@inproceedings{17426, abstract = {The robustness of neural networks against input perturbations with bounded magnitude represents a serious concern in the deployment of deep learning models in safety-critical systems. Recently, the scientific community has focused on enhancing certifiable robustness guarantees by crafting 1-Lipschitz neural networks that leverage Lipschitz bounded dense and convolutional layers. Although different methods have been proposed in the literature to achieve this goal, understanding the performance of such methods is not straightforward, since different metrics can be relevant (e.g., training time, memory usage, accuracy, certifiable robustness) for different applications. For this reason, this work provides a thorough theoretical and empirical comparison between methods by evaluating them in terms of memory usage, speed, and certifiable robust accuracy. The paper also provides some guidelines and recommendations to support the user in selecting the methods that work best depending on the available resources. We provide code at https://github.com/berndprach/1LipschitzLayersCompared.}, author = {Prach, Bernd and Brau, Fabio and Buttazzo, Giorgio and Lampert, Christoph}, booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition}, location = {Seattle, WA, United States}, pages = {24574--24583}, publisher = {Computer Vision Foundation}, title = {{1-Lipschitz layers compared: Memory, speed, and certifiable robustness}}, doi = {10.1109/CVPR52733.2024.02320}, year = {2024}, } @unpublished{18874, abstract = {Despite extensive research since the community learned about adversarial examples 10 years ago, we still do not know how to train high-accuracy classifiers that are guaranteed to be robust to small perturbations of their inputs. Previous works often argued that this might be because no classifier exists that is robust and accurate at the same time. However, in computer vision this assumption does not match reality where humans are usually accurate and robust on most tasks of interest. We offer an alternative explanation and show that in certain settings robust generalization is only possible with unrealistically large amounts of data. More precisely we find a setting where a robust classifier exists, it is easy to learn an accurate classifier, yet it requires an exponential amount of data to learn a robust classifier. Based on this theoretical result, we explore how well robust classifiers generalize on datasets such as CIFAR-10. We come to the conclusion that on this datasets, the limitation of current robust models also lies in the generalization, and that they require a lot of data to do well on the test set. We also show that the problem is not in the expressiveness or generalization capabilities of current architectures, and that there are low magnitude features in the data which are useful for non-robust generalization but are not available for robust classifiers.}, author = {Prach, Bernd and Lampert, Christoph}, booktitle = {arXiv}, title = {{Intriguing properties of robust classification}}, doi = {10.48550/arXiv.2412.04245}, year = {2024}, } @inproceedings{17634, abstract = {System behaviors are traditionally evaluated through binary classifications of correctness, which do not suffice for properties involving quantitative aspects of systems and executions. Quantitative automata offer a more nuanced approach, mapping each execution to a real number by incorporating weighted transitions and value functions generalizing acceptance conditions. In this paper, we introduce QuAK, the first tool designed to automate the analysis of quantitative automata. QuAK currently supports a variety of quantitative automaton types, including Inf, Sup, LimInf, LimSup, LimInfAvg, and LimSupAvg automata, and implements decision procedures for problems such as emptiness, universality, inclusion, equivalence, as well as for checking whether an automaton is safe, live, or constant. Additionally, QuAK is able to compute extremal values when possible, construct safety-liveness decompositions, and monitor system behaviors. We demonstrate the effectiveness of QuAK through experiments focusing on the inclusion, constant-function check, and monitoring problems.}, author = {Chalupa, Marek and Henzinger, Thomas A and Mazzocchi, Nicolas Adrien and Sarac, Naci E}, booktitle = {12th International Symposium on Leveraging Applications of Formal Methods, Verification and Validation}, isbn = {9783031753862}, issn = {1611-3349}, location = {Crete, Greece}, pages = {3--20}, publisher = {Springer Nature}, title = {{QuAK: Quantitative Automata Kit}}, doi = {10.1007/978-3-031-75387-9_1}, volume = {15222}, year = {2024}, } @phdthesis{18681, author = {Tavakoli, Mojtaba}, isbn = {978-3-99078-048-0}, issn = {2663-337X}, pages = {230}, publisher = {Institute of Science and Technology Austria}, title = {{Developing molecular and structural tools for studying brain architecture with super resolution expansion microscopy. LICONN: Molecularly-informed connectomics reconstruction with light microscopy}}, doi = {10.15479/at:ista:18681}, year = {2024}, } @phdthesis{18132, abstract = {In this thesis, we are dealing with both arithmetic and geometric problems coming from the study of rational points with a particular focus on function fields over finite fields: (1) Using the