Seminars

Tal Linzen, an associate professor of linguistics and data science at New York University and a research scientist at Google, will speak in MIT's Siegel Family Quest for Intelligence seminar series. His work combines behavioral experiments and computational methods to study language learning and comprehension, alongside large-language-model post-training, evaluation, and interpretability.

Yuko Munakata presents evidence challenging capacity-only accounts of self-control and explains how experience, effort, and expected payoff shape executive-function engagement and later outcomes.

Zohar Ringel of the Hebrew University of Jerusalem presents a renormalisation-group approach to deep learning based on neural-network field theories. The seminar examines scaling laws, the removal of non-learnable field modes, benign overfitting and open questions about feature learning. The confirmed Quantum Matter seminar takes place in the Bob Room at Perimeter Institute on 8 September 2026, from 15:30 to 17:00 Toronto time.

Nick Turk-Browne presents research using infant functional MRI to investigate how the hippocampus supports learning and memory early in life. The seminar connects neural development, memory formation and infantile amnesia through new methods for studying awake infants.

Traditional work in the study of human reward-based learning involves designing an experimental task---often inspired by Reinforcement Learning (RL) theory---and fits a small set of computational models---often inspired by RL algorithms---to that dataset. For example, researchers often model human behavior on bandit tasks using variants of Q-learning. While this approach has been highly productive, leading to landmark discoveries such as the dopamine reward prediction error hypothesis, it also has limitations. This talk focuses on the lack of generalizability of such models: Even if they closely fit behavior on the original task, models derived from the one-task-one-model paradigm usually predict behavior on other tasks quite poorly. I argue that this lack of generalizability is a fundamental problem for the cognitive sciences: we intuitively expect our models to be robust to superficial task differences, such as variations in the number of choice options, reward probabilities, or the exact kind of non-stationarity. I will propose potential solutions to this problem along two dimensions: the behavioral dataset and the computational model. Regarding computational models, I will introduce work in which we moved beyond the limitations of hand-crafted one-off models by employing flexible, data-driven methods. These methods allowed us to compare classes of models instead of individual model instances, allowing us to cover the space of possible models more exhaustively, and innovate cognitive mechanisms very efficiently. For the behavioral dataset, we move from using single learning tasks to a comprehensive task space that encompasses most existing paradigms in the literature, while closing the gaps between them in a near-continuous fashion. Our results suggest that more general models in conjunction with broader datasets can pave the road toward increasingly general models of human reward-based learning and decision making, and a persistent departure from many aspects of RL theory. Presented in the van Vreeswijk Theoretical Neuroscience Seminar series (formerly WWTNS) on 2026-05-20. Recording duration: 00:51:19.

Reward expectations – one’s prediction about the likelihood of future outcomes - play a central role in shaping the satisfaction derived from those outcomes. Most existing research treats expectations as static, assuming they remain fixed in time. However, real-life expectations are often dynamic, fluctuating as new information becomes available. For example, during a soccer game, your expectations of seeing your team winning will likely rise and fall as the game unfolds. In the main part of this talk, I will present a series of studies demonstrating that human expectations can be tracked at sub-second timescales. Using slot machines as a case study, we leverage the continuous deceleration of the reels to elicit moment-by-moment fluctuations in rewardexpectations. To capture these dynamics, we take complementary approaches: we use the high temporal resolution of electroencephalography (EEG) to track neural signatures of evolving expectations, and we develop a novel behavioral paradigm (“Slot or Not”) designed to measure changes in expectations via betting behavior. Across four studies, we show that expectations fluctuate continuously and can be tracked both behaviorally and neurally. Extending these findings, a subsequent intracranial study shows that the human orbitofrontal cortex (OFC) encodes the moment-by-moment changes of reward expectations. In the second part of this talk, I will return to the relationship between expectations andsatisfaction. If expectations shape satisfaction, and if they are best conceptualized as dynamic trajectories rather than static quantities, a key question arises: does the trajectory leading up to an outcome influence how that outcome is evaluated? I will outline a new research direction aimed at formalizing this relationship using computational modeling. This is ongoing work, and I welcome feedback on how best to formalize these ideas. Finally, I will discuss potential extensions of this framework to psychopathology, asking whether alterations in dynamic expectations may characterize conditions such as Major Depressive Disorder and Gambling disorder. Together, this work introduces a new framework for studying expectations as dynamic processes, offering a richer understanding Presented in the van Vreeswijk Theoretical Neuroscience Seminar series (formerly WWTNS) on 2026-04-29. Recording duration: 00:42:25.

