The PhD student will be part of the ANR project 'REFINED' involving the Laboratory of Embedded Artificial Intelligence in CEA List in Paris, the Multispeech research team In LORIA, Nancy, and the Hearing Institute in Paris. The project aims at studying new Deep Learning based methods to improve hearing acuity of ANSD patients. A cohort of ANSD volunteers will be tested to identify spectro-temporal auditory and extra-auditory cues correlated with the speech perception. Additionally, the benefits of neural networks will be studied. However, current artificial intelligence methods are too complex to be applied to processors with low computing and memory capacities: compression and optimization methods are needed.

University of Wurzburg, the newly established Center for Artificial Intelligence and Data Science (CAIDAS) has two Faculty openings: (1) Applied Super-Resolution Professorship (Full W3 or Associate W2 with tenure track W3). Possible research areas are: super-resolution, inverse problems, restoration, spectral imaging, machine learning for microscopy/telescopy, machine learning for photogrammetry. (2) Digital Media Processing Professorship (Junior W1 with tenure track W2). Possible research areas are: inverse problems, super-resolution, restoration, computational photography, compression, mobile/edge AI, Augmented/Mixed Reality, machine learning for photogrammetry.

At the newly established Center for Artificial Intelligence and Data Science (CAIDAS) from University of Wurzburg, Germany, we have two Faculty openings. The salaries and the packages are competitive. (1) Applied Super-Resolution Professorship (Full W3 or Associate W2 with tenure track W3). Possible research areas are: super-resolution, inverse problems, restoration, spectral imaging, machine learning for microscopy/telescopy, machine learning for photogrammetry. (2) Digital Media Processing Professorship (Junior W1 with tenure track W2). Possible research areas are: inverse problems, super-resolution, restoration, computational photography, compression, mobile/edge AI, Augmented/Mixed Reality, machine learning for photogrammetry.

This webinar addressed computational perspectives on how animals and humans make decisions, spanning normative, descriptive, and mechanistic models. Sam Gershman (Harvard) presented a capacity-limited reinforcement learning framework in which policies are compressed under an information bottleneck constraint. This approach predicts pervasive perseveration, stimulus‐independent “default” actions, and trade-offs between complexity and reward. Such policy compression reconciles observed action stochasticity and response time patterns with an optimal balance between learning capacity and performance. Jonathan Pillow (Princeton) discussed flexible descriptive models for tracking time-varying policies in animals. He introduced dynamic Generalized Linear Models (Sidetrack) and hidden Markov models (GLM-HMMs) that capture day-to-day and trial-to-trial fluctuations in choice behavior, including abrupt switches between “engaged” and “disengaged” states. These models provide new insights into how animals’ strategies evolve under learning. Finally, Kenji Doya (OIST) highlighted the importance of unifying reinforcement learning with Bayesian inference, exploring how cortical-basal ganglia networks might implement model-based and model-free strategies. He also described Japan’s Brain/MINDS 2.0 and Digital Brain initiatives, aiming to integrate multimodal data and computational principles into cohesive “digital brains.”

In the real world, object arrangement follows a number of rules. Some of the rules pertain to the spatial relations between objects and scenes (i.e., syntactic rules) and others about the contextual relations (i.e., semantic rules). Research has shown that violation of semantic rules influences interval timing with the duration of scenes containing such violations to be overestimated as compared to scenes with no violations. However, no study has yet investigated whether both semantic and syntactic violations can affect timing in the same way. Furthermore, it is unclear whether the effect of scene violations on timing is due to attentional or other cognitive accounts. Using an oddball paradigm and real-world scenes with or without semantic and syntactic violations, we conducted two experiments on whether time dilation will be obtained in the presence of any type of scene violation and the role of attention in any such effect. Our results from Experiment 1 showed that time dilation indeed occurred in the presence of syntactic violations, while time compression was observed for semantic violations. In Experiment 2, we further investigated whether these estimations were driven by attentional accounts, by utilizing a contrast manipulation of the target objects. The results showed that an increased contrast led to duration overestimation for both semantic and syntactic oddballs. Together, our results indicate that scene violations differentially affect timing due to violation processing differences and, moreover, their effect on timing seems to be sensitive to attentional manipulations such as target contrast.

This webinar will focus on spike sorting analysis with SpikeInterface, an open-source framework for the analysis of extracellular electrophysiology data. After a brief introduction of the project (~30 mins) highlighting the basics of the SpikeInterface software and advanced features (e.g., data compression, quality metrics, drift correction, cloud visualization), we will have an extensive hands-on tutorial (~90 mins) showing how to use SpikeInterface in a real-world scenario. After attending the webinar, you will: (1) have a global overview of the different steps involved in a processing pipeline; (2) know how to write a complete analysis pipeline with SpikeInterface.

Myopia is predicted to affect 50% of all people worldwide by 2050, and is a risk factor for significant, potentially blinding ocular pathologies, such as retinal detachment and glaucoma. Thus, there is significant motivation to better understand the process of myopigenesis and to develop effective anti-myopigenic treatments. In nearly all cases of human myopia, scleral remodeling is an obligate step in the axial elongation that characterizes the condition. Here I will describe the development of a biomechanical assay based on transient unconfined compression of scleral samples. By treating the scleral as a poroelastic material, one can determine scleral biomechanical properties from extremely small samples, such as obtained from the mouse eye. These properties provide proxy measures of scleral remodeling, and have allowed us to identify all-trans retinoic acid (atRA) as a myopigenic stimulus in mice. I will also describe nascent collaborative work on modeling the transport of atRA in the eye.

