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.

Understanding the neural mechanisms of reward-guided learning is a long-standing goal of computational neuroscience. Recent methodological innovations enable us to collect ever larger neural and behavioral datasets. This presents opportunities to achieve greater understanding of learning in the brain at scale, as well as methodological challenges. In the first part of the talk, I will discuss our recent insights into the mechanisms by which zebra finch songbirds learn to sing. Dopamine has been long thought to guide reward-based trial-and-error learning by encoding reward prediction errors. However, it is unknown whether the learning of natural behaviours, such as developmental vocal learning, occurs through dopamine-based reinforcement. Longitudinal recordings of dopamine and bird songs reveal that dopamine activity is indeed consistent with encoding a reward prediction error during naturalistic learning. In the second part of the talk, I will talk about recent work we are doing at DeepMind to develop tools for automatically discovering interpretable models of behavior directly from animal choice data. Our method, dubbed CogFunSearch, uses LLMs within an evolutionary search process in order to "discover" novel models in the form of Python programs that excel at accurately predicting animal behavior during reward-guided learning. The discovered programs reveal novel patterns of learning and choice behavior that update our understanding of how the brain solves reinforcement learning problems.

Sleep is an active state critical for processing emotional memories encoded during waking in both humans and animals. There is a remarkable overlap between the brain structures and circuits active during sleep, particularly rapid eye-movement (REM) sleep, and the those encoding emotions. Accordingly, disruptions in sleep quality or quantity, including REM sleep, are often associated with, and precede the onset of, nearly all affective psychiatric and mood disorders. In this context, a major biomedical challenge is to better understand the underlying mechanisms of the relationship between (REM) sleep and emotion encoding to improve treatments for mental health. This lecture will summarize our investigation of the cellular and circuit mechanisms underlying sleep architecture, sleep oscillations, and local brain dynamics across sleep-wake states using electrophysiological recordings combined with single-cell calcium imaging or optogenetics. The presentation will detail the discovery of a 'somato-dendritic decoupling'in prefrontal cortex pyramidal neurons underlying REM sleep-dependent stabilization of optimal emotional memory traces. This decoupling reflects a tonic inhibition at the somas of pyramidal cells, occurring simultaneously with a selective disinhibition of their dendritic arbors selectively during REM sleep. Recent findings on REM sleep-dependent subcortical inputs and neuromodulation of this decoupling will be discussed in the context of synaptic plasticity and the optimization of emotional responses in the maintenance of mental health.

When we attend a demanding task, our performance is poor at low arousal (when drowsy) or high arousal (when anxious), but we achieve optimal performance at intermediate arousal. This celebrated Yerkes-Dodson inverted-U law relating performance and arousal is colloquially referred to as being "in the zone." In this talk, I will elucidate the behavioral and neural mechanisms linking arousal and performance under the Yerkes-Dodson law in a mouse model. During decision-making tasks, mice express an array of discrete strategies, whereby the optimal strategy occurs at intermediate arousal, measured by pupil, consistent with the inverted-U law. Population recordings from the auditory cortex (A1) further revealed that sound encoding is optimal at intermediate arousal. To explain the computational principle underlying this inverted-U law, we modeled the A1 circuit as a spiking network with excitatory/inhibitory clusters, based on the observed functional clusters in A1. Arousal induced a transition from a multi-attractor (low arousal) to a single attractor phase (high arousal), and performance is optimized at the transition point. The model also predicts stimulus- and arousal-induced modulations of neural variability, which we confirmed in the data. Our theory suggests that a single unifying dynamical principle, phase transitions in metastable dynamics, underlies both the inverted-U law of optimal performance and state-dependent modulations of neural variability.

The human medial temporal lobe contains neurons that respond selectively to the semantic contents of a presented stimulus. These "concept cells" may respond to very different pictures of a given person and even to their written or spoken name. Their response latency is far longer than necessary for object recognition, they follow subjective, conscious perception, and they are found in brain regions that are crucial for declarative memory formation. It has thus been hypothesized that they may represent the semantic "building blocks" of episodic memories. In this talk I will present data from single unit recordings in the hippocampus, entorhinal cortex, parahippocampal cortex, and amygdala during paradigms involving object recognition and conscious perception as well as encoding of episodic memories in order to characterize the role of concept cells in these cognitive functions.

Understanding the neural mechanisms of reward-guided learning is a long-standing goal of computational neuroscience. Recent methodological innovations enable us to collect ever larger neural and behavioral datasets. This presents opportunities to achieve greater understanding of learning in the brain at scale, as well as methodological challenges. In the first part of the talk, I will discuss our recent insights into the mechanisms by which zebra finch songbirds learn to sing. Dopamine has been long thought to guide reward-based trial-and-error learning by encoding reward prediction errors. However, it is unknown whether the learning of natural behaviours, such as developmental vocal learning, occurs through dopamine-based reinforcement. Longitudinal recordings of dopamine and bird songs reveal that dopamine activity is indeed consistent with encoding a reward prediction error during naturalistic learning. In the second part of the talk, I will talk about recent work we are doing at DeepMind to develop tools for automatically discovering interpretable models of behavior directly from animal choice data. Our method, dubbed CogFunSearch, uses LLMs within an evolutionary search process in order to "discover" novel models in the form of Python programs that excel at accurately predicting animal behavior during reward-guided learning. The discovered programs reveal novel patterns of learning and choice behavior that update our understanding of how the brain solves reinforcement learning problems.