circle method we produce upper bounds for the number of rational points of bounded height on diagonal cubic surfaces and fourfolds over Fq(t). This is based on joint work with Leonhard Hochfilzer. (2) We study rational points on smooth complete intersections X defined by cubic and quadratic hypersurfaces over Fq(t). We refine the Farey dissection of the “unit square” developed by Vishe [202] and use the circle method with a Kloosterman refinement to establish an asymptotic formula for the number of rational points of bounded height on X when dim(X) ≥ 23. Under the same hypotheses, we also verify weak approximation. (3) In joint work with Hochfilzer, we obtain upper bounds for the number of rational points of bounded height on del Pezzo surfaces of low degree over any global field. Our approach is to take hyperplane sections, which reduces the problem to uniform estimates for the number of rational points on curves. (4) We develop a version of the circle method capable of counting Fq-points on jet schemes of moduli spaces of rational curves on hypersurfaces. Combining this with a spreading out argument and a result of Mustaţă [150], this allows us to show that these moduli spaces only have canonical singularities under suitable assumptions on the degree and the dimension. In addition, we give an overview of guiding questions and conjectures in the field of rational points and explain the basic mechanism underlying the circle method. }, author = {Glas, Jakob}, issn = {2663-337X}, pages = {195}, publisher = {Institute of Science and Technology Austria}, title = {{Counting rational points over function fields}}, doi = {10.15479/at:ista:18132}, year = {2024}, } @article{14687, abstract = {The short history of research on Li-O2 batteries has seen a remarkable number of mechanistic U-turns over the years. From the initial use of carbonate electrolytes, that were then found to be entirely unsuitable, to the belief that (su)peroxide was solely responsible for degradation, before the more reactive singlet oxygen was found to form, to the hypothesis that capacity depends on a competing surface/solution mechanism before a practically exclusive solution mechanism was identified. Herein, we argue for an ever-fresh look at the reported data without bias towards supposedly established explanations. We explain how the latest findings on rate and capacity limits, as well as the origin of side reactions, are connected via the disproportionation (DISP) step in the (dis)charge mechanism. Therefrom, directions emerge for the design of electrolytes and mediators on how to suppress side reactions and to enable high rate and high reversible capacity.}, author = {Jethwa, Rajesh B and Mondal, Soumyadip and Pant, Bhargavi and Freunberger, Stefan Alexander}, issn = {1521-3773}, journal = {Angewandte Chemie International Edition}, keywords = {General Chemistry, Catalysis}, number = {28}, publisher = {Wiley}, title = {{To DISP or not? The far‐reaching reaction mechanisms underpinning Lithium‐air batteries}}, doi = {10.1002/anie.202316476}, volume = {63}, year = {2024}, } @unpublished{20071, abstract = {Farkas established that a system of linear inequalities has a solution if and only if we cannot obtain a contradiction by taking a linear combination of the inequalities. We state and formally prove several Farkas-like theorems over linearly ordered fields in Lean 4. Furthermore, we extend duality theory to the case when some coefficients are allowed to take "infinite values".}, author = {Dvorak, Martin and Kolmogorov, Vladimir}, booktitle = {arXiv}, keywords = {Farkas lemma, linear programming, extended reals, calculus of inductive constructions}, title = {{Duality theory in linear optimization and its extensions -- formally verified}}, doi = {10.48550/arXiv.2409.08119}, year = {2024}, } @article{18266, abstract = {Matrix games are the most basic model in game theory, and yet robustness with respect to small perturbations of the matrix entries is not fully understood. In this paper, we introduce value positivity and uniform value positivity, two properties that refine the notion of optimality in the context of polynomially perturbed matrix games. The first concept captures how the value depends on the perturbation parameter, and the second consists of the existence of a fixed strategy that guarantees the value of the unperturbed matrix game for every sufficiently small positive parameter. We provide polynomial-time algorithms to check whether a polynomially perturbed matrix game satisfies these properties. We further provide the functional form for a parameterized optimal strategy and the value function. Finally, we translate our results to linear programming and stochastic games, where value positivity is related to the existence of robust solutions.}, author = {Chatterjee, Krishnendu and Oliu-Barton, Miquel and Saona Urmeneta, Raimundo J}, issn = {1526-5471}, journal = {Mathematics of Operations Research}, number = {4}, pages = {2433--3282}, publisher = {Institute for Operations Research and the Management Sciences}, title = {{Value-positivity for matrix games}}, doi = {10.1287/moor.2022.0332}, volume = {50}, year = {2024}, } @article{14542, abstract = {It is a remarkable property of BCS theory that the ratio of the energy gap at zero temperature Ξ and the critical temperature Tc is (approximately) given by a universal constant, independent of the microscopic details of the fermionic interaction. This universality has rigorously been proven quite recently in three spatial dimensions and three different limiting regimes: weak coupling, low density and high density. The goal of this short note is to extend the universal behavior to lower dimensions d=1,2 and give an exemplary proof in the weak coupling limit.