Neurons in the rodent hippocampus appear to encode the position of the animal in physical space during movement. Individual ``place cells'' fire in restricted sub-regions of an environment, a feature often taken as evidence that the hippocampus encodes a map of space that subserves navigation. But these same neurons exhibit complex responses to many other variables that defy explanation by position alone, and the hippocampus is known to be more broadly critical for memory formation. Here we elaborate and test a theory of hippocampal coding which produces place cells as a general consequence of efficient memory coding. We constructed neural networks that actively exploit the correlations between memories in order to learn compressed representations of experience. Place cells readily emerged in the trained model, due to the correlations in sensory input between experiences at nearby locations. Notably, these properties were highly sensitive to the compressibility of the sensory environment, with place field size and population coding level in dynamic opposition to optimally encode the correlations between experiences. The effects of learning were also strongly biphasic: nearby locations are represented more similarly following training, while locations with intermediate similarity become increasingly decorrelated, both distance-dependent effects that scaled with the compressibility of the input features. Using virtual reality and 2-photon functional calcium imaging in head-fixed mice, we recorded the simultaneous activity of thousands of hippocampal neurons during virtual exploration to test these predictions. Varying the compressibility of sensory information in the environment produced systematic changes in place cell properties that reflected the changing input statistics, consistent with the theory. We similarly identified representational plasticity during learning, which produced a distance-dependent exchange between compression and pattern separation. These results motivate a more domain-general interpretation of hippocampal computation, one that is naturally compatible with earlier theories on the circuit's importance for episodic memory formation. Work done in collaboration with James Priestley, Lorenzo Posani, Marcus Benna, Attila Losonczy.

Changes in the geometry and topology of self-assembled membranes underlie diverse processes across cellular biology and engineering. Similar to lipid bilayers, monolayer colloidal membranes studied by the Sharma (IISc Bangalore) and Dogic (UCSB) Labs have in-plane fluid-like dynamics and out-of-plane bending elasticity, but their open edges and micron length scale provide a tractable system to study the equilibrium energetics and dynamic pathways of membrane assembly and reconfiguration. First, we discuss how doping colloidal membranes with short miscible rods transforms disk-shaped membranes into saddle-shaped minimal surfaces with complex edge structures. Theoretical modeling demonstrates that their formation is driven by increasing positive Gaussian modulus, which in turn is controlled by the fraction of short rods. Further coalescence of saddle-shaped surfaces leads to exotic topologically distinct structures, including shapes similar to catenoids, tri-noids, four-noids, and higher order structures. We then mathematically explore the mechanics of these catenoid-like structures subject to an external axial force and elucidate their intimate connection to two problems whose solutions date back to Euler: the shape of an area-minimizing soap film and the buckling of a slender rod under compression. A perturbation theory argument directly relates the tensions of membranes to the stability properties of minimal surfaces. We also investigate the effects of including a Gaussian curvature modulus, which, for small enough membranes, causes the axial force to diverge as the ring separation approaches its maximal value.

Efficient coding has served as a guiding principle in understanding the neural code. To date, however, it has been explored mainly in the context of peripheral sensory cells with simple tuning curves. By contrast, ‘deeper’ neurons such as grid cells come with more complex tuning properties which imply a different, yet highly efficient, strategy for representing information. I will show that a highly efficient code is not specific to a population of neurons with finely tuned response properties: it emerges robustly in a shallow network with random synapses. Here, the geometry of population responses implies that optimality obtains from a tradeoff between two qualitatively different types of error: ‘local’ errors (common to classical neural population codes) and ‘global’ (or ‘catastrophic’) errors. This tradeoff leads to efficient compression of information from a high-dimensional representation to a low-dimensional one. After describing the theoretical framework, I will use it to re-interpret recordings of motor cortex in behaving monkey. Our framework addresses the encoding of (sensory) information; if time allows, I will comment on ongoing work that focuses on decoding from the perspective of efficient coding.

Tali's work emphasized the tradeoff between compression and information preservation. In this talk I will explore this theme in the context of deep learning. Artificial neural networks have recently revolutionized the field of machine learning. However, we still do not have sufficient theoretical understanding of how such models can be successfully learned. Two specific questions in this context are: how can neural nets be learned despite the non-convexity of the learning problem, and how can they generalize well despite often having more parameters than training data. I will describe our recent work showing that gradient-descent optimization indeed leads to 'simpler' models, where simplicity is captured by lower weight norm and in some cases clustering of weight vectors. We demonstrate this for several teacher and student architectures, including learning linear teachers with ReLU networks, learning boolean functions and learning convolutional pattern detection architectures.

Our world is full of living creatures that must share information to survive and reproduce. As humans, we easily forget how hard it is to communicate within natural environments. So how do organisms solve this challenge, using only natural resources? Ideas from computer science, physics and mathematics, such as energetic cost, compression, and detectability, define universal criteria that almost all communication systems must meet. We use insect swarms as a model system for identifying how organisms harness the dynamics of communication signals, perform spatiotemporal integration of these signals, and propagate those signals to neighboring organisms. In this talk I will focus on two types of communication in insect swarms: visual communication, in which fireflies communicate over long distances using light signals, and chemical communication, in which bees serve as signal amplifiers to propagate pheromone-based information about the queen’s location.