Columbia University

Talk Title: TBD

Abstract: TBD

Harvard University

Talk Title: Offline cortical reactivations of recent sensory experiences

Abstract: Many theories of offline memory consolidation posit that the pattern of neurons activated during a salient sensory experience will be faithfully reactivated, thereby stabilizing the entire pattern. However, sensory-evoked patterns are not stable, but instead drift across repeated experiences. To investigate potential roles of reactivations in the stabilization and/or drift of sensory representations, we imaged calcium activity of thousands of excitatory neurons in mouse lateral visual cortex. Presentation of a stimulus resulted in transient, stimulus-specific reactivations during the following minute that were often coupled with hippocampal sharp-wave ripples. These cortical reactivations depended on local circuit activity, as they were abolished by cortical silencing during the preceding stimulus. Surprisingly, reactivations that occurred early in a session systemically differed from preceding activity patterns evoked by the stimulus. These reactivations were more similar to future patterns evoked by the stimulus, thereby predicting within-day and across-day representational drift. In particular, neurons that participated more or less in early reactivations than in stimulus response patterns gradually increased or decreased their stimulus responses later in the session. The rate and content of these reactivations was sufficient to accurately predict the dynamics and context of future changes in stimulus responses and, surprisingly, the decreasing similarity of cortical population responses to distinct stimuli. We hypothesize that activity patterns during sensory cortical reactivations may contribute to a gradual drift and separation in sensory response patterns, and may thereby enhance sensory discrimination.

Lab website

University of Oxford

Talk Title: Structuring behaviour in the PFC

Abstract: It looks like the PFC has an explicit representation of the future. There are now several groups showing this, including some data from our lab that I presented at the last lakes conference. This year I will present some theory work that tries to show what you can do with this kind of representation, and how it might make some very specific predictions about the connectivity in the PFC, where synaptic connections should instantiate a world model. It turns out if you do this right, you can get planning as inference for free in attractor dynamics. We hope it turns out to be true because it would be very pretty if so.

University of Oxford

Talk Title: Predictive coding: Effective learning with local plasticity

Abstract:Synaptic plasticity underlying leaning in biological neural networks only relies on information locally available to a synapse such as activity of presynaptic and postsynaptic neurons. This is a fundamental constraint on biological neural networks which shapes how they are organized. Hence it is important to understand how effective learning in large networks of neurons can be achieved within the constraint of local plasticity. This talk will review predictive coding, which is an influential model describing information processing in hierarchically organized cortical circuits. The talk will demonstrate that predictive coding network can learn equally or more effectively than artificial neural networks trained with backpropagation despite relying only on the local Hebbian plasticity. It will be also shown that when predictive coding networks are trained with inputs similar to those received by particular cortical regions (visual cortex and entorhinal cortex), then the neurons in the model develop representations seen in these regions (direction selectivity in visual cortex and grid cells in entorhinal cortex). This suggests that the diverse neural codes seen in different cortical regions can arise from the same learning algorithm, simply due to the difference in inputs they receive.

Lab website

Princeton University

Talk Title: TBD

Abstract: TBD

University of Washington

Talk Title: The Hippocampal Code as a Dynamic Internal Model of Constructed Experience

Abstract: A long history of research has demonstrated that the hippocampus plays a critical role in memory formation; however, we currently have an incomplete understanding of the mechanisms that support this function. While a large body of rigorous research in rodents supports the premise that hippocampal neurons encode allocentric location, accumulating evidence suggests that representations of place can be altered by task demands and by experience of rewards within the environment. To advance our understanding of the mechanisms of learning in the primate brain, we need a comprehensive framework that identifies the dynamics of hippocampal activity during both spatial and nonspatial forms of learning. I will discuss a series of experiments that have examined neural activity in the hippocampus in monkeys performing behavioral tasks including foraging and spatial memory tasks in a virtual environment. Data from these experiments demonstrate that hippocampal activity tracks the semantics of task structure and changes with task demands, providing a neural instantiation of a flexible mental model that extends to non-spatial domains and potentially serves as a scaffold for memory formation.

Lab Website

Imperial College London

Talk Title: Estimating the uncertainty of inputs with prediction-error circuits

Abstract:At any moment, our brains receive a stream of sensory stimuli arising from the world we interact with. Simultaneously, neural circuits are shaped by feedback signals carrying predictions about the same inputs we experience. Those feedforward and feedback inputs often do not perfectly match. Thus, our brains have the challenging task of integrating these conflicting streams of information according to their reliabilities. However, how neural circuits keep track of both the stimulus and prediction uncertainty is not well understood. Here, we propose a network model whose core is a hierarchical prediction-error circuit. We show that our network can estimate the variance of the sensory stimuli and the uncertainty of the prediction using the activity of negative and positive prediction-error neurons. In line with previous hypotheses, we demonstrate that neural circuits rely strongly on feedback predictions if the perceived stimuli are noisy and the underlying generative process, that is, the environment is stable. Moreover, we show that predictions modulate neural activity at the onset of a new stimulus, even if this sensory information is reliable. In our network, the uncertainty estimation, and, hence, how much we rely on predictions, can be influenced by perturbing the intricate interplay of different inhibitory interneurons. We, therefore, investigate the contribution of those inhibitory interneurons to the weighting of feedforward and feedback inputs. Finally, we show that our network can be linked to biased perception and unravel how stimulus and prediction uncertainty contribute to the contraction bias.

