@article{20975, abstract = {Galaxy clusters are the most massive, gravitationally bound structures in the Universe. They emerged through hierarchical structure formation of large-scale dark matter and baryon overdensities. Early galaxy ‘proto-clusters’ are believed to have substantially contributed to the cosmic star-formation rate density and served as ‘hotspots’ for the reionization of the intergalactic medium. Our understanding of the formation of these structures at the earliest cosmic epochs is, however, limited to sparse observations of their galaxy members or is based on phenomenological models and cosmological simulations. Here we report the detection of a large and coherent structure of neutral atomic hydrogen gas (H i) extending from a galaxy proto-cluster at redshift z = 5.4, one billion years after the Big Bang. The presence of this H i gas is revealed by strong damped Lyman-α absorption features observed in several background-galaxy spectra. Although the sight lines overall probe a large range in H i column densities, NHI = 1020 cm−2 to 1023.5 cm−2, they are similar across nearby sight lines, demonstrating that they probe the same dense neutral gas. This observation of a dense large-scale overdensity of cold neutral gas challenges current cosmological simulations and has strong implications for the reionization topology of the Universe.}, author = {Heintz, Kasper E. and Bennett, Jake S. and Oesch, Pascal A. and Sneppen, Albert and Rennehan, Douglas and Pollock, Clara L. and Witstok, Joris and Smit, Renske and Vejlgaard, Simone and Terp, Chamilla and Koca, Umran S. and Brammer, Gabriel B. and Finlator, Kristian and Hayes, Matthew J. and Sijacki, Debora and Naidu, Rohan P. and Matthee, Jorryt J and Valentino, Francesco and Tanvir, Nial R. and Jakobsson, Páll and Laursen, Peter and Watson, Darach J. and Davé, Romeel and Keating, Laura C. and Covelo-Paz, Alba}, issn = {2397-3366}, journal = {Nature Astronomy}, pages = {448--456}, publisher = {Springer Nature}, title = {{A dense web of neutral gas in a galaxy proto-cluster post-reionization}}, doi = {10.1038/s41550-025-02745-x}, volume = {10}, year = {2026}, } @article{21036, abstract = {Forests under livestock grazing sustain important ecosystem services but face potential trade-offs between production and ecological integrity. While the effects of grazing on individual forest attributes are well documented, their integrated consequences remain poorly understood, particularly in temperate forest ecosystems. We evaluated the combined influence of livestock grazing intensity and canopy cover on individual attributes and ecosystem multifunctionality in native Nothofagus forests of Tierra del Fuego, Argentina. Across eight ranches spanning two agroecological regions (Ecotone and Mountain Range), we quantified forest regeneration, understorey richness and biomass, and soil properties, integrating them into a multifunctionality index. Using generalized linear mixed models, we found strong context-dependence: in the Mountain Range, higher grazing intensity reduced seedling and sapling density, organic matter content, coarse woody debris, and overall multifunctionality. In the Ecotone, these effects of livestock use intensity were attenuated, and canopy cover diminished sapling density and multifunctionality, but moderate cover enhanced understorey. Our results extend multifunctionality research from grazed grasslands to grazed temperate forests and show that ecological responses and trade-offs vary across landscape units. We conclude that the Mountain Range is more vulnerable to grazing, requiring stricter management, whereas the Ecotone retains greater capacity to sustain multifunctionality under controlled livestock use intensity. These findings underscore the importance of region-specific silvopastoral strategies that reconcile food production with forest conservation in southern Patagonia and comparable temperate forest landscapes worldwide.}, author = {Rodríguez, Paula and Cruz Alonso, Verónica and Romano, Silvina and Bustamante, Gimena and Soler Schaller, Rosina Matilde}, issn = {0167-8809}, journal = {Agriculture, Ecosystems and Environment}, publisher = {Elsevier}, title = {{Context-dependent effects of livestock grazing on forest attributes and ecosystem multifunctionality in Nothofagus forests}}, doi = {10.1016/j.agee.2026.110219}, volume = {400}, year = {2026}, } @article{21132, abstract = {We unify the variational hypocoercivity framework established by D. Albritton, S. Armstrong, J.