Useful internal representations should explain the patterns of regularities and dependencies among observations. Probabilistic graphical models promise a principled way to uncover latent factors as such, but they are hard to scale to handle high-dimensional sensory observations and complicated dependencies structures. Neural-networks, on the other hand, excel at approximating complicated high-dimensional functions, but their internal representations do not easily lend themselves to a probabilistic interpretation. Despite some successes, a general unified approach is still missing for integrating the two approaches. I will describe a novel approach towards merging adaptive neural-network components into a probabilistic framework, based on three core ideas. The first is to train a set of networks to collectively perform inference, leveraging the ability of pattern-recognition methods to amortise complicated transformations. The second is to constrain the way in which the outputs of these networks are interpreted, transformed, and combined together. These constraints, together with the learning objective itself, are derived directly from probabilistic considerations encoded in a graphical model. Finally, the third core idea is that of recognition-parametrisation, allowing the inference ("recognition") procedure to directly define the model itself, without requiring an explicit "generative" decoder. Presented in the van Vreeswijk Theoretical Neuroscience Seminar series (formerly WWTNS) on 2026-03-04. Recording duration: 00:48:26.

Much is known about the neural representations of current environmental location and direction within the hippocampal formation, but use of such a “cognitive map” requires the online representation of desired locations and how to get there, and the neural basis for this function has been more elusive. I will discuss how “theta sweeps” of place and grid cell firing encode the current location (at early phases of each theta cycle) while, at later phases, sampling around the forward direction during exploration and indicating the direction to desired locations during goal-directed navigation. I will show how a relatively simple attractor model captures these results, but requires inputs signalling movement-direction and goal-direction.I will discuss why it is useful to consider the hippocampus as a generative model (in which head-direction, rather than movement-direction, is required, to translate egocentric sensory inputs to allocentric latent representations and back again) in explaining its roles in both spatial cognition and memory consolidation. “Replay sequences” are thought to support offline consolidation, and likely resemble theta sweeps more than behavioural experience. I will finish (given time) by considering how human memory consolidation can be seen as extraction of latent variables from replay via self-supervised learning, and how this perspective explains aspects of human memory such as gist-based distortions, imagination and planning. Presented in the van Vreeswijk Theoretical Neuroscience Seminar series (formerly WWTNS) on 2026-02-25. Recording duration: 00:46:52.

Although stress can be considered as an ongoing process that helps an organism to cope with present and future challenges, when it is too intense or uncontrollable, it can lead to adverse consequences for physical and mental health. Social stress specifically, is a highly prevalent traumatic experience, present in multiple contexts, such as war, bullying and interpersonal violence, and it has been linked with increased risk for major depression and anxiety disorders. Nevertheless, not all individuals exposed to strong stressful events develop psychopathology, with the mechanisms of resilience and vulnerability being still under investigation. During this talk, I will identify key gaps in our knowledge about stress vulnerability and I will present our recent data from our contextual fear learning protocol based on social defeat stress in mice.

1. Can we reconstruct images that a person saw, directly from their fMRI brain recordings? 2. Can we reconstruct the training data that a deep-network trained on, directly from the parameters of the network? The answer to both of these intriguing questions is “Yes!” In this talk I will present some of our work in both domains. I will then show how combining the power of Brains and Machines can lead to significant breakthroughs in both areas, and potentially bridge the gap between Minds and Machines. Finally, I will show how combining the power of Multiple Brains (with NO shared data) may lead to new breakthrough discoveries in Brain-Science, and allow mapping of information between different brains. Presented in the van Vreeswijk Theoretical Neuroscience Seminar series (formerly WWTNS) on 2026-02-18. Recording duration: 00:50:24.