Allen Institute for Neural Dynamics

Talk Title: Structure and function of locus coeruleus norepinephrine neurons

Abstract: Norepinephrine (NE) is released through most of the nervous system from neurons in the locus coeruleus (LC). We found a relationship between the structure and function of these neurons in mice. Axonal projections of individual neurons were largely confined within brain regions and correlated with a gradient of gene expression in LC. LC-NE neurons showed two types of action potentials, wide (type I) and narrow (type II). Ascending projections were mostly type I and descending projections were mostly type II. In a behavioral task requiring ongoing learning, type I neurons were excited by reward prediction error, a key signal driving learning, and predicted future changes in behavior. Type II neurons were excited by lack of reward and predicted the opposite changes in future behavior. Together, these observations reveal a modular structure and function of a neurotransmitter system and show that it acts as a quantitative learning signal.

New York University

Talk Title: Multi-regional circuit mechanisms of value-based decisions

Abstract: The value of the environment determines animals’ motivational states and sets expectations for error-based learning. But how are values computed? We developed a novel temporal wagering task with latent structure, and used high-throughput behavioral training to obtain well-powered behavioral datasets from hundreds of rats that learned the structure of the task. We found that rats use distinct value computations for sequential decisions within single trials. Moreover, these sequential decisions are supported by different brain regions, suggesting that distinct neural circuits support specific types of value computations. I will discuss our ongoing efforts to delineate how distributed circuits in the orbitofrontal cortex and striatum coordinate complex value-based decisions.

Lab website

Allen Institute for Neural Dynamics

Talk Title: An optical approach for studying synaptic learning rules

Abstract: Learning goal-directed movements is associated with new activity patterns in the motor cortex, which become more reliable as performance improves. Hebbian theory proposes that reward information modulates synaptic plasticity by reinforcing activity patterns that lead to successful outcomes, but experimental support for this hypothesis is lacking. We combine optical mapping of causal connectivity with a single-neuron brain-computer interface (BCI) to test this idea. We find that plasticity is shaped by performance feedback. Specifically, we find that changes in connectivity across the population depend on the initial activity of the conditioned neuron—the neuron controlling the BCI. When the conditioned neuron starts out poorly aligned with the target, connections between co-active neurons tend to weaken. When it is already somewhat aligned, these connections strengthen instead. This suggests that learning reshapes connectivity based on reward prediction errors, reinforcing useful patterns while suppressing maladaptive ones. These results can be explained by recurrent networks trained with a novel 3-factor learning rule, in which reward prediction errors (RPEs) gate synaptic modifications. Lastly, I will discuss efforts to identify candidate neuromodulators that may convey the RPE signal to the cortex.

Lab website

Columbia University

Talk Title: TBD

Abstract: TBD

Georgia Institute of Technology

University of California, Berkeley

Talk Title: Flexible circuit computations of the Drosophila head direction network

Abstract: TBD

Website profile

Stanford University

Talk Title: Learning and adapting the structure of neural maps on behavioral timescales

Abstract: Over the last several decades, the tractable response properties of parahippocampal neurons have provided a new access key to understanding the cognitive process of self-localization: the ability to know where you are currently located in space. Defined by functionally discrete response properties, neurons across multiple brain regions are proposed to provide the basis for an internal neural map of space, which enables animals to perform path-integration based spatial navigation and supports the formation of spatial memories. My lab focuses on understanding the mechanisms that generate this neural map of space and how this map is used to support behavior. In this talk, I’ll discuss how our internal neural maps of space adapt and exhibit plasticity in novel environments.