-C. Mourrat, and M. Novack [2], with the notion of second-order lifts of reversible diffusion processes, recently introduced by A. Eberle and the second author [30]. We give an abstract, yet fully constructive, presentation of the theory, so that it can be applied to a large class of linear kinetic equations. As this hypocoercivity technique does not twist the reference norm, we can recover accurate and sharp convergence rates in various models. Among those, adaptive Langevin dynamics (ALD) is discussed in full detail and we show that for near-quadratic potentials, with suitable choices of parameters, it is a near-optimal second-order lift of the overdamped Langevin dynamics. As a further consequence, we observe that the Generalised Langevin Equation (GLE) is also a second-order lift, as the standard (kinetic) Langevin dynamics are, of the overdamped Langevin dynamics. Then, convergence of (GLE) cannot exceed ballistic speed, i.e. the square root of the rate of the overdamped regime. We illustrate this phenomenon with explicit computations in a benchmark Gaussian case.}, author = {Brigati, Giovanni and Lörler, Francis and Wang, Lihan}, issn = {1937-5077}, journal = {Kinetic and Related Models}, pages = {34--55}, publisher = {American Institute of Mathematical Sciences}, title = {{Hypocoercivity meets lifts}}, doi = {10.3934/krm.2025020}, volume = {20}, year = {2026}, } @article{20818, abstract = {This study demonstrates that Marchantia non-canonical PINs are predominantly localized to the plasma membrane, with MpPINX and MpPINW exhibiting asymmetric distribution. A newly identified miniW domain within the MpPINW hydrophilic loop governs subcellular trafficking and asymmetric PM localization of non-canonical PINs in Marchantia.}, author = {Tang, Han and Smoljan, Adrijana and Zou, Minxia and Zhang, Yuzhou and Lu, Kuan Ju and Friml, Jiří}, issn = {1365-3040}, journal = {Plant Cell and Environment}, number = {3}, pages = {1505--1508}, publisher = {Wiley}, title = {{The miniW domain directs polarized membrane localization of non-canonical PINs in Marchantia polymorpha}}, doi = {10.1111/pce.70295}, volume = {49}, year = {2026}, } @article{21453, abstract = {1. Collective behaviours are a fascinating study area due to the emergent properties that can only arise in groups of interacting individuals. However, their quantitative study is often impaired by technical difficulties, creating either low-quality and sparse data or impractical data amounts, particularly when capturing large groups over long periods of time. Common challenges arise from recording group members with as little obscuring of each other as possible, as well as in generating manageable data amounts with as high as possible information content. 2. We here provide a multicomponent system that allows to record, analyse and simulate the long-term spatiotemporal activity patterns of insect collectives, especially ant colonies. Our Ant Observing System, ALTAA, comprises a flat-nest design to prevent occlusion of individuals, a recording system running on a low-power single-board-computer, and a set of computer programmes performing quantitative analyses to guide the formation and validation of rules underlying the observed collective patterns. Our system is scalable in that it allows parallel, continuous observation of a high number of colonies using low memory space, with colony maintenance requirements (e.g. feeding, nest humidity) being achieved at lowest possible disturbance by the experimenter. 3. We showcase the potential of the system in a study using the black garden ant, Lasius niger, where we analyse the spatiotemporal effects of different group sizes (1, 6, 10 ants), brood (larvae) presence or absence, as well as of different nest geometries, over a period of 1 week. We show that the ants' motion activity has a weak periodicity in the range of 20 to 120 min promoted by larval presence, and that ants are spatially attracted to their larvae, the water source and the walls. We also find that the presence of nestmates lowers an individual ant's motion activity. Observed data are compared to simulations of the temporal activity of the ants. 4. ALTAA provides a powerful toolkit to quantify and interpret spatial and temporal collective activity patterns in (social) insects over extended periods.}, author = {Oh, Jinook and Cremer, Sylvia}, issn = {2041-210X}, journal = {Methods in Ecology and Evolution}, number = {4}, pages = {1218--1234}, publisher = {Wiley}, title = {{ALTAA: Analysis of long-term activity patterns in ant colonies}}, doi = {10.1111/2041-210x.70277}, volume = {17}, year = {2026}, } @article{21295, abstract = {Depending on the type of flow, the transition to turbulence can take one of two forms: either turbulence arises from a sequence of instabilities or from the spatial proliferation of transiently chaotic domains, a process analogous to directed percolation. The former scenario is commonly referred to as a supercritical transition and frequently encountered in flows destabilized by body forces, whereas the latter subcritical transition is common in shear