Biology subjects

Mante, V.

Publications and source records attributed to Mante, V..

15 recordsLinked to original sources

bioRxiv2026

Weight initialization shapes task organization in recurrent neural networks

Flexibly recombining computational modules is essential for biological and artificial neural networks to rapidly adapt to changing environments. This requires modules to be shared across tasks rather than rigidly segregated, yet what determines this organization remains unknown. Previous work suggests that weight initialization shapes whether networks learn task-specific or generic representations, but it is unclear whether this extends to recurrent networks and, more importantly, to network connectivity. Here, we systematically vary the initial weight variance of recurrent neural networks and study them using a framework that allows us to identify the functionally relevant connectivity subspaces for each computational module. We find that networks with low initial weight variance converge to solutions in which different subtasks rely on largely overlapping weight subspaces, whereas high-variance networks implement subtasks in higher-dimensional, more segregated weight subspaces. Our results also provide mechanistic insights with implications for interpreting biological neural circuits and for designing efficient recurrent architectures.

neuroscience↗

bioRxiv2025

Category learning disentangles representation of trial events in hippocampus CA1

The hippocampus (HC) is known to encode task-relevant variables, including latent ones, capturing and parsing critical information from sequences into episodes. This is thought to be the basis to form a cognitive map of possible abstract states beyond mere perceptual details, akin a state machine, with predictive value, contributing to an internal world model. However, its specific role in categorization tasks in general, where learning a category may depend on independent, unordered (non-sequential) experienced examples, remains unclear. Here, we investigate CA1 population coding during a categorization paradigm in mice in combination with calcium imaging at different stages of category training with interleaved generalization tests. Our results reveal that hippocampal coding changes critically throughout the different stages of categorization training. Specifically, the disentangling of choice and outcome variables emerges as factorized, abstract, representations and fundamentally distinguishes simple discrimination from categorization training. Furthermore, this factorized geometry relates to improved behavioral performance. Our findings suggest that trial encoding on HC adapts in response to the category structure presented by representing critical events into the appropriate computational format to support generalization of category membership.

neuroscience↗

bioRxiv2025

War Trauma Impairs Prenatal Brain Development: Evidence from Genetically-Informed Brain Imaging

The global rise in armed conflicts has raised concerns about their long-term consequences. While the immediate mental and physical impacts of war are well documented, the effects on unborn children of expecting mothers remain poorly understood. Here, we leverage a large dataset (N=37,856) combining historical World War II bombing records with brain imaging and genetic data to examine whether in-utero exposure to air raids is associated with lasting differences in adult brain structure, cognition, and physical health. Using a pre-registered quasi-experimental design controlling for genetic and environmental factors, we find that exposure during the last trimester within 2 km of bombings is associated with lower gray matter volume, which partially mediates adverse effects on intelligence and obesity. These findings suggest a plausible pathway linking prenatal trauma to long-term differences in neurodevelopmental and functional outcomes, underscoring the vulnerability of the fetal brain to acute stress and highlighting the importance of protecting pregnant women during armed conflicts.

neuroscience↗

bioRxiv2025

Data-Driven Reduced Modeling of Recurrent Neural Networks

Artificial Recurrent Neural Networks (RNNs) are widely used in neuroscience to model the collective activity of neurons during behavioral tasks. The high dimensionality of their parameter and activity spaces, however, often make it challenging to infer and interpret the fundamental features of their dynamics. In this study, we employ recent nonlinear dynamical system techniques to uncover the core dynamics of several RNNs used in contemporary neuroscience. Specifically, using a data-driven approach, we identify Spectral Submanifolds (SSMs), i.e., low-dimensional attracting invariant manifolds tangent to the eigenspaces of fixed points. The internal dynamics of SSMs serve as nonlinear models that reduce the dimensionality of the full RNNs by orders of magnitude. Through low-dimensional, SSM-reduced models, we give mathematically precise definitions of line and ring attractors, which are intuitive concepts commonly used to explain decision-making and working memory. The new level of understanding of RNNs obtained from SSM reduction enables the interpretation of mathematically well-defined and robust structures in neuronal dynamics, leading to novel predictions about the neural computations underlying behavior.

neuroscience↗

bioRxiv2023

Inferring context-dependent computations through linear approximations of prefrontal cortex dynamics