Memories of emotionally-salient events are long-lasting, guiding behavior from minutes to years after learning. The prelimbic cortex (PL) is required for fear memory retrieval across time and is densely interconnected with many subcortical and cortical areas involved in recent and remote memory recall, including the temporal association area (TeA). While the behavioral expression of a memory may remain constant over time, the neural activity mediating memory-guided behavior is dynamic. In PL, different neurons underlie recent and remote memory retrieval and remote memory-encoding neurons have preferential functional connectivity with cortical association areas, including TeA. TeA plays a preferential role in remote compared to recent memory retrieval, yet how TeA circuits drive remote memory retrieval remains poorly understood. Here we used a combination of activity-dependent neuronal tagging, viral circuit mapping and miniscope imaging to investigate the role of the PL-TeA circuit in fear memory retrieval across time in mice. We show that PL memory ensembles recruit PL-TeA neurons across time, and that PL-TeA neurons have enhanced encoding of salient cues and behaviors at remote timepoints. This recruitment depends upon ongoing synaptic activity in the learning-activated PL ensemble. Our results reveal a novel circuit encoding remote memory and provide insight into the principles of memory circuit reorganization across time.

Humans and other animals approach reward and avoid punishment and pay attention to cues predicting these events. Such motivated behavior thus appears to be guided by value, which directs behavior towards or away from positively or negatively valenced outcomes. Moreover, it is facilitated by (top-down) salience, which enhances attention to behaviorally relevant learned cues predicting the occurrence of valenced outcomes. Using human neuroimaging, we recently separated value (ventral striatum, posterior ventromedial prefrontal cortex) from salience (anterior ventromedial cortex, occipital cortex) in the domain of liquid reward and punishment. Moreover, we investigated potential drivers of learned salience: the probability and uncertainty with which valenced and non-valenced outcomes occur. We find that the brain dissociates valenced from non-valenced probability and uncertainty, which indicates that reinforcement matters for the brain, in addition to information provided by probability and uncertainty alone, regardless of valence. Finally, we assessed learning signals (unsigned prediction errors) that may underpin the acquisition of salience. Particularly the insula appears to be central for this function, encoding a subjective salience prediction error, similarly at the time of positively and negatively valenced outcomes. However, it appears to employ domain-specific time constants, leading to stronger salience signals in the aversive than the appetitive domain at the time of cues. These findings explain why previous research associated the insula with both valence-independent salience processing and with preferential encoding of the aversive domain. More generally, the distinction of value and salience appears to provide a useful framework for capturing the neural basis of motivated behavior.

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 learn and represent 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 the hippocampus 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.

Recent insights from auditory neuroscience provide a new perspective on how the brain encodes speech. Using these recent insights, I will provide an overview of key factors underpinning individual differences in children’s development of language and phonology, providing a context for exploring atypical reading development (dyslexia). Children with dyslexia are relatively insensitive to acoustic cues related to speech rhythm patterns. This lack of rhythmic sensitivity is related to the atypical neural encoding of rhythm patterns in speech by the brain. I will describe our recent data from infants as well as children, demonstrating developmental continuity in the key neural variables.

Trends in NeuroAI is a reading group hosted by the MedARC Neuroimaging & AI lab (https://medarc.ai/fmri). Title: Brain-optimized inference improves reconstructions of fMRI brain activity Abstract: The release of large datasets and developments in AI have led to dramatic improvements in decoding methods that reconstruct seen images from human brain activity. We evaluate the prospect of further improving recent decoding methods by optimizing for consistency between reconstructions and brain activity during inference. We sample seed reconstructions from a base decoding method, then iteratively refine these reconstructions using a brain-optimized encoding model that maps images to brain activity. At each iteration, we sample a small library of images from an image distribution (a diffusion model) conditioned on a seed reconstruction from the previous iteration. We select those that best approximate the measured brain activity when passed through our encoding model, and use these images for structural guidance during the generation of the small library in the next iteration. We reduce the stochasticity of the image distribution at each iteration, and stop when a criterion on the "width" of the image distribution is met. We show that when this process is applied to recent decoding methods, it outperforms the base decoding method as measured by human raters, a variety of image feature metrics, and alignment to brain activity. These results demonstrate that reconstruction quality can be significantly improved by explicitly aligning decoding distributions to brain activity distributions, even when the seed reconstruction is output from a state-of-the-art decoding algorithm. Interestingly, the rate of refinement varies systematically across visual cortex, with earlier visual areas generally converging more slowly and preferring narrower image distributions, relative to higher-level brain areas. Brain-optimized inference thus offers a succinct and novel method for improving reconstructions and exploring the diversity of representations across visual brain areas. Speaker: Reese Kneeland is a Ph.D. student at the University of Minnesota working in the Naselaris lab. Paper link: https://arxiv.org/abs/2312.07705

Subsecond timing underlies nearly all sensory and motor activities across species and is critical to survival. While subsecond temporal information has been found across cortical and subcortical regions, it is unclear if it is generated locally and intrinsically or if it is a read out of a centralized clock-like mechanism. Indeed, mechanisms of subsecond timing at the circuit level are largely obscure. Primary sensory areas are well-suited to address these question as they have early access to sensory information and provide minimal processing to it: if temporal information is found in these regions, it is likely to be generated intrinsically and locally. We test this hypothesis by training mice to perform an audio-visual temporal pattern sensory discrimination task as we use 2-photon calcium imaging, a technique capable of recording population level activity at single cell resolution, to record activity in primary visual cortex (V1). We have found significant changes in network dynamics through mice’s learning of the task from naive to middle to expert levels. Changes in network dynamics and behavioral performance are well accounted for by an intrinsic model of timing in which the trajectory of q network through high dimensional state space represents temporal sensory information. Conversely, while we found evidence of other temporal encoding models, such as oscillatory activity, we did not find that they accounted for increased performance but were in fact correlated with the intrinsic model itself. These results provide insight into how subsecond temporal information is encoded mechanistically at the circuit level.