flows. Both cases are inherently continuous in a sense that the transformation from ordered laminar to fully turbulent fluid motion is only accomplished gradually with flow speed. Here we show that these established transition types do not account for the more general setting of shear flows subject to body forces. The combination of the two continuous scenarios leads to the attenuation of spatial coupling; with increasing forcing amplitude, the transition becomes increasingly sharp and eventually discontinuous. We argue that the suppression of laminar–turbulent coexistence and the approach towards a discontinuous phase transition potentially apply to a broad range of situations including flows subject to, for example, buoyancy, centrifugal or electromagnetic forces.}, author = {Yang, Bowen and Zhuang, Yi and Yalniz, Gökhan and Vasudevan, Mukund and Marensi, Elena and Hof, Björn}, issn = {1745-2481}, journal = {Nature Physics}, pages = {424--429}, publisher = {Springer Nature}, title = {{Discontinuous transition to shear flow turbulence}}, doi = {10.1038/s41567-025-03166-3}, volume = {22}, year = {2026}, } @article{21483, abstract = {Embryogenesis in the model plant Arabidopsis thaliana provides a framework for understanding how cell polarity and patterning coordinate with hormonal signalling to establish the plant body plan. Following fertilisation, the zygote divides asymmetrically to generate apical and basal lineages, establishing the apical–basal axis that defines future shoot and root poles. Genetic and molecular analyses of classical mutants including gnom, monopteros (mp), bodenlos (bdl) and topless revealed that localised auxin biosynthesis, directional transport and downstream transcriptional responses are central to apical–basal axis establishment and organ initiation. The main components of this regulation are polarly localised PIN auxin transporters and downstream modules involving MONOPTEROS and WUSCHEL-RELATED HOMEOBOX transcription factors. Advances in microscopy have transformed the study of Arabidopsis embryogenesis: fluorescence-compatible clearing reagents and three-dimensional reconstructions now permit quantitative analyses of cell geometry, division orientation, and cytoskeletal dynamics. Live ovule imaging setups with confocal laser scanning and multiphoton microscopes enable real-time observation of embryo development, while laser-assisted cell ablation can be used to probe cell-to-cell communication and fate plasticity. Together, these methodological breakthroughs position Arabidopsis embryos as a prime model for dissecting the chemical and biophysical cues that shape plant development.}, author = {Babic, David and Zupunski, Milan and Friml, Jiří}, issn = {1469-8137}, journal = {New Phytologist}, number = {3}, pages = {1483--1491}, publisher = {Wiley}, title = {{Imaging and genetic toolbox to study Arabidopsis embryogenesis}}, doi = {10.1111/nph.71072}, volume = {250}, year = {2026}, } @article{21484, abstract = {An individual's phenotype reflects a complex interplay of the direct effects of their DNA, epigenetic modifications of their DNA induced by their parents, and indirect effects of their parents' DNA. Here, we derive how the genetic variance within a population is changed under the influence of indirect maternal, paternal and parent-of-origin effects under random mating. We also consider indirect effects of a sibling, in particular how the genetic variance is altered when looking at the phenotypic difference between two siblings. The calculations are then extended to include assortative mating (AM), which alters the variance by inducing increased homozygosity and correlations within and across loci. AM likely leads to covariance of parental genetic effects, a measure of the similarity of parents in the indirect effects they have on their children. We propose that this assortment for parental characteristics, where biological parents create similar environments for their children, can create shared parental effects across traits and the appearance of cross-trait AM. Our theory shows how the resemblance among relatives increases under both AM, indirect and parent-of-origin effects. When our model is used to predict correlations among relatives in human height, we find that explaining the patterns observed in real data requires both indirect genetic effects and assortative mating. The degree to which direct, indirect and epigenetic effects shape the phenotypic variance of complex traits remains an open question that requires large-scale family data to be resolved.