"Sebastian U."},{"full_name":"Horváth, Samuel","last_name":"Horváth","first_name":"Samuel"},{"full_name":"Takáč, Martin","last_name":"Takáč","first_name":"Martin"}],"page":"2539-2547","_id":"20302","date_created":"2025-09-07T22:01:35Z"},{"publication":"42nd International Conference on Machine Learning","volume":267,"acknowledgement":"AP. This project has received funding from the European Union’s Horizon 2020 research and innovation programme under the Marie Skłodowska-Curie Grant Agreement No. 101034413.\r\nFL. This research was funded in whole or in part by the Austrian Science Fund (FWF) 10.55776/COE12. For open access purposes, the author has applied a CC BY public\r\ncopyright license to any author accepted manuscript version arising from this submission.","day":"01","arxiv":1,"year":"2025","corr_author":"1","OA_place":"publisher","language":[{"iso":"eng"}],"conference":{"start_date":"2025-07-13","end_date":"2025-07-19","name":"ICML: International Conference on Machine Learning","location":"Vancouver, Canada"},"oa":1,"intvolume":"       267","file_date_updated":"2025-12-16T12:21:49Z","date_created":"2025-12-14T23:02:04Z","_id":"20817","page":"48962-48973","publication_identifier":{"eissn":["2640-3498"]},"publisher":"ML Research Press","file":[{"file_size":993381,"content_type":"application/pdf","date_updated":"2025-12-16T12:21:49Z","date_created":"2025-12-16T12:21:49Z","file_name":"2025_ICML_Pervez.pdf","creator":"dernst","relation":"main_file","file_id":"20827","access_level":"open_access","success":1,"checksum":"933cb673fb41416f537278fb990df6c3"}],"scopus_import":"1","title":"Mechanistic PDE networks for discovery of governing equations","quality_controlled":"1","author":[{"full_name":"Pervez, Adeel A","id":"fca6d90c-d47f-11ee-bc87-93ff51604981","last_name":"Pervez","first_name":"Adeel A"},{"full_name":"Gavves, Efstratios","last_name":"Gavves","first_name":"Efstratios"},{"full_name":"Locatello, Francesco","last_name":"Locatello","id":"26cfd52f-2483-11ee-8040-88983bcc06d4","first_name":"Francesco","orcid":"0000-0002-4850-0683"}],"user_id":"2DF688A6-F248-11E8-B48F-1D18A9856A87","ec_funded":1,"publication_status":"published","date_updated":"2026-09-16T07:17:18Z","article_processing_charge":"No","has_accepted_license":"1","department":[{"_id":"FrLo"}],"alternative_title":["PMLR"],"date_published":"2025-05-01T00:00:00Z","tmp":{"legal_code_url":"https://creativecommons.org/licenses/by/4.0/legalcode","name":"Creative Commons Attribution 4.0 International Public License (CC-BY 4.0)","short":"CC BY (4.0)","image":"/images/cc_by.png"},"ddc":["000"],"citation":{"short":"A.A. Pervez, E. Gavves, F. Locatello, in:, 42nd International Conference on Machine Learning, ML Research Press, 2025, pp. 48962–48973.","ieee":"A. A. Pervez, E. Gavves, and F. Locatello, “Mechanistic PDE networks for discovery of governing equations,” in <i>42nd International Conference on Machine Learning</i>, Vancouver, Canada, 2025, vol. 267, pp. 48962–48973.","chicago":"Pervez, Adeel A, Efstratios Gavves, and Francesco Locatello. “Mechanistic PDE Networks for Discovery of Governing Equations.” In <i>42nd International Conference on Machine Learning</i>, 267:48962–73. ML Research Press, 2025.","ista":"Pervez AA, Gavves E, Locatello F. 2025. Mechanistic PDE networks for discovery of governing equations. 42nd International Conference on Machine Learning. ICML: International Conference on Machine Learning, PMLR, vol. 267, 48962–48973.","ama":"Pervez AA, Gavves E, Locatello F. Mechanistic PDE networks for discovery of governing equations. In: <i>42nd International Conference on Machine Learning</i>. Vol 267. ML Research Press; 2025:48962-48973.","apa":"Pervez, A. A., Gavves, E., &#38; Locatello, F. (2025). Mechanistic PDE networks for discovery of governing equations. In <i>42nd International Conference on Machine Learning</i> (Vol. 267, pp. 48962–48973). Vancouver, Canada: ML Research Press.","mla":"Pervez, Adeel A., et al. “Mechanistic PDE Networks for Discovery of Governing Equations.” <i>42nd International Conference on Machine Learning</i>, vol. 267, ML Research Press, 2025, pp. 48962–73."