The critical brain hypothesis states that the brain can benefit from operating close to a second-order phase transition. While it has been shown that several computational aspects of sensory information processing (e.g., sensitivity to input) are optimal in this regime, it is still unclear whether these computational benefits of criticality can be leveraged by neural systems performing behaviorally relevant computations. To address this question, we investigate signatures of criticality in networks optimized to perform efficient encoding. We consider a network of leaky integrate-and-fire neurons with synaptic transmission delays and input noise. Previously, it was shown that the performance of such networks varies non-monotonically with the noise amplitude. Interestingly, we find that in the vicinity of the optimal noise level for efficient coding, the network dynamics exhibits signatures of criticality, namely, the distribution of avalanche sizes follows a power law. When the noise amplitude is too low or too high for efficient coding, the network appears either super-critical or sub-critical, respectively. This result suggests that two influential, and previously disparate theories of neural processing optimization—efficient coding, and criticality—may be intimately related

Precisely estimating event timing is essential for survival, yet temporal distortions are ubiquitous in our daily sensory experience. Here, we tested whether the relative position, duration, and distance in time of two sequentially-organized events—standard S, with constant duration, and comparison C, with duration varying trial-by-trial—are causal factors in generating temporal distortions. We found that temporal distortions emerge when the first event is shorter than the second event. Importantly, a significant interaction suggests that a longer inter-stimulus interval (ISI) helps to counteract such serial distortion effect only when the constant S is in the first position, but not if the unpredictable C is in the first position. These results imply the existence of a perceptual bias in perceiving ordered event durations, mechanistically contributing to distortion in time perception. Our results clarify the mechanisms generating time distortions by identifying a hitherto unknown duration-dependent encoding inefficiency in human serial temporal perception, something akin to a strong prior that can be overridden for highly predictable sensory events but unfolds for unpredictable ones.

Brains can gracefully weed out irrelevant stimuli to guide behavior. This feat is believed to rely on a progressive selection of task-relevant stimuli across the cortical hierarchy, but the specific across-area interactions enabling stimulus selection are still unclear. Here, we propose that population gating, occurring within A1 but controlled by top-down inputs from mPFC, can support across-area stimulus selection. Examining single-unit activity recorded while rats performed an auditory context-dependent task, we found that A1 encoded relevant and irrelevant stimuli along a common dimension of its neural space. Yet, the relevant stimulus encoding was enhanced along an extra dimension. In turn, mPFC encoded only the stimulus relevant to the ongoing context. To identify candidate mechanisms for stimulus selection within A1, we reverse-engineered low-rank RNNs trained on a similar task. Our analyses predicted that two context-modulated neural populations gated their preferred stimulus in opposite contexts, which we confirmed in further analyses of A1. Finally, we show in a two-region RNN how population gating within A1 could be controlled by top-down inputs from PFC, enabling flexible across-area communication despite fixed inter-areal connectivity.

Several species of insects have been observed to perform accurate path integration, constantly updating a vector memory of their location relative to a starting position, which they can use to take a direct return path. Foraging insects such as bees and ants are also able to store and recall the vectors to return to food locations, and to take novel shortcuts between these locations. Other insects, such as dung beetles, are observed to integrate multimodal directional cues in a manner well described by vector addition. All these processes appear to be functions of the Central Complex, a highly conserved and strongly structured circuit in the insect brain. Modelling this circuit, at the single neuron level, suggests it has general capabilities for vector encoding, vector memory, vector addition and vector rotation that can support a wide range of directed and navigational behaviours.

The human middle temporal complex (hMT+) has a crucial biological relevance for the processing and detection of direction and speed of motion in visual stimuli. In both humans and monkeys, it has been extensively investigated in terms of its retinotopic properties and selectivity for direction of moving stimuli; however, only in recent years there has been an increasing interest in how neurons in MT encode the speed of motion. In this talk, I will explore the proposed mechanism of speed encoding questioning whether hMT+ neuronal populations encode the stimulus speed directly, or whether they separate motion into its spatial and temporal components. I will characterize how neuronal populations in hMT+ encode the speed of moving visual stimuli using electrocorticography ECoG and 7T fMRI. I will illustrate that the neuronal populations measured in hMT+ are not directly tuned to stimulus speed, but instead encode speed through separate and independent spatial and temporal frequency tuning. Finally, I will suggest that this mechanism may play a role in evaluating multisensory responses for visual, tactile and auditory stimuli in hMT+.

A complete neuroscience requires multi-level theories that address phenomena ranging from higher-level cognitive behaviors to activities within a cell. Unfortunately, we don't have cognitive models of behavior whose components can be decomposed into the neural dynamics that give rise to behavior, leaving an explanatory gap. Here, we decompose SUSTAIN, a clustering model of concept learning, into neuron-like units (SUSTAIN-d; decomposed). Instead of abstract constructs (clusters), SUSTAIN-d has a pool of neuron-like units. With millions of units, a key challenge is how to bridge from abstract constructs such as clusters to neurons, whilst retaining high-level behavior. How does the brain coordinate neural activity during learning? Inspired by algorithms that capture flocking behavior in birds, we introduce a neural flocking learning rule to coordinate units that collectively form higher-level mental constructs ("virtual clusters"), neural representations (concept, place and grid cell-like assemblies), and parallels recurrent hippocampal activity. The decomposed model shows how brain-scale neural populations coordinate to form assemblies encoding concept and spatial representations, and why many neurons are required for robust performance. Our account provides a multi-level explanation for how cognition and symbol-like representations are supported by coordinated neural assemblies formed through learning.

Humans and other animals can estimate rapidly the number of items in a scene, flashes or tones in a sequence and motor actions. Adaptation techniques provide clear evidence in humans for the existence of specialized numerosity mechanisms that make up the numbersense. This sense of number is truly general, encoding the numerosity of both spatial arrays and sequential sets, in vision and audition, and interacting strongly with action. The adaptation (cross-sensory and cross-format) acts on sensory mechanisms rather than decisional processes, pointing to a truly general sense.