The complex neural population activity of prefrontal cortex (PFC) is a hallmark of cognitive processes. How these rich dynamics emerge and support neural computations is largely unknown. Here, we infer mechanisms underlying the context-dependent selection and integration of sensory inputs by fitting dynamical models to PFC population responses of behaving monkeys. A class of models implementing linear dynamics driven by external inputs accurately captured the PFC responses within each context, achieving performance comparable to models without linear constraints. Two distinct mechanisms of input selection and integration were equally consistent with the data. One implemented context-dependent recurrent dynamics, as previously proposed, and relied on transient input amplification. The other relied on the subtle contextual modulation of the inputs, providing quantitative constraints on the attentional effects in sensory areas required to explain flexible PFC responses and behavior. Both mechanisms consistently revealed properties of inputs and recurrent dynamics missing in more simplified, incomplete descriptions of PFC responses. By revealing mechanisms consistent with rich cortical dynamics, our modeling approach provides a principled and general framework to link neural population activity and computation.

neuroscience↗

bioRxiv2023

Population-level neural correlates of flexible avoidance learning in medial prefrontal cortex

The medial prefrontal cortex (mPFC) has been proposed to link sensory inputs and behavioral outputs to mediate the execution of learned behaviors. However, how such a link is implemented has remained unclear. To measure prefrontal neural correlates of sensory stimuli and learned behaviors, we performed population calcium imaging during a novel tone-signaled active avoidance paradigm in mice. We developed a novel analysis approach based on dimensionality reduction and decoding that allowed us to identify and isolate population activity patterns related the tone stimulus, learned avoidance actions and general motion. While tone-related activity was not informative about behavior, avoidance-related activity was predictive of upcoming avoidance actions. Moreover, avoidance-related activity distinguished between two different learned avoidance actions, consistent with a model in which mPFC contributes to the selection between different goal-directed actions. Overall, our results suggest that mPFC circuit dynamics transform sensory inputs into specific behavioral outputs through distributed population-level computations.

neuroscience↗

bioRxiv2022

A new theoretical framework jointly explains behavioral and neural variability across subjects performing flexible decision-making

The ability to flexibly switch our response to external stimuli according to contextual information is critical for successful interactions with a complex world. Context-dependent computations are necessary across many domains1-3, yet their neural implementations remain poorly understood. Here we developed a novel behavioral task in rats to study context-dependent selection and accumulation of evidence for decision-making4-6. Under assumptions supported by both monkey and rat data, we first show mathematically that a network can solve this problem through a combination of three defined components. These components can be identified and tested directly with experimental data. We further show that existing electrophysiological and modeling data are compatible with the full variety of possible combinations of these components, suggesting that different individuals could use different component combinations. To study variability across individual subjects, we developed automated, high-throughput methods to train rats on our task, and we trained many subjects on it. Consistent with theoretical predictions, neural and behavioral analyses revealed substantial heterogeneity across rats, despite uniformly good task performance. Our theory further predicts a specific link between behavioral and neural signatures, which was robustly supported in the data. In summary, our results provide a new experimentally-supported theoretical framework to analyze individual variability in biological and artificial systems performing flexible decision-making tasks, they open the door to cellular-resolution studies of individual variability in higher cognition, and they provide insights into neural mechanisms of context-dependent computation more generally.

neuroscience↗

bioRxiv2022

Primate pre-arcuate cortex actively maintains persistent representations of saccades from plans to outcomes

Dorso-lateral prefrontal cortex (dlPFC) in primates plays a key role in the acquisition and execution of flexible, goal-directed behaviors. Recordings in monkey dlPFC have revealed possible neural correlates of the underlying cognitive processes like attention, planning, or decision-making, both at the single-neuron and population levels. Integrating these observations into a coherent picture of dlPFC function is challenging, as these studies typically focused on neural activity in relation to a few, specific events within a single, fully learned behavioral task. Here we obtain a more comprehensive description of dlPFC activity from a large dataset of population recordings in monkeys across a variety of behavioral contexts. We characterized neural activity in relation to saccades that monkeys made freely, or at different stages of training in multiple tasks involving instructed saccades, perceptual discriminations, and reward-based decisions. Across all contexts, we observed reliable and strong modulations of neural activity in relation to a retrospective representation of the most recent saccadic movement. Prospective, planning-like activity was instead limited to task-related, delayed saccades that were directly eligible for a reward. The link between prospective and retrospective representations was highly structured, potentially reflecting a hard-wired feature of saccade responses in these areas. Only prospective representations were modulated by the recent behavioral history, but neither representations were modulated by learning occurring over days, despite obvious concurrent behavioral changes. Dorso-lateral PFC thus combines tightly linked flexible and rigid representations, with a dominant contribution from retrospective signals maintaining the memory of past actions.