}, author = {Henheik, Sven Joscha and Lauritsen, Asbjørn Bækgaard and Roos, Barbara}, issn = {1793-6659}, journal = {Reviews in Mathematical Physics}, number = {9}, publisher = {World Scientific Publishing}, title = {{Universality in low-dimensional BCS theory}}, doi = {10.1142/s0129055x2360005x}, volume = {36}, year = {2024}, } @phdthesis{17485, abstract = {Large language models (LLMs) have made tremendous progress in the past few years, from being able to generate coherent text to matching or surpassing humans in a wide variety of creative, knowledge or reasoning tasks. Much of this can be attributed to massively increased scale, both in the size of the model as well as the amount of training data, from 100s of millions to 100s of billions, or even trillions. This trend is expected to continue, which, although exciting, also raises major practical concerns. Already today's 100+ billion parameter LLMs require top-of-the-line hardware just to run. Hence, it is clear that sustaining these developments will require significant efficiency advances. Historically, one of the most practical ways of improving model efficiency has been compression, especially in the form of sparsity or quantization. While this has been studied extensively in the past, existing accurate methods are all designed for models around 100 million parameters; scaling them up to ones literally 1000x larger is highly challenging. In this thesis, we introduce a new unified sparsification and quantization approach OBC, which through additional algorithmic enhancements leads to GPTQ and SparseGPT, the first techniques fast and accurate enough to compress 100+ billion parameter models to 4- or even 3-bit precision and 50% weight-sparsity, respectively. Additionally, we show how weight-only quantizion does not just bring space savings but also up to 4.5x faster generation speed, via custom GPU kernels. In fact, we show for the first time that it is possible to develop an FP16 times INT4 mixed-precision matrix multiplication kernel, called Marlin, which comes close to simultaneously maximizing both memory and compute utilization, making weight-only quantization highly practical even for multi-user serving. Further, we demonstrate that GPTQ can be scaled to widely overparametrized trillion-parameter models, where extreme sub-1-bit compression rates can be achieved without any inference slow-down, by co-designing a bespoke entropy coding scheme together with an efficient kernel. Finally, we also study compression from the perspective of someone with access to massive amounts of compute resources for training large models completely from scratch. Here the key questions evolve around the joint scaling behavior between compression, model size, and amount of training data used. Based on extensive experimental results for both vision and text models, we introduce the first scaling law which accurately captures the relationship between weight-sparsity, number of non-zero weights and data. This further allows us to characterize the optimal sparsity, which we find to increase the longer a fixed cost model is being trained. Overall, this thesis presents contributions to three different angles of large model efficiency: affordable but accurate algorithms, highly efficient systems implementations, and fundamental scaling laws for compressed training.}, author = {Frantar, Elias}, issn = {2663-337X}, pages = {129}, publisher = {Institute of Science and Technology Austria}, title = {{Compressing large neural networks: Algorithms, systems and scaling laws}}, doi = {10.15479/at:ista:17485}, year = {2024}, } @phdthesis{17465, abstract = {In the modern age of machine learning, artificial neural networks have become an integral part of many practical systems. One of the key ingredients of the success of the deep learning approach is recent computational advances which allowed the training of models with billions of parameters on large-scale data. Such over-parameterized and data-hungry regimes pose a challenge for the theoretical analysis of modern models since “classical” statistical wisdom is no longer applicable. In this view, it is paramount to extend or develop new machinery that will allow tackling the neural network analysis under new challenging asymptotic regimes, which is the focus of this thesis. Large neural network systems are usually optimized via “local” search algorithms, such as stochastic gradient descent (SGD). However, given the high-dimensional nature of the parameter space, it is a priori not clear why such a crude “local” approach works so remarkably well in practice. We take a step towards demystifying this phenomenon by showing that the landscape of the SGD training dynamics exhibits a few beneficial properties for the optimization. First, we show that along the SGD trajectory an over-parameterized network is dropout stable. The emergence of dropout stability