Previous experiments have shown that a comparison of two written narratives highlights their shared relational structure, which in turn facilitates the retrieval of analogous narratives from the past (e.g., Gentner, Loewenstein, Thompson, & Forbus, 2009). However, analogical retrieval occurs across domains that appear more conceptually distant than merely different narratives, and the deepest analogies use matches in higher-order relational structure. The present study investigated whether comparison can facilitate analogical retrieval of higher-order relations across written narratives and abstract symbolic problems. Participants read stories which became retrieval targets after a delay, cued by either analogous stories or letter-strings. In Experiment 1 we replicated Gentner et al. who used narrative retrieval cues, and also found preliminary evidence for retrieval between narrative and symbolic domains. In Experiment 2 we found clear evidence that a comparison of analogous letter-string problems facilitated the retrieval of source stories with analogous higher-order relations. Experiment 3 replicated the retrieval results of Experiment 2 but with a longer delay between encoding and recall, and a greater number of distractor source stories. These experiments offer support for the schema induction account of analogical retrieval (Gentner et al., 2009) and show that the schemas abstracted from comparison of narratives can be transferred to non-semantic symbolic domains.

The rising and setting of the sun is accompanied by changes in both the irradiance and the spectral distribution of the sky. Since the discovery of intrinsically photosensitive retinal ganglion cells (ipRGCs) 20 years ago, considerable progress has been made in understanding melanopsin's contributions to encoding irradiance. Much less is known about the cone inputs to ipRGCs and how they could encode changes in the color of the sky. I will summarize our recent connectomic investigation into the cone-opponent inputs to primate ipRGCs and the implications of this work on our understanding of circadian photoentrainment and the evolution of color vision.

Humans, as a social species, have an increased ability to detect and perceive visual elements involved in social exchanges, such as faces and eyes. The gaze, in particular, conveys information crucial for social interactions and social cognition. Researchers have hypothesized that in order to engage in dynamic face-to-face communication in real time, our brains must quickly and automatically process the direction of another person's gaze. There is evidence that direct gaze improves face encoding and attention capture and that direct gaze is perceived and processed more quickly than averted gaze. These results are summarized as the "direct gaze effect". However, in the recent literature, there is evidence to suggest that the mode of visual information processing modulates the direct gaze effect. In this presentation, I argue that top-down processing, and specifically the relevance of eye features to the task, promotes the early preferential processing of direct versus indirect gaze. On the basis of several recent evidences, I propose that low task relevance of eye features will prevent differences in eye direction processing between gaze directions because its encoding will be superficial. Differential processing of direct and indirect gaze will only occur when the eyes are relevant to the task. To assess the implication of task relevance on the temporality of cognitive processing, we will measure event-related potentials (ERPs) in response to facial stimuli. In this project, instead of typical ERP markers such as P1, N170 or P300, we will measure lateralized ERPs (lERPS) such as lateralized N170 and N2pc, which are markers of early face encoding and attentional deployment respectively. I hypothesize that the relevance of the eye feature task is crucial in the direct gaze effect and propose to revisit previous studies, which had questioned the existence of the direct gaze effect. This claim will be illustrate with different past studies and recent preliminary data of my lab. Overall, I propose a systematic evaluation of the role of top-down processing in early direct gaze perception in order to understand the impact of context on gaze perception and, at a larger scope, on social cognition.

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.

Many engineered and biological systems must trade off performance and energy use, and the brain is no exception. While there are theories on how activity levels are controlled in biological networks through feedback control (homeostasis), it is not clear what the effects on population coding are, and therefore how performance and energy can be traded off. In this talk we will consider this tradeoff in auto-encoding networks, in which there is a clear definition of performance (the coding loss). We first show how SNNs follow a characteristic trade-off curve between activity levels and coding loss, but that standard networks need to be retrained to achieve different tradeoff points. We next formalize this tradeoff with a joint loss function incorporating coding loss (performance) and activity loss (energy use). From this loss we derive a class of spiking networks which coordinates its spiking to minimize both the activity and coding losses -- and as a result can dynamically adjust its coding precision and energy use. The network utilizes several known activity control mechanisms for this --- threshold adaptation and feedback inhibition --- and elucidates their potential function within neural circuits. Using geometric intuition, we demonstrate how these mechanisms regulate coding precision, and thereby performance. Lastly, we consider how these insights could be transferred to trained SNNs. Overall, this work addresses a key energy-coding trade-off which is often overlooked in network studies, expands on our understanding of homeostasis in biological SNNs, as well as provides a clear framework for considering performance and energy use in artificial SNNs.

During perception the visual system integrates current sensory evidence with previously acquired knowledge of the visual world. Presumably this computation relies on internal recurrent interactions. We record populations of neurons from the primary visual cortex of cats and macaque monkeys and find evidence for adaptive internal responses to structured stimulation that change on both slow and fast timescales. In the first experiment, we present abstract images, only briefly, a protocol known to produce strong and persistent recurrent responses in the primary visual cortex. We show that repetitive presentations of a large randomized set of images leads to enhanced stimulus encoding on a timescale of minutes to hours. The enhanced encoding preserves the representational details required for image reconstruction and can be detected in post-exposure spontaneous activity. In a second experiment, we show that the encoding of natural scenes across populations of V1 neurons is improved, over a timescale of hundreds of milliseconds, with the allocation of spatial attention. Given the hierarchical organization of the visual cortex, contextual information from the higher levels of the processing hierarchy, reflecting high-level image regularities, can inform the activity in V1 through feedback. We hypothesize that these fast attentional boosts in stimulus encoding rely on recurrent computations that capitalize on the presence of high-level visual features in natural scenes. We design control images dominated by low-level features and show that, in agreement with our hypothesis, the attentional benefits in stimulus encoding vanish. We conclude that, in the visual system, powerful recurrent processes optimize neuronal responses, already at the earliest stages of cortical processing.