neuroscience↗

bioRxiv2022

Operative dimensions in unconstrained connectivity of recurrent neural networks

Recurrent Neural Networks (RNNs) are commonly used models to study neural computation. However, a comprehensive understanding of how dynamics in RNNs emerge from the underlying connectivity is largely lacking. Previous work derived such an understanding for RNNs fulfilling very specific constraints on their connectivity, but it is unclear whether the resulting insights apply more generally. Here we study how network dynamics are related to network connectivity in RNNs trained without any specific constraints on several tasks previously employed in neuroscience. Despite the apparent high-dimensional connectivity of these RNNs, we show that a low-dimensional, functionally relevant subspace of the weight matrix can be found through the identification of operative dimensions, which we define as components of the connectivity whose removal has a large influence on local RNN dynamics. We find that a weight matrix built from only a few operative dimensions is sufficient for the RNNs to operate with the original performance, implying that much of the high-dimensional structure of the trained connectivity is functionally irrelevant. The existence of a low-dimensional, operative subspace in the weight matrix simplifies the challenge of linking connectivity to network dynamics and suggests that independent network functions may be placed in specific, separate subspaces of the weight matrix to avoid catastrophic forgetting in continual learning.

neuroscience↗

bioRxiv2021

Dendritic integration of sensory and reward information facilitates learning

Learning goal-directed behavior requires association of pertinent sensory stimuli with behaviorally relevant outcomes. In the mammalian neocortex, dendrites of pyramidal neurons are suitable association sites1-3 but how their activities adapt during learning remains elusive. Here, we track calcium signals in apical dendrites of layer 5 (L5) pyramidal neurons in mouse barrel cortex during texture discrimination learning4. We observe diverse task-related responses, either localized to branches or widespread throughout the apical tuft. However, even in expert mice, the tufts capability to discriminate go/no-go stimuli remains poor. Yet, we identify two prevailing response types in dendritic branches: 1) responses to unexpected outcome (reward) in naive mice that decrease with growing task proficiency, and 2) responses associated with salient sensory stimuli, especially the outcome-predicting texture touch, that strengthen upon learning. We demonstrate that these response types match distinct unsigned components of the temporal difference error5 by replicating our results with a reinforcement learning model. Moreover, optogenetic apical inhibition of L5 neurons during the outcome time window prevents naive animals from learning the task, consistent with the effect of equivalent model perturbation. Our findings suggest that salience signals in L5 apical dendrites facilitate the recruitment of task-relevant neurons via dendritic gain modulation.

neuroscience↗

bioRxiv2021

Residual dynamics resolves recurrent contributions to neural computation

Relating neural activity to behavior requires an understanding of how neural computations arise from the coordinated dynamics of distributed, recurrently connected neural populations. However, inferring the nature of recurrent dynamics from partial recordings of a neural circuit presents significant challenges. Here, we show that some of these challenges can be overcome by a fine-grained analysis of the dynamics of neural residuals, i.e. trial-by-trial variability around the mean neural population trajectory for a given task condition. Residual dynamics in macaque pre-frontal cortex (PFC) in a saccade-based perceptual decision-making task reveals recurrent dynamics that is time-dependent, but consistently stable, and suggests that pronounced rotational structure in PFC trajectories during saccades is driven by inputs from upstream areas. The properties of residual dynamics restrict the possible contributions of PFC to decision-making and saccade generation, and suggest a path towards fully characterizing distributed neural computations with large-scale neural recordings and targeted causal perturbations.