allows to conclude that the minima found by SGD are connected via a continuous path of small loss. This in turn means that the high-dimensional landscape of the neural network optimization problem is provably not so unfavourable to gradient-based training, due to mode connectivity. Next, we show that SGD for an over-parameterized network tends to find solutions that are functionally more “simple”. This in turn means that the SGD minima are more robust, since a less complicated solution will less likely overfit the data. More formally, for a prototypical example of a wide two-layer ReLU network on a 1d regression task we show that the SGD algorithm is implicitly selective in its choice of an interpolating solution. Namely, at convergence the neural network implements a piece-wise linear function with the number of linear regions depending only on the amount of training data. This is in contrast to a “smooth”-like behaviour which one would expect given such a severe over-parameterization of the model. Diverging from the generic supervised setting of classification and regression problems, we analyze an auto-encoder model that is commonly used for representation learning and data compression. Despite the wide applicability of the auto-encoding paradigm, the theoretical understanding of their behaviour is limited even in the simplistic shallow case. The related work is restricted to extreme asymptotic regimes in which the auto-encoder is either severely over-parameterized or under-parameterized. In contrast, we provide a tight characterization for the 1-bit compression of Gaussian signals in the challenging proportional regime, i.e., the input dimension and the size of the compressed representation obey the same asymptotics. We also show that gradient-based methods are able to find a globally optimal solution and that the predictions made for Gaussian data extrapolate beyond - to the case of compression of natural images. Next, we relax the Gaussian assumption and study more structured input sources. We show that the shallow model is sometimes agnostic to the structure of the data vii which results in a Gaussian-like behaviour. We prove that making the decoding component slightly less shallow is already enough to escape the “curse” of Gaussian performance. }, author = {Shevchenko, Aleksandr}, issn = {2663-337X}, pages = {232}, publisher = {Institute of Science and Technology Austria}, title = {{High-dimensional limits in artificial neural networks}}, doi = {10.15479/at:ista:17465}, year = {2024}, } @phdthesis{17881, abstract = {This work can be broadly classified into the study of critical phenomena in a one dimensional array of Josephson junctions. While we study quantum criticality when the array is in thermal equilibrium at zero bias, the non-equilibrium study involves understanding the bistability of the array at a critical non-zero bias. This work furthers our knowledge in understanding quantum critical behaviour at finite temperatures in a one dimensional Josephson array, while also establishing relaxation behaviour dual to that observed in a single Josephson junction. Chapter 1 briefly introduces the model to understand superconductor-insulator phase transition in a one dimensional Josephson array and points out the state of the field from where we started our zero-bias experiments. In this context it discusses the phase-charge duality observed in a Josephson array and its dual hysteretic behaviour to that of a single junction, setting the ground for our non-equilibrium study of the array. Chapter 2 shows the experimental setup and the chip layout of the device we measured. In chapter 3 we show that, unlike the typical quantum-critical broadening scenario, in one dimensional Josephson arrays temperature dramatically shifts the critical region. This shift leads to a regime of superconductivity at high temperature, arising from the melted zero-temperature insulator. Our results quantitatively explain the low-temperature onset of superconductivity in nominally insulating regimes, and the transition to the strongly insulating phase. We further present, to our knowledge, the first understanding of the onset of anomalous-metallic resistance saturation [30]. This work demonstrates a non-trivial interplay between thermal effects and quantum criticality. A practical consequence is that, counterintuitively, the coherence of high-impedance quantum circuits is expected to be stabilized by thermal fluctuations. In chapter 4, we show relaxation oscillations in a current-biased one dimensional array of Josephson junctions. These oscillations are well described by a circuit model, dual to the ordinary Josephson relaxation oscillations [72]. Injection locking these oscillations results in current plateaux. The relaxation step is found to obey a characteristic self-consistent relation, suggesting that it is governed by overheating effects. Chapter 5 describes the various checks and analysis we performed to support our conclusions made in chapters 3 and 4. Finally, chapter 6 describes the nanofabrication steps and the finite element electromagnetic simulations we performed to fabricate our devices.