The human middle temporal complex (hMT+) has a crucial biological relevance for the processing and detection of direction and speed of motion in visual stimuli. In both humans and monkeys, it has been extensively investigated in terms of its retinotopic properties and selectivity for direction of moving stimuli; however, only in recent years there has been an increasing interest in how neurons in MT encode the speed of motion. In this talk, I will explore the proposed mechanism of speed encoding questioning whether hMT+ neuronal populations encode the stimulus speed directly, or whether they separate motion into its spatial and temporal components. I will characterize how neuronal populations in hMT+ encode the speed of moving visual stimuli using electrocorticography ECoG and 7T fMRI. I will illustrate that the neuronal populations measured in hMT+ are not directly tuned to stimulus speed, but instead encode speed through separate and independent spatial and temporal frequency tuning. Finally, I will show that this mechanism plays a role in evaluating multisensory responses for visual, tactile and auditory motion stimuli in hMT+.

It has long been established that the hippocampus plays a crucial role for episodic memory. As opposed to the modular approach, now it is generally assumed that being a complex structure, the HC performs multiplex interconnected functions, whose hierarchical organization provides basis for the higher cognitive functions such as semantics-based encoding and retrieval. However, the «where, when and how» properties of distinct memory aspects within and outside the HC are still under debate. Here we used a visual associative memory task as a probe to test the hypothesis about the differential involvement of the rostral and caudal portions of the human hippocampus in memory encoding, recognition and associative recall. In epilepsy patients implanted with stereo-EEG, we show that at retrieval the rostral HC is selectively active for recognition memory, whereas the caudal HC is selectively active for the associative memory. Low frequency desynchronization and high frequency synchronization characterize the temporal dynamic in encoding and retrieval. Therefore, we describe here anatomical segregation in the hippocampal contributions to associative and recognition memory.

A central goal of vision science is to understand the principles underlying the perception and neural coding of the complex visual environment of our everyday experience. In the visual cortex, foundational work with artificial stimuli, and more recent work combining natural images and deep convolutional neural networks, have revealed much about the tuning of cortical neurons to specific image features. However, a major limitation of this existing work is its focus on single-neuron response strength to isolated images. First, during natural vision, the inputs to cortical neurons are not isolated but rather embedded in a rich spatial and temporal context. Second, the full structure of population activity—including the substantial trial-to-trial variability that is shared among neurons—determines encoded information and, ultimately, perception. In the first part of this talk, I will argue for a normative approach to study encoding of natural images in primary visual cortex (V1), which combines a detailed understanding of the sensory inputs with a theory of how those inputs should be represented. Specifically, we hypothesize that V1 response structure serves to approximate a probabilistic representation optimized to the statistics of natural visual inputs, and that contextual modulation is an integral aspect of achieving this goal. I will present a concrete computational framework that instantiates this hypothesis, and data recorded using multielectrode arrays in macaque V1 to test its predictions. In the second part, I will discuss how we are leveraging this framework to develop deep probabilistic algorithms for natural image and video segmentation.

Neural circuits exhibit complex activity patterns, both spontaneously and in response to external stimuli. Information encoding and learning in neural circuits depend on the ability of time-varying stimuli to control spontaneous network activity. In particular, variability arising from the sensitivity to initial conditions of recurrent cortical circuits can limit the information conveyed about the sensory input. Spiking and firing rate network models can exhibit such sensitivity to initial conditions that are reflected in their dynamic entropy rate and attractor dimensionality computed from their full Lyapunov spectrum. I will show how chaos in both spiking and rate networks depends on biophysical properties of neurons and the statistics of time-varying stimuli. In spiking networks, increasing the input rate or coupling strength aids in controlling the driven target circuit, which is reflected in both a reduced trial-to-trial variability and a decreased dynamic entropy rate. With sufficiently strong input, a transition towards complete network state control occurs. Surprisingly, this transition does not coincide with the transition from chaos to stability but occurs at even larger values of external input strength. Controllability of spiking activity is facilitated when neurons in the target circuit have a sharp spike onset, thus a high speed by which neurons launch into the action potential. I will also discuss chaos and controllability in firing-rate networks in the balanced state. For these, external control of recurrent dynamics strongly depends on correlations in the input. This phenomenon was studied with a non-stationary dynamic mean-field theory that determines how the activity statistics and the largest Lyapunov exponent depend on frequency and amplitude of the input, recurrent coupling strength, and network size. This shows that uncorrelated inputs facilitate learning in balanced networks. The results highlight the potential of Lyapunov spectrum analysis as a diagnostic for machine learning applications of recurrent networks. They are also relevant in light of recent advances in optogenetics that allow for time-dependent stimulation of a select population of neurons.

Human movement disorders and pharmacological studies have long suggested molecular dopamine modulates the pace of the internal clock. But how does the endogenous dopaminergic system influence the timing of our movements? We examined the relationship between dopaminergic signaling and the timing of reward-related, self-timed movements in mice. Animals were trained to initiate licking after a self-timed interval following a start cue; reward was delivered if the animal’s first lick fell within a rewarded window (3.3-7 s). The first-lick timing distributions exhibited the scalar property, and we leveraged the considerable variability in these distributions to determine how the activity of the dopaminergic system related to the animals’ timing. Surprisingly, dopaminergic signals ramped-up over seconds between the start-timing cue and the self-timed movement, with variable dynamics that predicted the movement/reward time, even on single trials. Steeply rising signals preceded early initiation, whereas slowly rising signals preceded later initiation. Higher baseline signals also predicted earlier self-timed movement. Optogenetic activation of dopamine neurons during self-timing did not trigger immediate movements, but rather caused systematic early-shifting of the timing distribution, whereas inhibition caused late-shifting, as if dopaminergic manipulation modulated the moment-to-moment probability of unleashing the planned movement. Consistent with this view, the dynamics of the endogenous dopaminergic signals quantitatively predicted the moment-by-moment probability of movement initiation. We conclude that ramping dopaminergic signals, potentially encoding dynamic reward expectation, probabilistically modulate the moment-by-moment decision of when to move. (Based on work from Hamilos et al., eLife, 2021).