}, author = {Krätschmer, Ilse and Robinson, Matthew Richard}, issn = {1943-2631}, journal = {Genetics}, number = {4}, publisher = {Oxford University Press}, title = {{A quantitative genetic model for indirect genetic effects and genomic imprinting under random and assortative mating}}, doi = {10.1093/genetics/iyag042}, volume = {232}, year = {2026}, } @article{21502, abstract = {The mammalian brain stores glucose, the main circulating energy substrate, as glycogen. In rodents, the cerebellum contains relatively high glycogen levels, yet its cellular and subcellular distribution remains poorly defined. Using monoclonal antibodies against glycogen, we examined its distribution in the mouse cerebellar cortex. Glycogen was predominantly localized to Bergmann glia (BG) processes in the molecular layer and was also detected in Purkinje cells (PCs), the principal cerebellar neurons. To assess the functional significance of cerebellar glycogen, we analyzed behavior in mice lacking glycogen synthase 1 (Gys1) in BG or PCs using a floxed Gys1 line. Gys1 deficiency in either PCs or GFAP-positive cells reduced anxiety-like behavior, whereas combined deletion caused PC degeneration and ataxia. These findings reveal a critical role for glycogen metabolism in both astrocytes and neurons in cerebellar function.}, author = {Akther, Sonam and Lee, Ashley Bomin and Konno, Ayumu and Asiminas, Antonis and Vittani, Marta and Mishima, Tsuneko and Hirai, Hirokazu and Meehan, Claire Francesca and Duran, Jordi and Guinovart, Joan and Ashida, Hitoshi and Morita, Tsuyoshi and Baba, Otto and Shigemoto, Ryuichi and Nedergaard, Maiken and Hirase, Hajime}, issn = {2589-0042}, journal = {iScience}, number = {4}, publisher = {Elsevier}, title = {{Distribution and functional significance of rodent cerebellar glycogen}}, doi = {10.1016/j.isci.2026.115192}, volume = {29}, year = {2026}, } @article{21486, abstract = {Sex-chromosome systems are highly variable across animals, but how they transition from one to another is not well understood. Diptera have undergone multiple sex-chromosome turnovers and expansions while maintaining their general chromosomal content, which makes them an ideal clade to study such transitions. We analyzed more than 100 dipteran whole-genome assemblies and identified 4 new lineages that underwent sex-chromosome turnover (in addition to the 5 previously reported). We find that the majority of turnovers happened in the group Schizophora, which tend to have fewer genes on Muller element F (the chromosome homologous to the ancestral insect X chromosome) than lower dipterans, a factor previously hypothesized to facilitate turnover. Most derived X chromosomes have higher GC content than autosomes, consistent with a high prevalence of male achiasmy in Diptera. In addition, an excess of gene movement out of the X is detected for most of these new X chromosomes, and many of these moved genes have high testis expression in Drosophila, suggesting that out-of-X gene movement contributes to the long-term demasculinization of X chromosomes.}, author = {Layana Franco, Lorena Alexandra and Toups, Melissa A and Vicoso, Beatriz}, issn = {2056-3744}, journal = {Evolution Letters}, number = {3}, publisher = {Oxford University Press}, title = {{Causes and consequences of sex-chromosome turnovers in Diptera}}, doi = {10.1093/evlett/qrag003}, volume = {10}, year = {2026}, } @article{21436, abstract = {The cobalt-intercalated transition metal dichalcogenide CoxTaS2 hosts a rich landscape of magnetic phases that depend sensitively on x. While the stoichiometric compound with x = 1/3 exhibits a single magnetic transition, samples with x≤0.325 display two transitions with an anomalous Hall effect (AHE) emerging in the lower temperature phase. Here, we resolve the spin structure in each phase by employing a suite of magneto-optical probes that include the discovery of anomalous magneto-birefringence: a spontaneous time-reversal sensitive rotation of the principal optic axes. A symmetry-based analysis identifies the AHE-active phase as an anisotropic (2+1)Q state, in which magnetic modulation at one wavevector (Q) differs in symmetry from that at the remaining two. The (2+1)Q state naturally exhibits scalar spin chirality as a mechanism for the AHE and expands the classification of multi-Q magnetic phases.}, author = {Kruppe, Jonathon and Rodriguez, Josue and Xu, Catherine and Analytis, James and Orenstein, Joseph and Sunko, Veronika}, issn = {2397-4648}, journal = {npj Quantum Materials}, publisher = {Springer Nature}, title = {{Anisotropic multi-Q order in CoxTaS2}}, doi = {10.1038/s41535-026-00856-w}, volume = {11}, year = {2026}, } @article{21472, abstract = {We study the ground state energy of a gas of spin 1/2 fermions with repulsive short-range interactions. We derive an upper bound that agrees, at low density e, with the Huang–Yang conjecture. The latter captures the first three terms in an asymptotic low-density expansion, and in particular the Huang–Yang correction term of order e^7/3. Our trial state is constructed using an adaptation of the bosonic Bogoliubov theory to the Fermi system, where the correlation structure of fermionic particles is incorporated by quasi-bosonic Bogoliubov transformations. In the latter, it is important to consider a modified zero-energy scattering equation that takes into account the presence of the Fermi sea, in the spirit of the Bethe–Goldstone equation.