},"project":[{"call_identifier":"H2020","_id":"fc2ed2f7-9c52-11eb-aca3-c01059dda49c","grant_number":"101034413","name":"IST-BRIDGE: International postdoctoral program"},{"grant_number":"COE12","name":"Bilateral Artificial Intelligence (Locatello)","_id":"a392d8f0-b034-11f1-b88e-d3025c6734f6"}],"status":"public","OA_type":"gold","month":"05","abstract":[{"lang":"eng","text":"We present Mechanistic PDE Networks -- a model for discovery of governing partial differential equations from data. Mechanistic PDE Networks represent spatiotemporal data as space-time dependent linear partial differential equations in neural network hidden representations. The represented PDEs are then solved and decoded for specific tasks. The learned PDE representations naturally express the spatiotemporal dynamics in data in neural network hidden space, enabling increased modeling power. Solving the PDE representations in a compute and memory-efficient way, however, is a significant challenge. We develop a native, GPU-capable, parallel, sparse and differentiable multigrid solver specialized for linear partial differential equations that acts as a module in Mechanistic PDE Networks. Leveraging the PDE solver we propose a discovery architecture that can discovers nonlinear PDEs in complex settings, while being robust to noise. We validate PDE discovery on a number of PDEs including reaction-diffusion and Navier-Stokes equations."}],"related_material":{"link":[{"url":"https://github.com/ alpz/mech-nn-discovery-pde","relation":"software"}]},"oa_version":"Published Version","external_id":{"arxiv":["2502.18377"]},"type":"conference"},{"quality_controlled":"1","title":"Neural collapse beyond the unconstrained features model: Landscape, dynamics, and generalization in the mean-field regime","author":[{"id":"1a5914c2-896a-11ed-bdf8-fb80621a0635","full_name":"Wu, Diyuan","last_name":"Wu","first_name":"Diyuan"},{"id":"27EB676C-8706-11E9-9510-7717E6697425","full_name":"Mondelli, Marco","last_name":"Mondelli","first_name":"Marco","orcid":"0000-0002-3242-7020"}],"publisher":"ML Research Press","publication_identifier":{"eissn":["2640-3498"]},"file":[{"success":1,"checksum":"c5ce8b1c83e33dc3a11122f4910deb67","creator":"dernst","relation":"main_file","file_id":"21337","access_level":"open_access","file_size":3994385,"content_type":"application/pdf","file_name":"2025_ICML_Wu.pdf","date_created":"2026-02-19T08:28:22Z","date_updated":"2026-02-19T08:28:22Z"}],"date_created":"2026-02-18T12:02:45Z","_id":"21326","page":"67499-67536","file_date_updated":"2026-02-19T08:28:22Z","oa":1,"conference":{"name":"ICML: International Conference on Machine Learning","location":"Vancouver, Canada","end_date":"2025-07-19","start_date":"2025-07-13"},"intvolume":"       267","year":"2025","language":[{"iso":"eng"}],"OA_place":"publisher","corr_author":"1","arxiv":1,"day":"30","acknowledgement":"This research was funded in whole or in part by the Austrian Science Fund (FWF) 10.55776/COE12. For the purpose of open access, the authors have applied a CC BY public\r\ncopyright license to any Author Accepted Manuscript version arising from this submission. The authors would like to thank Peter Sukenık for general helpful discussions and for pointing out that all the stationary points are approximately proportional in the case without entropic regularization. ","volume":267,"publication":"Proceedings of the 42nd International Conference on Machine Learning","oa_version":"Published Version","external_id":{"arxiv":["2501.19104"]},"type":"conference","month":"07","abstract":[{"lang":"eng","text":"Neural Collapse is a phenomenon where the last-layer representations of a well-trained neural network