}, author = {Mukhopadhyay, Soham}, isbn = {978-3-99078-043-5}, issn = {2663-337X}, pages = {82}, publisher = {Institute of Science and Technology Austria}, title = {{Thermal effects in one dimensional Josephson chains}}, doi = {10.15479/at:ista:17881}, year = {2024}, } @unpublished{18057, abstract = {We report relaxation oscillations in a one-dimensional array of Josephson junctions. The oscillations are circuit-dual to those ordinarily observed in single junctions. The dual circuit quantitatively accounts for temporal dynamics of the array, including the dependence on biasing conditions. Injection locking the oscillations results in well-developed current plateaux. A thermal model explains the relaxation step of the oscillations.}, author = {Mukhopadhyay, Soham and Lancheros Naranjo, Diego A and Senior, Jorden L and Higginbotham, Andrew P}, booktitle = {arXiv}, title = {{Dual relaxation oscillations in a Josephson junction array}}, doi = {10.48550/arXiv.2408.07829}, year = {2024}, } @phdthesis{17119, abstract = {Genomes are shaped by natural selection at the level of the organism, as genomic variants that have a beneficial effect on the viability or fecundity of their carriers are on average expected to be passed on to more offspring than less beneficial alleles. However, selection also favors genomic variants that drive their own transmission to the next generation above the mendelian expectation of 50 percent in heterozygotes, even if these self-promoting variants are less beneficial to the organism than other variants at the same locus. Such variants, called meiotic drivers, are found in diverse taxa, and often impose fitness costs on their host organisms. As meiotic drivers often require multiple genes and sequences for transmission ratio distortion, they are often found in regions of low recombination, such as inversions, which prevent their recombination with the non-driving homologous regions. Reduced recombination rates are expected to lead to the accumulation of deleterious mutations, which may affect hundreds of genes trapped in the inversions of meiotic drivers. Although the observed fitness costs of self-promoting haplotypes are thought to possibly reflect sequence degeneration, no study has systematically investigated the level of degeneration on a meiotic driver. Further, the low rates of recombination between driving and non-driving haplotypes have limited the power of traditional genetic studies in uncovering the gene content of meiotic drivers, and made the the identification of the genes causing transmission ratio distortion difficult. After an introduction to meiotic drivers in Chapter 1, this thesis presents three studies that make use of next generation sequencing data to characterize the sequence and expression evolution of genes on the t-haplotype, a large and ancient meiotic driver in house mice that is transmitted to up to 100% of the offspring in males heterozygous for it. Chapter 2 presents a comprehensive assessment of the t-haplotype’s sequence evolution, which shows signs of sequence degeneration counteracted by occasional recombination with the non-driving homolog over large parts of the meiotic driver, proposing an explanation for its long-term survival. Chapter 3 investigates the sequence and expression evolution of genes on the t-haplotype, and finds widespread expression and copy number changes and signs of less efficient purifying selection compared to the genes on the non-driving homolog. Further, this chapter finds candidates for involvment in drive: two positively selected genes on the t-haplotype, and the discovery of a t-specific gene duplicate, which was gained from another chromosome, and which acquired novel sequence and testis-specific expression on the t-haplotype. Finally, Chapter 4 provides unprecedented insights into the gene expression landscape in testes of t-carrier mice, using single nucleus sequencing. Cell-resolved RNA-sequencing allows the comparison of expression in spermatids carrying or not carrying the t-haplotype as well as the timing of t-haplotype-induced expression changes along spermatogenesis. This study shows the timing of previously found drive-associated genes, and uncovers novel candidate genes and biological processes that may underlie the complex biology of transmission ratio distortion of the t-haplotype. Chapter 5 synthesizes the findings of the three studies, and discusses them in the context of the current state of meiotic drive research.}, author = {Kelemen, Réka K}, isbn = {978-3-99078-039-8}, issn = {2663-337X}, keywords = {meiotic driver, neofunctionalization, single nucleus sequencing}, pages = {105}, publisher = {Institute of Science and Technology Austria}, title = {{Characterizing the sequence and expression evolution of the t-haplotype, a model meiotic driver}}, doi = {10.15479/at:ista:17119}, year = {2024}, } @phdthesis{18477, abstract = {ADAR1 is broadly expressed across various tissues and is vital in regulating pathways associated with innate immune responses. ADAR1 marks double-stranded RNA as "self" through its A-to-I editing activity, effectively repressing autoimmunity and maintaining immune tolerance. This editing process has been detected at millions of sites across the human genome. However, the mechanism underlying ADAR1's substrate selectivity properties remains largely unclear, with much of the current knowledge derived from comparisons to its more extensively studied homolog, ADAR2. By studying ADAR1 in complex with its RNA