Estimating intervals is essential for adaptive behavior and decision-making. Although several theoretical models have been proposed to explain how the brain keeps track of time, there is still no evidence toward a single one. It is often hard to compare different models due to their overlap in behavioral predictions. For this reason, several studies have looked for neural signatures of temporal processing using methods such as electrophysiological recordings (EEG). However, for this strategy to work, it is essential to have consistent EEG markers of temporal processing. In this talk, I'll present results from several studies investigating how temporal information is encoded in the EEG signal. Specifically, across different experiments, we have investigated whether different neural signatures of temporal processing (such as the CNV, the LPC, and early ERPs): 1. Depend on the task to be executed (whether or not it is a temporal task or different types of temporal tasks); 2. Are encoding the physical duration of an interval or how much longer/shorter an interval is relative to a reference. Lastly, I will discuss how these results are consistent with recent proposals that approximate temporal processing with decisional models.

Topographic maps are a prominent feature of brain organization, reflecting local and large-scale representation of the sensory surface. ​​Traditionally, such representations in early visual areas are conceived as retinotopic maps preserving ego-centric retinal spatial location while ensuring that other features of visual input are uniformly represented for every location in space. I will discuss our recent findings of a striking departure from this simple mapping in the secondary visual area (V2) of the tree shrew that is best described as a sinusoidal transformation of the visual field. This sinusoidal topography is ideal for achieving uniform coverage in an elongated area like V2 as predicted by mathematical models designed for wiring minimization, and provides a novel explanation for stripe-like patterns of intra-cortical connections and functional response properties in V2. Our findings suggest that cortical circuits flexibly implement solutions to sensory surface representation, with dramatic consequences for large-scale cortical organization. Furthermore our work challenges the framework of relatively independent encoding of location and features in the visual system, showing instead location-dependent feature sensitivity produced by specialized processing of different features in different spatial locations. In the second part of the talk, I will propose that location-dependent feature sensitivity is a fundamental organizing principle of the visual system that achieves efficient representation of positional regularities in visual input, and reflects the evolutionary selection of sensory and motor circuits to optimally represent behaviorally relevant information. The relevant papers can be found here: V2 retinotopy (Sedigh-Sarvestani et al. Neuron, 2021) Location-dependent feature sensitivity (Sedigh-Sarvestani et al. Under Review, 2022)

To better understand human scene understanding, we extracted features from images using CLIP, a neural network model of visual concept trained with supervision from natural language. We then constructed voxelwise encoding models to explain whole brain responses arising from viewing natural images from the Natural Scenes Dataset (NSD) - a large-scale fMRI dataset collected at 7T. Our results reveal that CLIP, as compared to convolution based image classification models such as ResNet or AlexNet, as well as language models such as BERT, gives rise to representations that enable better prediction performance - up to a 0.86 correlation with test data and an r-square of 0.75 - in higher-level visual cortex in humans. Moreover, CLIP representations explain distinctly unique variance in these higher-level visual areas as compared to models trained with only images or text. Control experiments show that the improvement in prediction observed with CLIP is not due to architectural differences (transformer vs. convolution) or to the encoding of image captions per se (vs. single object labels). Together our results indicate that CLIP and, more generally, multimodal models trained jointly on images and text, may serve as better candidate models of representation in human higher-level visual cortex. The bridge between language and vision provided by jointly trained models such as CLIP also opens up new and more semantically-rich ways of interpreting the visual brain.

In the deep reinforcement learning (RL) community, motor control problems are usually approached from a reward-based learning perspective. However, humans are often believed to learn motor control through directed error-based learning. Within this learning setting, the control system is assumed to have access to exact error signals and their gradients with respect to the control signal. This is unlike reward-based learning, in which errors are assumed to be unsigned, encoding relative successes and failures. Here, we try to understand the relation between these two approaches, reward- and error- based learning, and ballistic arm reaches. To do so, we test canonical (deep) RL algorithms on a well-known sensorimotor perturbation in neuroscience: mirror-reversal of visual feedback during arm reaching. This test leads us to propose a potentially novel RL algorithm, denoted as model-based deterministic policy gradient (MB-DPG). This RL algorithm draws inspiration from error-based learning to qualitatively reproduce human reaching performance under mirror-reversal. Next, we show MB-DPG outperforms the other canonical (deep) RL algorithms on a single- and a multi- target ballistic reaching task, based on a biomechanical model of the human arm. Finally, we propose MB-DPG may provide an efficient computational framework to help explain error-based learning in neuroscience.

The hippocampus is crucial for spatial navigation and episodic memory formation. Hippocampal place cells exhibit spatially selective activity within an environment and have been proposed to form the neural basis of a cognitive map of space that supports these mnemonic functions. However, the direct influence of place cell activity on spatial navigation behaviour has not yet been demonstrated. Using an ‘all-optical’ combination of simultaneous two-photon calcium imaging and two-photon holographically targeted optogenetics, we identified and selectively activated place cells that encoded behaviourally relevant locations in a virtual reality environment. Targeted stimulation of a small number of place cells was sufficient to bias the behaviour of animals during a spatial memory task, providing causal evidence that hippocampal place cells actively support spatial navigation and memory. Time permitting, I will also describe new experiments aimed at understanding the fundamental encoding mechanism that supports episodic memory, focussing on the role of hippocampal sequences across multiple timescales and behaviours.

When we remember a stimulus we rarely maintain a full fidelity representation of the observed item. Our working memory instead maintains a mixture of the observed feature values and categorical/gist information. I will discuss evidence from computational models supporting a mix of categorical and fine-detail information in working memory. Having established the need for two memory formats in working memory, I will discuss whether categorical and fine-detailed information for a stimulus are represented separately or as a single unified representation. Computational models of these two potential cognitive structures make differing predictions about the pattern of responses in visual working memory recall tests. The present study required participants to remember the orientation of stimuli for later reproduction. The pattern of responses are used to test the competing representational structures and to quantify the relative amount of fine-detailed and categorical information maintained. The effects of set size, encoding time, serial order, and response order on memory precision, categorical information, and guessing rates are also explored. (This is a 60 min talk).