Brandeis University

Talk Title: The role of dendrite-targeting interneurons in the formation of hippocampal representations

Abstract: A crucial function of the brain is to produce useful representations of the external world. These representations are used to form a cellular memory of past experiences, which can be recruited to guide present behaviors. However, the nature of the neuronal code and the mechanisms used to create, maintain, and recall neuronal representations remain unsolved. As a result, we still know very little about how representations embedded within neuronal circuits are actually used by the brain to guide goal-directed, adaptive behaviors. We previously demonstrated that a non-Hebbian type of synaptic plasticity, behavioral timescale synaptic plasticity (BTSP), has a fundamental role in forming experience-dependent representations in hippocampal area CA1. BTSP has several distinct characteristics, including that it is driven by dendritic plateau potentials (plateaus) instead of APs. Axons from layer 3 entorhinal cortex (EC) impinge onto the site of dendritic plateau initiation, the apical tuft of CA1 pyramidal neurons, and convey an instructive signal directing learning-related activity changes in the hippocampal CA1 network. In this talk, I will focus on recent data investigating the role of dendrite-targeting interneurons in regulating BTSP induction and the development of the spatial representation during learning. I will present preliminary experimental results suggesting that both feedforward and feedback dendrite-targeting inhibitory interneurons increase their activity with learning, potentially counter-balancing the increased activity of entorhinal inputs and limiting the number of CA1 pyramidal neurons that acquire place fields.

Lab website

Sainsbury Wellcome Centre

Talk Title:Neural circuits for predictive visual processing

Abstract: TBD

Lab Website

Friedrich Miescher Institute for Biomedical Research

Talk Title: A cortical circuit approach to psychosis

Abstract: The cortical dynamics underlying changes in conscious perception are largely unknown. By screening for common effects of diverse psychoactive drugs, we identified cell-type-specific activity patterns consistently altered by these substances. The primary effects were observed in spontaneous activity patterns spanning the dorsal cortex, rather than in sensory or movement-related activity. While the functional significance of these large-scale changes requires further investigation, distinct patterns emerged for antipsychotics, psychedelics, and anesthetics. Critically, all drug classes modulated functional communication within the layer 5 cortical network, supporting the hypothesis that this layer plays a key role in conscious processing.

Lab website

Ashok Litwin-Kumar

The École polytechnique fédérale de Lausanne

Talk Title: TBD

Abstract: TBD

California Institute of Technology

Talk Title: Rapid learning of complex tasks

Abstract: Animals can learn complex tasks involving multiple decisions after just a few successes. Learning often proceeds discontinuously: the animal’s performance may jump from one minute to the next, as though it had suddenly grasped the task’s solution. Little is known about the neural circuits or cellular mechanisms that support such rapid learning. Here I will report on studies of mice navigating a complex labyrinth, starting from their very first experience to expert performance a few hours later. I will touch on behavioral phenomena of sudden learning, a model of how it might work, neural recordings to test such a model, and some unexpected results from animals that lack the entire dorsal forebrain.

University of California San Diego

Talk Title: TBD

Abstract: TBD

Princeton University

Talk Title: Circuit Mechanisms for Dynamic Social Interactions

Abstract: Our research uncovers the neural mechanisms underlying how animals process dynamic sensory cues from a social partner, make decisions, and pattern the appropriate action for the current context. To uncover these mechanisms, we have developed new methods for quantification and computational modeling of behavior, as well as for brain-wide neural recording, and we combine these with the genetic and neural circuit tools of the Drosophila model system. We also recently generated the first whole-brain connectome for Drosophila, and I will discuss how we are leveraging this resource, to connect circuit architecture and activity at brain scale to behavior.

Lab website

Allen Institute for Neural Dynamics

Talk Title: Distributed context coding during sensory task switching

Abstract: TBD

Ecole Normale Superieure Paris

Talk Title: Structured Excitatory-Inhibitory Networks

Abstract: Networks of excitatory and inhibitory (EI) neurons form a canonical circuit in the brain. Classical theoretical analyses of dynamics in EI networks have revealed key principles such as EI balance or paradoxical responses to external inputs. These seminal results assume that synaptic strengths depend on the type of neurons they connect but are otherwise statistically independent. However, recent synaptic physiology datasets have uncovered connectivity patterns that deviate significantly from independent connection models. Simultaneously, studies of task-trained recurrent networks have emphasized the role of connectivity structure in implementing neural computations. Despite these findings, integrating detailed connectivity structures into mean-field theories of EI networks remains a substantial challenge. In this talk, I will outline a theoretical approach to understanding dynamics in structured EI networks by employing a low-rank approximation based on an analytical computation of the dominant eigenvalues of the full connectivity matrix. I will illustrate this approach by investigating the effects of pair-wise connectivity motifs on linear dynamics in EI networks. Specifically, I will present recent results demonstrating that an over-representation of chain motifs induces a strong positive eigenvalue in inhibition-dominated networks, generating a potential instability that challenges classical EI balance criteria. Furthermore, by examining the effects of external input, we found that chain motifs can, on their own, induce paradoxical responses, wherein an increased input to inhibitory neurons leads to a counterintuitive decrease in their activity through recurrent feedback mechanisms. Altogether, our theoretical approach opens new avenues for relating recorded connectivity structures with dynamics and computations in biological networks.