substrates and applying a combination of biochemical techniques and structural studies using CryoEM, we aim to gain a more comprehensive understanding of the substrate selectivity characteristics of ADAR1. In this thesis, the purification protocol for ADAR1 was successfully optimized, resulting in the first report in the literature to achieve high protein purity and activity. This advancement enabled the investigation of complex formation between ADAR1 and various RNA substrates, leading to the identification of optimal conditions for preparing the cryoEM sample. However, despite comprehensive optimization of the cryo-EM conditions, the resulting data lacked the desired quality, highlighting the need for similar rigorous optimization of the RNA substrates to facilitate structural studies of the ADAR1-RNA complex. The study was complemented by AlphaFold predictions, which provided some insights into this mechanism. Moreover, during this project I established a collaboration with a research group focused on studying ADAR homologs. Notably ADAR homologs were identified in bivalve species, and it was further demonstrated that ADAR and its A-to-I editing activity are upregulated in Pacific oysters during infections with Ostreid herpesvirus-1—a highly infectious virus that leads to significant losses in oyster populations globally. I successfully purified oyster ADAR and prepared in vitro edited RNA for nanopore sequencing—a direct sequencing technology capable of detecting modified nucleotides without the need for reverse transcription. The collaborators initiated optimization of this nanopore-based approach. However, current technological limitations still constrain the reliable detection of modified nucleotides. The project also examined the impact of RNA editing on RNA binding and filament formation by MDA5, a key cytosolic dsRNA sensor that triggers an interferon response. A primary target of ADAR1's editing activity is RNA derived from repetitive elements present in the genome, particularly Alu elements forming double-stranded RNA. When unedited, these RNA sequences are recognized by MDA5. However, the mechanisms by which MDA5 interacts with Alu RNAs, as well as the role of A-to-I editing in influencing this binding, are still not well understood. The interaction between MDA5 and Alu elements, was successfully established. This was achieved through the testing of different RNA variants and the evaluation of filament formation using binding techniques and electron microscopy imaging. This groundwork has set the conditions for further evaluation using CryoEM. Furthermore, the effects of A-to-I editing on the binding properties of MDA5 with Alu RNA were investigated. Given the recent research that has provided new insights into MDA5's interaction with dsRNA, it is essential to revise the experimental setup to integrate these findings before moving forward with the CryoEM sample analysis.}, author = {Kaczmarek, Beata M}, isbn = {978-3-99078-045-9}, issn = {2663-337X}, pages = {124}, publisher = {Institute of Science and Technology Austria}, title = {{Biochemical and structural insights into ADAR1 RNA editing}}, doi = {10.15479/at:ista:18477}, year = {2024}, } @phdthesis{15352, abstract = {Epilepsy affects about 50 to 65 million people globally. It summarizes a spectrum of neurological disorders that have in common a hyperactivity of the neuronal network resulting in seizures. A common assumption is that an imbalance between neuronal excitation and inhibition is a key mechanism in seizure generation and epileptogeneisis. In at least one-third of the patients, current therapies have proven unsuccessful in treating seizure progression. One potential reason could be that the therapies only focus on neurons. Recent studies suggest that neuronal hyperactivity causes a microglial response, which reinstates brain homeostasis. Additionally, interactions between microglia and neurons have been shown to inhibit neuronal firing and dampen seizure activity. However, the exact relationship between microglia and seizure progression in epilepsy is yet to be elucidated. A main bottleneck is that several studies investigate microglia dynamics in ex vivo slice models, which can severely affect the microglia dynamics due to their rapid response to environmental changes. On the other hand, in vivo studies focus mostly on behavior characterization of the epileptic seizure phenotype and their long-term consequences on microglia activity leaving out the direct consequences of acute seizure activity on microglia dynamics. Here, we perform a pilot study to combine electroencephalography (EEG) and in vivo live imaging to directly monitor and correlate the onset of seizure activity with microglia response. To induce seizures, we take advantage of the kainic acid (KA) model, which represents similar neuropathological and electroencephalographic features seen in human patients with temporal lobe epilepsy (TLE). After confirmation of induction of the seizure and microglia activity in the hippocampus as a focal point, we investigated whether these changes also reached the primary visual cortex (V1) as a secondary generalized seizure activity. Indeed, we found that microglia changed their morphology at high doses of KA in the V1. Next, we optimized each of the two methodological components: for the EEG recording, our initial attempts under the microscope suffered from extensive electrical noise, which overlaid the actual signal. Thus, we built a customized Faraday-cage and confirmed that the signal-to-noise ratio was sufficiently reduced to be able to record brain oscillatory activity. For the in vivo live imaging of microglia, we had to optimize the imaging parameters, so that we would be able to detect microglial processes in a sufficient resolution to track their process changes. Finally, we combined both methodologies with the KA model. We confirmed that KA induced seizure activity and found first indication that those correlate with microglia volume changes. Overall, we have developed a first methodological approach, which allows the analysis of the acute effects of seizure onset on microglia. Future studies will have to continue to optimize the drift during imaging recording and the post-image analysis. }, author = {Murmann, Julie Stefanie}, issn = {2791-4585}, pages = {54}, publisher = {Institute of Science and Technology Austria}, title = {{Investigating acute microglia response to seizure activity in vivo: Combining 2-Photon imaging and EEG recording}}, doi = {10.15479/at:ista:15352}, year = {2024}, } @phdthesis{14821, abstract = {The hippocampus is central to memory formation, storage and retrieval over many timescales. Neurons in this brain area are highly selective to spatial position as well as to many other variables of the environment. It is believed that the selectivity patterns of hippocampal neurons reflect the structure of tasks an animal performs. However, especially at timescales longer than a few minutes or hours it is not fully known how these representations evolve, nor how they map to behaviour in the process. In this thesis, I monitored the evolution of hippocampal representations in a novel spatial-associative memory task for rats. Reward locations were associated with global sensory cues (i.e. context); animals had to remember the associations and dig for food in those locations only. I used in vivo electrophysiology to record the activity of the hippocampus dorsal CA1 neurons during the learning period of a few days. I report here a novel and simple method to classify behaviour performance to account for individual variability in learning speed and spurious performance unrelated to true task rule learning. Using this classification I was then able to investigate neural responses on different stages of learning matched across animals. On the first day of learning, I observed a fast formation of single-cell selectivity to task variables which remained stable over days. I also observed that reward tuning was not a single process but dependent on task-related cognitive load. At the population level, a linear decoding approach revealed a hierarchy in the representation of task variables that changed with learning. In the high-dimensional space of population activity, the representation of contexts was specific to each position in the maze, and could thus be better decoded if the position was known. The decoding of position did not improve with knowledge of other variables. As learning progressed, the hippocampal code underwent a reorganisation of high-variance directions in population activity, identified by principal component analysis. I found that dominant dimensions started carrying increasing amounts of information about task context specifically at those positions where it mattered for task performance. When I contrasted this with variables less relevant to task performance (e.g. movement direction), I did not observe differences in decoding quality over positions nor a reduction of dimensionality with learning. Overall, the largest changes in CA1 neural response with task learning happened in a matter of a few trials; over days, changes undetectable in single-cell statistics were responsible for re-structuring the hierarchy of neural representations at the population level; these changes were task-specific and reflected different stages of learning. This indicates that complex task learning may involve different magnitudes of response modulation in CA1, which happen at specific time scales linked to behaviour.}, author = {Chiossi, Heloisa}, issn = {2663-337X}, pages = {89}, publisher = {Institute of Science and Technology Austria}, title = {{Adaptive hierarchical representations in the hippocampus}}, doi = {10.15479/at:ista:14821}, year = {2024}, } @phdthesis{15101, abstract = {The coupling between presynaptic Ca2+ channels and release sensors is a key factor that determines speed and efficacy of synapse transmission. At some excitatory synapses, channel–sensor coupling becomes tighter during development, and tightening is often associated with a switch in the reliance on different Ca2+ channel subtypes. However, the coupling topography at many synapses remains unknown, and it is unclear how it changes during development. To address this question, we analyzed the coupling configuration at the cerebellar basket cell (BC) to Purkinje cell (PC) synapse at different developmental stages, combining biophysical analysis, structural analysis, and modeling. Quantal analysis of BC–PC indicated that release probability decreased, while the number of functional sites increased during development. Although transmitter