}, author = {Giacomelli, Emanuela L. and Hainzl, Christian and Nam, Phan Thành and Seiringer, Robert}, issn = {1097-0312}, journal = {Communications on Pure and Applied Mathematics}, number = {8}, pages = {1919--1972}, publisher = {Wiley}, title = {{The Huang–Yang formula for the low-density Fermi gas: Upper bound}}, doi = {10.1002/cpa.70040}, volume = {79}, year = {2026}, } @article{21504, abstract = {Selecting an appropriate divergence measure is a critical aspect of machine learning, as it directly impacts model performance. Among the most widely used, we find the Kullback–Leibler (KL) divergence, originally introduced in kinetic theory as a measure of relative entropy between probability distributions. Just as in machine learning, the ability to quantify the proximity of probability distributions plays a central role in kinetic theory. In this paper, we present a comparative review of divergence measures rooted in kinetic theory, highlighting their theoretical foundations and exploring their potential applications in machine learning and artificial intelligence.}, author = {Auricchio, Gennaro and Brigati, Giovanni and Giudici, Paolo and Toscani, Giuseppe}, issn = {1793-6314}, journal = {Mathematical Models and Methods in Applied Sciences}, number = {6}, pages = {1185--1233}, publisher = {World Scientific Publishing}, title = {{From kinetic theory to AI: A rediscovery of high-dimensional divergences and their properties}}, doi = {10.1142/S0218202526410010}, volume = {36}, year = {2026}, } @article{21721, abstract = {Swimming bacteria move through a fluid by actuating their moving body parts. They are force-free and can be described as hydrodynamic force dipoles: pushers or pullers. This modelling description is broadly used in biological physics and active matter research, and it has successfully predicted, for example, the superfluid behaviour of suspensions of pushers or the bend instability and emergence of turbulent flows in active nematics. However, this description accounts only for the translational motion of the swimming body and neglects the effects of hydrodynamic torque dipoles, which are relevant to bacteria with rotary motor-driven flagella, such as swimming Escherichia coli. Here we show that the torque dipole of confined swimming E. coli can power the persistent rotation of symmetric discs. The torque dipole leads to a traction force on the discs, an additive mechanism that is both contactless and independent of the orientation of the bacteria. Our results indicate that the torque dipole of swimming E. coli is notable in confined geometries, which is relevant to bacterial transport through porous materials, biofilms and the development of chiral fluids.}, author = {Grober, Daniel B and Dhar, Tanumoy and Saintillan, David and Palacci, Jérémie A}, issn = {1745-2481}, journal = {Nature Physics}, pages = {620--627}, publisher = {Springer Nature}, title = {{The hydrodynamic torque dipole from rotary bacterial flagella powers symmetric discs}}, doi = {10.1038/s41567-026-03189-4}, volume = {22}, year = {2026}, } @article{21716, abstract = {Male germline development in plants is highly sensitive to heat stress, with elevated temperatures frequently impairing male fertility and consequently reducing seed production. Indeed, recent global warming has decreased major crop yields, emphasizing the urgent need to elucidate the molecular and cellular mechanisms underlying heat-induced male sterility. This review synthesizes current knowledge on how heat stress disrupts microsporogenesis and microgametogenesis, and how plants counteract these stresses through diverse thermotolerance mechanisms. We emphasize temperature-sensitive processes, including meiotic progression in male germ cells, programmed cell death of somatic tapetal nurse cells, and post-meiotic pollen tube development. We further discuss how epigenetic regulators enhance thermotolerance by reprogramming DNA methylation landscapes and modulating histone variant distribution. Finally, we propose future directions aimed at understanding the mechanisms of reproductive thermotolerance from the epigenetic perspective.