Current machine learning systems consume vastly more energy than biological brains. Neuromorphic systems aim to overcome this difference by mimicking the brain’s information coding via discrete voltage spikes. However, it remains unclear how both artificial and natural networks of spiking neurons can learn energy-efficient information processing strategies. Here we propose Predictive Coding Light (PCL), a recurrent hierarchical spiking neural network for unsupervised representation learning. In contrast to previous predictive coding approaches, PCL does not transmit prediction errors to higher processing stages. Instead, it suppresses the most predictable spikes and transmits a compressed representation of the input. Using only biologically plausible spike-timing based learning rules, PCL reproduces a wealth of findings on information processing in visual cortex and permits strong performance in downstream classification tasks. Overall, PCL offers a new approach to predictive coding and its implementation in natural and artificial spiking neural networks

Robert Gower introduces Polar Express for the polar decomposition and matrix sign function, motivated by Muon neural-network training. Using only matrix-matrix products makes the method suited to high-throughput GPUs. Each iteration adapts its polynomial update through minimax optimization, building on Chen and Chow and Nakatsukasa and Freund. Worst-case error minimization gives rapid initial and asymptotic convergence. The talk addresses finite-precision implementation in bfloat16 and reports improved validation loss when training GPT-2 on one billion FineWeb tokens across several learning rates.

Lior Horesh presents tensor-tensor algebra designed to retain key properties of matrix algebra while representing multidimensional correlations. An Eckart–Young-like tensor representation theorem underpins computationally feasible, provably optimal decompositions. Matrix-mimetic operations allow existing computational workflows to be adapted to tensors. Examples include tensorized neural-network structures and tensor graph convolutional networks for time-evolving graphs. The discussion concludes with tensor group symmetries and extensions to equivariant learning.

The hippocampus is thought to build a cognitive map that supports navigation, memory, and planning, but what defines such a map and how it is used remain debated. In this talk, I will present computational models in which hippocampal-like representations emerge in recurrent neural networks trained to predict sequences of sensory observations. While spatially tuned units reliably arise, they are not sufficient to form a cognitive map. Instead, map-like representations emerge when recurrent dynamics support multi-step prediction, yielding a population-level encoding of environmental geometry. Once learned, these representations can autonomously generate offline trajectories biased by recent experience, capturing key features of hippocampal replay. I will then show how these representations guide behavior in navigation tasks. In a hippocampal–striatal model facing visual ambiguity, access to hippocampal activity enables rapid learning and flexible adaptation. Place-like coding supports self-localization, while population-level hippocampal states can be used to derive intrinsic learning signals that estimate progress toward a remembered goal, improving performance beyond full sensory observability. Together, these results suggest that cognitive maps arise from predictive recurrent dynamics and support behavior through both localization and internally generated learning signals. Presented in the van Vreeswijk Theoretical Neuroscience Seminar series (formerly WWTNS) on 2026-01-21. Recording duration: 00:46:20.

David Baker presents recent advances in designing proteins for functions that natural evolution has not been required to solve. The approach begins with a desired structure or activity and uses deep learning to propose amino acid sequences expected to realise it. The resulting designs are translated into synthetic genes, produced as proteins and tested experimentally. The lecture connects computational design with laboratory characterisation, explaining how the combination can create new molecular functions relevant to medicine, technology and sustainability. Its focus is the construction and evaluation of proteins from scratch, extending the protein-design research recognised by the 2024 chemistry prize.