Neuronal computations are matched to optimally encode the sensory information that is available and relevant for the animal. However, the physical distribution of sensory information is often shaped by the animal’s own behavior. One prominent example is the encoding of optic flow fields that are generated during self-motion of the animal and will therefore depend on the type of locomotion. How evolution has matched computational resources to the behavioral constraints of an animal is not known. Here we use in vivo two photon imaging to record from a population of >3.500 local-direction selective cells. Our data show that the local direction-selective T4/T5 neurons in Drosophila form a population code that is matched to represent optic flow fields generated during translational and rotational self-motion of the fly. This coding principle for optic flow is reminiscent to the population code of local direction-selective ganglion cells in the mouse retina, where four direction-selective ganglion cells encode four different axes of self-motion encountered during walking (Sabbah et al., 2017). However, in flies we find six different subtypes of T4 and T5 cells that, at the population level, represent six axes of self-motion of the fly. The four uniformly tuned T4/T5 subtypes described previously represent a local snapshot (Maisak et al. 2013). The encoding of six types of optic flow in the fly as compared to four types of optic flow in mice might be matched to the high degrees of freedom encountered during flight. Thus, a population code for optic flow appears to be a general coding principle of visual systems, resulting from convergent evolution, but matching the individual ethological constraints of the animal.

Developmental epileptic encephalopathies are early-onset epilepsies, often refractory to therapy, with developmental delay or regression. These disorders carry poor neurodevelopmental prognosis, with long-term refractory epilepsy and persistent cognitive, behavioral and motor deficits. Mutations in the CACNA1A gene, encoding the pore-forming α1 subunit of CaV2.1 voltage-gated calcium channels, result in a spectrum of neurological disorders, including severe, early-onset epileptic encephalopathies. Recent work from the Rossignol lab helped characterize the phenotypic spectrum of CACNA1A-related epilepsies in humans. Using conditional genetics and novel animal models, the Rossignol lab unveiled some of the underlying pathophysiological mechanisms, including critical deficits in cortical inhibition, resulting in seizures and a range of cognitive-behavioral deficits. Importantly, Dr. Rossignol’s team demonstrated that the targeted activation of specific GABAergic interneuron populations in selected cortical regions prevents motor seizures and reverts attention deficits and cognitive rigidity in mouse models of the disorder. These recent findings open novel avenues for the treatment of these severe CACNA1A-associated neurodevelopmental disorders.

Modular structures in myriad forms — genetic, structural, functional — are ubiquitous in the brain. While modularization may be shaped by genetic instruction or extensive learning, the mechanisms of module emergence are poorly understood. Here, we explore complementary mechanisms in the form of bottom-up dynamics that push systems spontaneously toward modularization. As a paradigmatic example of modularity in the brain, we focus on the grid cell system. Grid cells of the mammalian medial entorhinal cortex (mEC) exhibit periodic lattice-like tuning curves in their encoding of space as animals navigate the world. Nearby grid cells have identical lattice periods, but at larger separations along the long axis of mEC the period jumps in discrete steps so that the full set of periods cluster into 5-7 discrete modules. These modules endow the grid code with many striking properties such as an exponential capacity to represent space and unprecedented robustness to noise. However, the formation of discrete modules is puzzling given that biophysical properties of mEC stellate cells (including inhibitory inputs from PV interneurons, time constants of EPSPs, intrinsic resonance frequency and differences in gene expression) vary smoothly in continuous topographic gradients along the mEC. How does discreteness in grid modules arise from continuous gradients? We propose a novel mechanism involving two simple types of lateral interaction that leads a continuous network to robustly decompose into discrete functional modules. We show analytically that this mechanism is a generic multi-scale linear instability that converts smooth gradients into discrete modules via a topological “peak selection” process. Further, this model generates detailed predictions about the sequence of adjacent period ratios, and explains existing grid cell data better than existing models. Thus, we contribute a robust new principle for bottom-up module formation in biology, and show that it might be leveraged by grid cells in the brain.

Large-scale neural recordings contain high-dimensional structure that cannot be easily captured by existing data visualization methods. We therefore developed an embedding algorithm called Rastermap, which captures highly nonlinear relationships between neurons, and provides useful visualizations by assigning each neuron to a location in the embedding space. Compared to standard algorithms such as t-SNE and UMAP, Rastermap finds finer and higher dimensional patterns of neural variability, as measured by quantitative benchmarks. We applied Rastermap to a variety of datasets, including spontaneous neural activity, neural activity during a virtual reality task, widefield neural imaging data during a 2AFC task, artificial neural activity from an agent playing atari games, and neural responses to visual textures. We found within these datasets unique subpopulations of neurons encoding abstract properties of the environment.