}, author = {Nagai, Hiroki and Feng, Xiaoqi}, issn = {1879-0356}, journal = {Current Opinion in Plant Biology}, number = {6}, publisher = {Elsevier}, title = {{Genetic and epigenetic mechanisms underlying male reproductive thermotolerance}}, doi = {10.1016/j.pbi.2026.102881}, volume = {91}, year = {2026}, } @article{21726, abstract = {Quantum control of the many-body wavefunction is a central challenge in quantum materials research, as it could yield a precise control knob to manipulate emergent phenomena. Floquet engineering, the coherent dressing of quantum states with periodic non-resonant optical fields, has become an important strategy for quantum control. Most applications to solid-state systems have targeted weakly interacting or single-ion states, leaving the manipulation of many-body wavefunctions largely unexplored. Here we use Floquet engineering to achieve quantum control of a strongly correlated Hubbard exciton in the one-dimensional Mott insulator Sr2CuO3. A non-resonant mid-infrared optical field coherently dresses the exciton wavefunction, driving its rotation between bright and dark states. We use resonant third-harmonic generation to quantify ultrafast π/2 rotations on the Bloch sphere spanned by these exciton states. Our work advances the quest towards programmable control of correlated states and exciton-based quantum sensing.}, author = {Baykusheva, Denitsa Rangelova and Carmichael, Deven and Weber, Clara S. and Lu, I. Te and Glerean, Filippo and Meng, Tepie and De Oliveira, Pedro B.M. and Homes, Christopher C. and Zaliznyak, Igor A. and Gu, G. D. and Dean, Mark P.M. and Rubio, Angel and Kennes, Dante M. and Claassen, Martin and Mitrano, Matteo}, issn = {1476-4660}, journal = {Nature Materials}, pages = {937--943}, publisher = {Springer Nature}, title = {{Quantum control of Hubbard excitons}}, doi = {10.1038/s41563-026-02517-6}, volume = {25}, year = {2026}, } @article{21730, abstract = {Hydrogen peroxide (H2O2) is a crucial member of the reactive oxygen species (ROS) family, playing roles in cellular signalling and immune responses in human health. Moreover, it is a potential biomarker of diabetes when present in aberrant concentrations. Therefore, monitoring trace levels of H2O2 has become a research hotspot for analytical and sensor chemists. In this context, we report a rhodamine-based fluorescent probe (RN), which shows excellent fluorescent enhancement at 555 nm upon the addition of H2O2 along with a low limit of detection (LOD) of 0.67 ppm and fast response (∼2 min). The probe is highly selective for H2O2, showing no fluorescence enhancement with other ROS. RN is synthesised in a one-pot chemical reaction using rhodamine 6G (R6G) and 4,7,10-trioxa-1,13-tridecanediamine (TTDA). H2O2 detection in pre-treated milk samples proves its real-world viability. We found that RN shows low cytotoxicity, which allowed us to successfully explore its potential to monitor H2O2 generation in a diabetic L929 skin cell line and diabetic mice liver tissue. This result demonstrates promising features for assessing early diabetic progression through fluorescence imaging.}, author = {Mondal, Moumita and Ghorai, Pravat and Samadder, Asmita and Freunberger, Stefan Alexander and Banerjee, Priyabrata}, issn = {2050-7518}, journal = {Journal of Materials Chemistry B}, number = {17}, pages = {5314--5322}, publisher = {Royal Society of Chemistry}, title = {{H2O2 responsive rhodamine-based probe for monitoring early-stage diabetes diagnosis}}, doi = {10.1039/d5tb02687c}, volume = {14}, year = {2026}, } @phdthesis{21854, abstract = {As neural-network-based models grow both in size and popularity, interest has grown in making the models smaller and more efficient to train. To that end, many methods have been proposed to prune models by reducing their number of nonzero parameters. Additionally, parameter-efficient fine-tuning, in which a much smaller number of parameters than the total contained in the model is updated during training, has become very popular, especially in the space of Large Language Models. At the same time, the increasingly routine deployment of machine learning in real-world applications has spurred a drive to make them more trustworthy - in the sense of, among other things, being unbiased, interpretable, and editable. In this thesis, we examine the interplay between efficiency and trustworthiness. First, we analyze the effects of model pruning on bias in computer vision models, demonstrating that increased sparsity leads to greater bias, largely as a function of increased model uncertainty in marginal cases. Based on this observation, we propose several bias mitigation techniques. Then, we demonstrate that example-specific model pruning can improve model interpretation methods while improving pruning efficiency to make example-specific model pruning feasible in real time. Then, we investigate the effectiveness of parameter-efficient and data-efficient model personalization via fine-tuning, demonstrating that it is highly feasible with very small computational and data resources. Finally, we consider efficiency