The Rockefeller University

Talk Title: TBD

Abstract: TBD

University of Washington

Talk Title: Assigning credit through the “other” connectome

Abstract: Learning in neural networks requires assigning the right values to thousands to trillions or more of individual connections, so that the network as a whole produces the desired behavior. Neuroscientists have gained insights into this “credit assignment” problem through decades of experimental, modeling, and theoretical studies. This has suggested key roles for synaptic eligibility traces and top-down feedback signals, among other factors. Here we study the potential contribution of another type of signaling that is being revealed in greater and greater fidelity by ongoing molecular and genomics studies. This is the set of modulatory pathways local to a given circuit, which form an intriguing second type of connectome overlayed on top of synaptic connectivity. We will share ongoing modeling and theoretical work that explores the possible roles of this local modulatory connectome in network learning. This is joint work with Helena Liu, Zihan Zhang, Uygar Sumbul, Stephen Smith, and Stefan Mihalas.

Lab website

Google DeepMind

Talk Title: Discovering symbolic cognitive models of animal behavior with CogFunSearch

Abstract: Symbolic models play a key role in cognitive science, expressing computationally precise hypotheses about how the brain implements a cognitive process. Identifying an appropriate model typically requires a great deal of effort and ingenuity on the part of a human scientist. Here, we adapt FunSearch [Romera-Paredes et al. (2024)], a recently developed tool that uses Large Language Models (LLMs) in an evolutionary algorithm, to automatically discover symbolic cognitive models that accurately capture human and animal behavior. We consider datasets from three species performing a classic reward-learning task that has been the focus of substantial modeling effort, and find that the discovered programs outperform state-of-the-art cognitive models for each. The discovered programs can readily be interpreted as hypotheses about human and animal cognition, instantiating interpretable symbolic learning and decision-making algorithms. Broadly, these results demonstrate the viability of using LLM-powered program synthesis to propose novel scientific hypotheses regarding mechanisms of human and animal cognition.

Lab website

University of Washington

Talk Title: The Midbrain Reticular Formation in visual decision-making

Abstract: TBD

Lab website

Max Planck Florida Institute for Neuroscience

Talk Title: TBD

Abstract: TBD

Princeton University

Talk Title: Dopamine neurons produce spatially localized decision substrates

Abstract: How does the activity of midbrain dopamine (DA) neurons influence behavior? A prominent, but incompletely tested, hypothesis is that the activity of ventral tegmental area (VTA) DA neurons generates downstream representations of state values (reward expectations) in ventral striatum (NAc), while substantia nigra pars compacta DA activity generates relative values (decision variables) in the dorsal striatum (DS). To test this model, we performed comprehensive striatal recordings in mice engaged in a trial-and-error probabilistic learning task where they continuously adapted their choices to obtain a reward of optogenetic stimulation of VTA DA neurons (paired with an auditory cue). Among several reinforcement learning models, we found that a variation of Q-learning best described choice trajectories. We then assessed neural representations of model-estimated state and relative action values, revealing for the first time that VTA DA stimulation is sufficient to generate downstream neural correlates of value. Surprisingly, both state and relative value representations were the strongest in the intermediate DS and the weakest in the NAc, contradicting the classic dorsal/ventral organization hypothesis. However, within the intermediate DS, state and action value representations were differentially organized, with the ventromedial region, which receives inputs from limbic cortical regions, preferentially encoding state value, and the dorsolateral subregion, which receives inputs from sensorimotor cortex, preferentially encoding relative values. A difference in timescales of value computation did not suffice to explain the relatively weak value correlates in NAc. Instead, we found that (1) VTA DA stimulation was sufficient to produce learned cue responses in NAc, (2) NAc DA but not DS DA can endow a cue with value to produce a conditioned reinforcer, and that (3) animals work for this cue, rather than VTA DA stimulation itself, in our main probabilistic learning task. Overall, this suggests that DA supports trial-and-error learning by endowing stimuli with value via the NAc, which in turn contributes to the generation of state and action values in DS.

Friedrich Miescher Institute for Biomedical Research

Talk Title: Representation learning through predictive synaptic plasticity

Abstract:Discriminating distinct objects and concepts from sensory stimuli is essential for survival. Our brains perform this processing in deep sensory networks shaped through plasticity. However, our understanding of the underlying plasticity mechanisms remains rudimentary. I will introduce Latent Predictive Learning (LPL), a plasticity model prescribing a local learning rule that combines Hebbian elements with predictive plasticity. I will show that deep neural networks equipped with LPL develop disentangled object representations without supervision. The same rule accurately captures neuronal selectivity changes observed in the primate inferotemporal cortex in response to altered visual experience. Finally, our model generalizes to spiking neural networks and naturally accounts for several experimentally observed properties of synaptic plasticity, including metaplasticity and spike-timing-dependent plasticity (STDP). LPL thus constitutes a plausible normative theory of representation learning in the brain while making concrete testable predictions.

Lab website