The long-term goal of our research is to understand the mechanisms of SUDEP, defined as Sudden, Unexpected, witnessed or unwitnessed, nontraumatic and non-drowning Death in patients with EPilepsy, excluding cases of documented status epilepticus. The majority of SUDEP patients die during sleep. SUDEP is the most devastating consequence of epilepsy, yet little is understood about its causes and no biomarkers exist to identify at risk patients. While SUDEP accounts for 7.5-20% of all epilepsy deaths, SUDEP risk in the genetic epilepsies varies with affected genes. Patients with ion channel gene variants have the highest SUDEP risk. Indirect evidence variably links SUDEP to seizure-induced apnea, pulmonary edema, dysregulation of cerebral circulation, autonomic dysfunction, and cardiac arrhythmias. Arrhythmias may be primary or secondary to hormonal or metabolic changes, or autonomic discharges. When SUDEP is compared to Sudden Cardiac Death secondary to Long QT Syndrome, especially to LQT3 linked to variants in the voltage-gated sodium channel (VGSC) gene SCN5A, there are parallels in the circumstances of death. To gain insight into SUDEP mechanisms, our approach has focused on channelopathies with high SUDEP incidence. One such disorder is Dravet syndrome (DS), a devastating form of developmental and epileptic encephalopathy (DEE) characterized by multiple pharmacoresistant seizure types, intellectual disability, ataxia, and increased mortality. While all patients with epilepsy are at risk for SUDEP, DS patients may have the highest risk, up to 20%, with a mean age at SUDEP of 4.6 years. Over 80% of DS is caused by de novo heterozygous loss-of-function (LOF) variants in SCN1A, encoding the VGSC Nav1.1  subunit, resulting in haploinsufficiency. A smaller cohort of patients with DS or a more severe DEE have inherited, homozygous LOF variants in SCN1B, encoding the VGSC 1/1B non-pore-forming subunits. A related DEE, Early Infantile EE (EIEE) type 13, is linked to de novo heterozygous gain-of-function variants in SCN8A, encoding the VGSC Nav1.6. VGSCs underlie the rising phase and propagation of action potentials in neurons and cardiac myocytes. SCN1A, SCN8A, and SCN1B are expressed in both the heart and brain of humans and mice. Because of this, we proposed that cardiac arrhythmias contribute to the mechanism of SUDEP in DEE. We have taken a novel approach to the development of therapeutics for DS in collaboration with Stoke Therapeutics. We employed Targeted Augmentation of Nuclear Gene Output (TANGO) technology, which modulates naturally occurring, non-productive splicing events to increase target gene and protein expression and ameliorate disease phenotype in a mouse model. We identified antisense oligonucleotides (ASOs) that specifically increase the expression of productive Scn1a transcript in human and mouse cell lines, as well as in mouse brain. We showed that a single intracerebroventricular dose of a lead ASO at postnatal day 2 or 14 reduced the incidence of electrographic seizures and SUDEP in the F1:129S-Scn1a+/- x C57BL/6J mouse model of DS. Increased expression of productive Scn1a transcript and NaV1.1 protein were confirmed in brains of treated mice. Our results suggest that TANGO may provide a unique, gene-specific approach for the treatment of DS.

Human vision is full of puzzles. Observers can grasp the essence of a scene in an instant, yet when probed for details they are at a loss. People have trouble finding their keys, yet they may be quite visible once found. How does one explain this combination of marvelous successes with quirky failures? I will describe our attempts to develop a unifying theory that brings a satisfying order to multiple phenomena. One key is to understand peripheral vision. A visual system cannot process everything with full fidelity, and therefore must lose some information. Peripheral vision must condense a mass of information into a succinct representation that nonetheless carries the information needed for vision at a glance. We have proposed that the visual system deals with limited capacity in part by representing its input in terms of a rich set of local image statistics, where the local regions grow — and the representation becomes less precise — with distance from fixation. This scheme trades off computation of sophisticated image features at the expense of spatial localization of those features. What are the implications of such an encoding scheme? Critical to our understanding has been the use of methodologies for visualizing the equivalence classes of the model. These visualizations allow one to quickly see that many of the puzzles of human vision may arise from a single encoding mechanism. They have suggested new experiments and predicted unexpected phenomena. Furthermore, visualization of the equivalence classes has facilitated the generation of testable model predictions, allowing us to study the effects of this relatively low-level encoding on a wide range of higher-level tasks. Peripheral vision helps explain many of the puzzles of vision, but some remain. By examining the phenomena that cannot be explained by peripheral vision, we gain insight into the nature of additional capacity limits in vision. In particular, I will suggest that decision processes face general-purpose limits on the complexity of the tasks they can perform at a given time.

During our everyday lives, much of what we experience is familiar and predictable. We typically follow the same morning routine, take the same route to work, and encounter the same colleagues. However, every once in a while, we encounter a surprising event that violates our expectations. When we encounter such violations of our expectations, it is adaptive to update our internal model of the world in order to make better predictions in the future. The hippocampus is thought to support both the learning of the predictable structure of our environment, as well as the detection and encoding of violations. However, the hippocampus is a complex and heterogeneous structure, composed of different subfields that are thought to subserve different functions. As such, it is not yet known how the hippocampus accomplishes the learning and updating of structured knowledge. Using behavioral methods and high-resolution fMRI, I'll show that during learning of repeated and predicted events, hippocampal subfields differentially integrate and separate event representations, thus learning the structure of ongoing experience. I then move on to discuss how when events violate our predictions, there is a shift in communication between hippocampal subfields, potentially allowing for efficient encoding of the novel and surprising information. If time permits, I'll present an additional behavioral study showing that violations of predictions promote detailed memories. Together, these studies advance our understanding of how we adaptively learn and update our knowledge.

Natural soundscapes such as from a forest, a busy restaurant, or a busy intersection are generally composed of a cacophony of sounds that the brain needs to interpret either independently or collectively. In certain instances sounds - such as from moving cars, sirens, and people talking - are perceived in unison and are recognized collectively as single sound (e.g., city noise). In other instances, such as for the cocktail party problem, multiple sounds compete for attention so that the surrounding background noise (e.g., speech babble) interferes with the perception of a single sound source (e.g., a single talker). I will describe results from my lab on the perception and neural representation of auditory textures. Textures, such as a from a babbling brook, restaurant noise, or speech babble are stationary sounds consisting of multiple independent sound sources that can be quantitatively defined by summary statistics of an auditory model (McDermott & Simoncelli 2011). How and where in the auditory system are summary statistics represented and the neural codes that potentially contribute towards their perception, however, are largely unknown. Using high-density multi-channel recordings from the auditory midbrain of unanesthetized rabbits and complementary perceptual studies on human listeners, I will first describe neural and perceptual strategies for encoding and perceiving auditory textures. I will demonstrate how distinct statistics of sounds, including the sound spectrum and high-order statistics related to the temporal and spectral correlation structure of sounds, contribute to texture perception and are reflected in neural activity. Using decoding methods I will then demonstrate how various low and high-order neural response statistics can differentially contribute towards a variety of auditory tasks including texture recognition, discrimination, and categorization. Finally, I will show examples from our recent studies on how high-order sound statistics and accompanying neural activity underlie difficulties for recognizing speech in background noise.