converge to a highly structured geometry. In this paper, we focus on its first (and most basic) property, known as NC1: the within-class variability vanishes. While prior theoretical studies establish the occurrence of NC1 via the data-agnostic unconstrained features model, our work adopts a data-specific perspective, analyzing NC1 in a three-layer neural network, with the first two layers operating in the mean-field regime and followed by a linear layer. In particular, we establish a fundamental connection between NC1 and the loss landscape: we prove that points with small empirical loss and gradient norm (thus, close to being stationary) approximately satisfy NC1, and the closeness to NC1 is controlled by the residual loss and gradient norm. We then show that (i) gradient flow on the mean squared error converges to NC1 solutions with small empirical loss, and (ii) for well-separated data distributions, both NC1 and vanishing test loss are achieved simultaneously. This aligns with the empirical observation that NC1 emerges during training while models attain near-zero test error. Overall, our results demonstrate that NC1 arises from gradient training due to the properties of the loss landscape, and they show the co-occurrence of NC1 and small test error for certain data distributions."}],"status":"public","project":[{"_id":"74caaef7-b034-11f1-8f2d-e0e993bb422e","grant_number":"COE12","name":"Bilateral Artificial Intelligence (Mondelli)"}],"OA_type":"gold","date_published":"2025-07-30T00:00:00Z","tmp":{"legal_code_url":"https://creativecommons.org/licenses/by/4.0/legalcode","name":"Creative Commons Attribution 4.0 International Public License (CC-BY 4.0)","short":"CC BY (4.0)","image":"/images/cc_by.png"},"ddc":["000"],"citation":{"mla":"Wu, Diyuan, and Marco Mondelli. “Neural Collapse beyond the Unconstrained Features Model: Landscape, Dynamics, and Generalization in the Mean-Field Regime.” <i>Proceedings of the 42nd International Conference on Machine Learning</i>, vol. 267, ML Research Press, 2025, pp. 67499–536.","apa":"Wu, D., &#38; Mondelli, M. (2025). Neural collapse beyond the unconstrained features model: Landscape, dynamics, and generalization in the mean-field regime. In <i>Proceedings of the 42nd International Conference on Machine Learning</i> (Vol. 267, pp. 67499–67536). Vancouver, Canada: ML Research Press.","ista":"Wu D, Mondelli M. 2025. Neural collapse beyond the unconstrained features model: Landscape, dynamics, and generalization in the mean-field regime. Proceedings of the 42nd International Conference on Machine Learning. ICML: International Conference on Machine Learning, PMLR, vol. 267, 67499–67536.","chicago":"Wu, Diyuan, and Marco Mondelli. “Neural Collapse beyond the Unconstrained Features Model: Landscape, Dynamics, and Generalization in the Mean-Field Regime.” In <i>Proceedings of the 42nd International Conference on Machine Learning</i>, 267:67499–536. ML Research Press, 2025.","ama":"Wu D, Mondelli M. Neural collapse beyond the unconstrained features model: Landscape, dynamics, and generalization in the mean-field regime. In: <i>Proceedings of the 42nd International Conference on Machine Learning</i>. Vol 267. ML Research Press; 2025:67499-67536.","short":"D. Wu, M. Mondelli, in:, Proceedings of the 42nd International Conference on Machine Learning, ML Research Press, 2025, pp. 67499–67536.","ieee":"D. Wu and M. Mondelli, “Neural collapse beyond the unconstrained features model: Landscape, dynamics, and generalization in the mean-field regime,” in <i>Proceedings of the 42nd International Conference on Machine Learning</i>, Vancouver, Canada, 2025, vol. 267, pp. 67499–67536."},"has_accepted_license":"1","article_processing_charge":"No","department":[{"_id":"MaMo"}],"alternative_title":["PMLR"],"user_id":"2DF688A6-F248-11E8-B48F-1D18A9856A87","date_updated":"2026-09-16T07:25:42Z","publication_status":"published"}]