in editing model knowledge using a custom synthetic data framework, demonstrating that parameter-efficient, low-rank fine-tuning frequently outperforms full-rank fine-tuning, and, additionally, that restricting which model blocks are fine-tuned frequently improves results. Together, the results in this thesis provide new insights and techniques for combining trustworthiness and efficiency during neural network inference and training. }, author = {Iofinova, Eugenia B}, issn = {2663-337X}, pages = {237}, publisher = {Institute of Science and Technology Austria}, title = {{On the utility and effects of efficiency in artificial neural networks}}, doi = {10.15479/AT-ISTA-21854}, year = {2026}, } @misc{21857, abstract = {The availability of powerful open-source large language models (LLMs) opens exciting use cases, such as using personal data to fine-tune these models to imitate a user’s unique writing style. Two key requirements for this functionality are personalization–in the sense that the output should recognizably reflect the user’s own writing style—and privacy–users may justifiably be wary of uploading extremely personal data, such as their email archive, to a third-party service. In this paper, we demonstrate the feasibility of training and running such an assistant, which we call Panza, on commodity hardware, for the specific use case of email generation. Panza’s personalization features are based on a combination of parameter-efficient fine-tuning using a variant of the Reverse Instructions technique [1] and Retrieval-Augmented Generation (RAG) [2]. We demonstrate that this combination allows us to fine-tune an LLM to reflect a user’s writing style using limited data, while executing on extremely limited resources, e.g. on a free Google Colab instance. Our key methodological contribution is the first detailed study of evaluation metrics for this task, and of how different choices of system components–the use of RAG and of different fine-tuning approaches–impact the system’s performance. Additionally, we demonstrate that very little data - under 100 email samples - are sufficient to create models that convincingly imitate humans, showcasing a previously unknown attack vector in language models. We are releasing the full Panza code as well as three new email datasets licensed for research use.}, author = {Nicolicioiu, Armand and Iofinova, Eugenia B and Jovanovic, Andrej and Kurtic, Eldar and Nikdan, Mahdi and Panferov, Andrei and Markov, Ilia and Shavit, Nir and Alistarh, Dan-Adrian}, booktitle = {Third Conference on Parsimony and Learning (Proceedings Track)}, keywords = {LLMs, PEFT, LoRA, personalization, efficient ML}, location = {Tübíngen, Germany}, publisher = {OpenReview}, title = {{Panza: Investigating the feasibility of fully-local personalized text generation}}, year = {2026}, } @unpublished{21859, abstract = {As artificial neural networks, and specifically large language models, have improved rapidly in capabilities and quality, they have increasingly been deployed in real-world applications, from customer service to Google search, despite the fact that they frequently make factually incorrect or undesirable statements. This trend has inspired practical and academic interest in model editing, that is, in adjusting the weights of the model to modify its likely outputs for queries relating to a specific fact or set of facts. This may be done either to amend a fact or set of facts, for instance, to fix a frequent error in the training data, or to suppress a fact or set of facts entirely, for instance, in case of dangerous knowledge. Multiple methods have been proposed to do such edits. However, at the same time, it has been shown that such model editing can be brittle and incomplete. Moreover the effectiveness of any model editing method necessarily depends on the data on which the model is trained, and, therefore, a good understanding of the interaction of the training data distribution and the way it is stored in the network is necessary and helpful to reliably perform model editing. However, working with large language models trained on real-world data does not allow us to understand this relationship or fully measure the effects of model editing. We therefore propose Behemoth, a fully synthetic data generation framework. To demonstrate the practical insights from the framework, we explore model editing in the context of simple tabular data, demonstrating surprising findings that, in some cases, echo real-world results, for instance, that in some cases restricting the update rank results in a more effective update.}, author = {Iofinova, Eugenia B and Alistarh, Dan-Adrian}, booktitle = {arXiv}, title = {{Behemoth: Benchmarking unlearning in LLMs using fully synthetic data}}, doi = {10.48550/arXiv.2601.23153}, year = {2026}, }