neuroscience↗

bioRxiv2020

SIPEC: the deep-learning Swiss knife for behavioral data analysis

The quantification of behaviors of interest from video data is commonly used to study brain function, the effects of pharmacological interventions, and genetic alterations. Existing approaches lack the capability to analyze the behavior of groups of animals in complex environments. We present a novel deep learning architecture for classifying individual and social animal behavior, even in complex environments directly from raw video frames, while requiring no intervention after initial human supervision. Our behavioral classifier is embedded in a pipeline (SIPEC) that performs segmentation, identification, pose-estimation, and classification of complex behavior, outperforming the state of the art. SIPEC successfully recognizes multiple behaviors of freely moving individual mice as well as socially interacting non-human primates in 3D, using data only from simple mono-vision cameras in home-cage setups.

neuroscience↗

bioRxiv2019

Prefrontal cortex exhibits multi-dimensional dynamic encoding during decision-making

Recent work has suggested that prefrontal cortex (PFC) plays a key role in context-dependent perceptual decision-making. Here we investigate population-level coding of decision variables in monkey PFC using a new method for identifying task-relevant dimensions of neural activity. Our analyses reveal that, in contrast to one-dimensional attractor models, PFC has a multi-dimensional code for decisions, context, and relevant as well as irrelevant sensory information. Moreover, these representations evolve in time, with an early linear accumulation phase followed by a phase with rotational dynamics. We identify the dimensions of neural activity associated with these phases, and show that they are not the product of distinct populations, but of a single population with broad tuning characteristics. Finally, we use model-based decoding to show that the transition from linear to rotational dynamics coincides with a sustained plateau in decoding accuracy, revealing that rotational dynamics in PFC preserve sensory as well as choice information for the duration of the stimulus integration period.

neuroscience↗

bioRxiv2019

Neighborhood-statistics reveal complex dynamics of song acquisition in the zebra finch

Motor behaviors are continually shaped by a variety of processes such as environmental influences, development, and learning1,2. The resulting behavioral changes are commonly quantified based on hand-picked features3-10 (e.g. syllable pitch11) and assuming discrete classes of behaviors (e.g. distinct syllables)3-5,9,10,12-17. Such methods may generalize poorly across behaviors and species and are necessarily biased. Here we present an account of behavioral change based on nearest-neighbor statistics18-23 that avoids such biases and apply it to song development in the juvenile zebra finch3. First, we introduce the concept of repertoire dating, whereby each syllable rendition is dated with a \"pseudo\" production-day corresponding to the day when similar renditions were typical in the behavioral repertoire. Differences in pseudo production-day across renditions isolate the components of vocal variability congruent with the long-term changes due to vocal learning and development. This variability is large, as about 10% of renditions have pseudo production-days falling more than 10 days into the future (anticipations) or into the past (regressions) relative to their actual production time. Second, we obtain a holistic, yet low-dimensional, description of vocal change in terms of a behavioral trajectory, which reproduces the pairwise similarities between renditions grouped by production time and pseudo production-day24. The behavioral trajectory reveals multiple, previously unrecognized components of behavioral change operating at distinct time-scales. These components interact differently across the behavioral repertoire--diurnal change in regressions undergoes only weak overnight consolidation4,5, whereas anticipations and typical renditions consolidate fully2,6,25. Our nearest-neighbor methods yield model-free descriptions of how behavior evolves relative to itself, rather than relative to a potentially arbitrary, experimenter-defined, goal3-5,11. Because of their generality, our methods appear well-suited to comparing learning across behaviors and species1,26-32, and between biological and artificial systems.

neuroscience↗

bioRxiv2019

Spatial and temporal structure of choice representations in primate prefrontal cortex

Divergent accounts of how choices are represented by neural populations have led to conflicting explanations of the underlying mechanisms of decision-making, ranging from persistent, attractor-based dynamics to transient, sequence-based dynamics. To evaluate these mechanisms, we characterize the spatial and temporal structure of choice representations in large neural populations in prefrontal cortex. We find that the pronounced diversity of choice responses across neurons reflects only a few, mostly persistent population patterns recruited at progressively later times before and after a choice. Brief sequential activity occurs during a saccadic choice, but is entirely absent in a delay preceding it. The diversity of choice responses, which could result from almost-random connectivity in the underlying circuits, instead largely reflects the topographical arrangement of response-field properties across the cortical surface. This spatial organization appears to form a fixed scaffold upon which the context-dependent representations of task-specific variables often observed in prefrontal cortex can be learned.

neuroscience↗