Neural activity is often analyzed with respect to external referents, such as the onset of a sensory stimulus or an overt motor action. Simultaneous recordings allow referencing neurons’ activity to each other and thus detecting signals that are internal to the organism. Further, multi-region simultaneous recordings allow observing how these internal signals are coordinated across the brain. Following this logic in rats performing a perceptual decision-making task, we recorded simultaneously from thousands of neurons across up to 20 brain regions at once. Here we report two internal signals which we found to profoundly shape decision-related neural dynamics and brain states. First, we decoded the continuously evolving decision state separately from each region, and found surprisingly large magnitude co-fluctuations in these measures. Dimensionality analysis showed these to be dominated by a single state variable, suggesting that only a single decision-making computation, not multiple parallel computations, are being carried out during the analyzed period. Second, we found that the precise time the subject commits to a decision – a covert event that we decoded from large-scale neural activity in primary motor cortex – was accompanied by a coordinated change, across the brain, from a decision formation to a post-commitment state. The two states differ substantially in their choice-predictive neural dynamics and in their inter-region correlations. Therefore, knowing the time of this state change on single trials is needed to correctly parse fundamentally different phases of decision-making. Overall, our data suggest that internally-referenced signals and state changes, not timelocked to external events but detectable through simultaneous recordings, are major features of neural activity during cognition. VVTNS 2026 Opening Lecture. Presented in the van Vreeswijk Theoretical Neuroscience Seminar series (formerly WWTNS) on 2026-01-07. Recording duration: 00:42:45.

Traditionally, touch is associated with exteroception and is rarely considered a relevant sensory cue for controlling movements in space, unlike vision. We developed a technique to isolate and measure tactile involvement in controlling sliding finger movements over a surface. Young adults traced a 2D shape with their index finger under direct or mirror-reversed visual feedback to create a conflict between visual and somatosensory inputs. In this context, increased reliance on somatosensory input compromises movement accuracy. Based on the hypothesis that tactile cues contribute to guiding hand movements when in contact with a surface, we predicted poorer performance when the participants traced with their bare finger compared to when their tactile sensation was dampened by a smooth, rigid finger splint. The results supported this prediction. EEG source analyses revealed smaller current in the source-localized somatosensory cortex during sensory conflict when the finger directly touched the surface. This finding supports the hypothesis that, in response to mirror-reversed visual feedback, the central nervous system selectively gated task-irrelevant somatosensory inputs, thereby mitigating, though not entirely resolving, the visuo-somatosensory conflict. Together, our results emphasize touch’s involvement in movement control over a surface, challenging the notion that vision predominantly governs goal-directed hand or finger movements.

Predictive processing offers a unifying view of neural computation, proposing that brains continuously anticipate sensory input and update internal models based on prediction errors. In this talk, I will present converging evidence for the computational mechanisms underlying this framework across human neuroscience and deep neural networks. I will begin with recent work showing that large-scale distributed prediction-error encoding in the human brain directly predicts how sensory representations reorganize through predictive learning. I will then turn to PredNet, a popular predictive coding inspired deep network that has been widely used to model real-world biological vision systems. Using dynamic stimuli generated with our Spatiotemporal Style Transfer algorithm, we demonstrate that PredNet relies primarily on low-level spatiotemporal structure and remains insensitive to high-level content, revealing limits in its generalization capacity. Finally, I will discuss new recurrent vision models that integrate top-down feedback connections with intrinsic neural variability, uncovering a dual mechanism for robust sensory coding in which neural variability decorrelates unit responses, while top-down feedback stabilizes network dynamics. Together, these results outline how prediction error signaling and top-down feedback pathways shape adaptive sensory processing in biological and artificial systems.

There is a fundamental tension between storing discrete traces of individual experiences, which allows recall of particular moments in our past without interference, and extracting regularities across these experiences, which supports generalization and prediction in similar situations in the future. One influential proposal for how the brain resolves this tension is that it separates the processes anatomically into Complementary Learning Systems, with the hippocampus rapidly encoding individual episodes and the neocortex slowly extracting regularities over days, months, and years. But this does not explain our ability to learn and generalize from new regularities in our environment quickly, often within minutes. We have put forward a neural network model of the hippocampus that suggests that the hippocampus itself may contain complementary learning systems, with one pathway specializing in the rapid learning of regularities and a separate pathway handling the region’s classic episodic memory functions. This proposal has broad implications for how we rapidly learn novel information of specific and generalized types, which we test across statistical learning, inference, and category learning paradigms. We also explore how this system interacts with slower-learning neocortical memory systems, with empirical and modeling investigations into how hippocampal replay shapes neocortical representations during sleep. Together, the work helps us understand how structured information in our environment is initially encoded and how it then transforms over time. Presented in the van Vreeswijk Theoretical Neuroscience Seminar series (formerly WWTNS) on 2025-12-03. Recording duration: 00:53:05.