release persistently relied on P/Q-type Ca2+ channels in the time period postnatal day 7–23, effects of the Ca2+ chelator EGTA and BAPTA applied by intracellular pipette perfusion decreased during development, indicative of tightening of source-sensor coupling. Furthermore, presynaptic action potentials became shorter during development, suggesting reduced efficacy of Ca2+ channel activation. Structural analysis by freeze-fracture replica labeling (FRL) and transmission electron microscopy (EM) indicated that presynaptic P/Q-type Ca2+ channels formed nanoclusters throughout development, whereas docked vesicles were only clustered at later developmental stages. The number of functional release sites correlated better with the AZ number early in development, but match better with the Ca2+ channel cluster number at later stages. Modeling suggested a developmental transformation from a more random to a more clustered coupling nanotopography. Thus, presynaptic signaling developmentally approaches a point-to-point configuration, optimizing speed, reliability, and energy efficiency of synaptic transmission.}, author = {Chen, JingJing}, issn = {2663-337X}, pages = {84}, publisher = {Institute of Science and Technology Austria}, title = {{Developmental transformation of nanodomain coupling between Ca2+ channels and release sensors at a central GABAergic synapse}}, doi = {10.15479/at:ista:15101}, year = {2024}, } @phdthesis{18471, abstract = {Spatial omics technologies are enriching our understanding of complex biological samples, by allowing us to study their molecular composition while preserving the spatial relationships between molecules in their native context. As the field continues to advance, there are technical challenges that need to be addressed in order to take full advantage of the spatial capabilities of these methods. In this work, I present two technical developments that I established for multiplexed error robust FISH (MERFISH) throughout my PhD: (1) pushing the spatial resolution limits to the nanoscale, and (2) adding rich tissue context to the mouse brain transcriptome. To achieve nanoscale resolution with MERFISH in cultured cells, I combined it with stimulated emission depletion (STED) and expansion microscopy (ExM) to achieve a spatial resolution as low as ~20 nm, and explored the compatibility of MERFISH with singlemolecule localization microscopy (SMLM) techniques. To visualize targeted mRNAs in mouse brain tissue, I applied the comprehensive analysis of tissues across scales (CATS) toolbox, which provides an unbiased morphological readout by labeling the extracellular domain. I successfully established this method, which we call CATS-MERFISH-ExM, to work with thick mouse brain slices, being able to extract transcriptomics information with 3D tissue context. CATS-MERFISH-ExM enabled us to identify cell types and further visualize the subcellular distribution of transcripts in mouse brain tissue, shedding light on the neuropil-specific transcriptome. This method provides integrated information on cellular structure and transcriptomes in situ, and could potentially be applied with other modalities, opening new avenues for scientific discovery. }, author = {Agudelo Duenas, Nathalie}, isbn = {978-3-99078-044-2}, issn = {2663-337X}, pages = {97}, publisher = {Institute of Science and Technology Austria}, title = {{Visualizing the neuronal transcriptional landscape with tissue context}}, doi = {10.15479/at:ista:18471}, year = {2024}, } @phdthesis{18574, abstract = {Biological vision is unlike a camera; rather than transmitting light information faithfully, early visual circuits process the visual scene to convey only the relevant information in an efficient manner. Consequentially, the nature of this visual processing then depends on what is the relevant information in a scene and on the notion of efficiency. In this work, I study how visual processing is modulated by two different variations in the visual scene. First, I discovered that in the mouse (Mus musculus) retina, Retinal Ganglion Cells in the upper and lower visual field have differences in the center surround structure of their receptive fields. Comparison with models of efficient coding show that this adaptation likely evolved to cope with the brightness gradient from the sky to the ground that is pervasive in natural scenes. In the second project, I study how the downstream neurons in the Superior Colliculus dynamically change their temporal selectivity depending on the ambient luminance and behavioral state. As the scene gets darker or when the animal is is less aroused, the neuronal responses get laggier, while still maintaining their relative timing with respect to the population. Overall, this work emphasises the need to understand visual processing in the context of specific demands of the animal in its the environment. The adaptive changes in the visual system, from the retinal ganglion cells to the superior colliculus, highlight the intricate ways in which biological vision optimizes the processing of visual information. }, author = {Gupta, Divyansh}, isbn = {978-3-99078-050-3}, issn = {2663-337X}, pages = {86}, publisher = {Institute of Science and Technology Austria}, title = {{Visual adaptations to natural statistics}}, doi = {10.15479/at:ista:18574}, year = {2024}, }