Biology subjects

Pouget, A.

Publications and source records attributed to Pouget, A..

At least 19 recordsLinked to original sources

bioRxiv2026

Structural Composition Enables Very Fast Learning

AO_SCPLOWBSTRACTC_SCPLOWThere is a gap between neuroscientific theories of learning and the speed of learning observed in many experiments. Since the Cognitive Revolution of the 1950s, compositionality has played a central role in efforts to bridge this gap. Roughly, a compositional system is one where distinct modules are combined according to a set of rules in order to accomplish complex tasks. Recently, significant progress has been made in understanding the emergence of modules in both biological and artificial neural systems. How, and under what conditions, the rules of module recombination are represented in these systems remains an open question. Here we present a neural model that can leverage these rules to dramatically speed up learning. We first show that when faced with multiple tasks which share subcomponents, models learn a low-dimensional representation that captures how subcomponents are reused across the task set. These low-dimensional spaces encode the structure that governs how modules should be recombined. Restricting learning to these subspaces greatly reduces the amount of experience needed to acquire a novel task, even when learning from reinforcement on single trials. In some cases, we can leverage the geometric regularities of these representations to reduce learning to a form of hypothesis testing over a small set of discrete points. Finally, we use this theory to model both behavioral and neural data from non-human primates performing a compositional task, and show that key features in this data are consistent with a model in which exploration during learning is restricted to these low-dimensional spaces. Overall, this work shows that the advantages of modularity in neural systems can be greatly improved upon when models represent the structure of module reuse. Both these features working in tandem lead to learning on timescales similar to biological intelligences, and hence provide a model for how such fast, adaptable behavior can emerge from systems of neurons.

neuroscience↗

bioRxiv2026

Augmenting the Bayesian Brain with learned and reusable world-model components for flexible cognition

The Bayesian Brain hypothesis assumes that cognition relies on internal generative models of the world, yet existing implementations remain constrained by pre-specified, task-specific generative structures and computationally heavy iterative inference schemes. Here, we introduce modular neural state-space models as a scalable realization of the Bayesian Brain, replacing fixed generative structures and pre-specified inference rules with learned world-model components and amortized neural updates. This framework preserves the core commitment to explaining observations through hidden causes while making inference learned and reusable rather than pre-specified and task-specific. Our modular implementation of these models affords learned components to be seamlessly recombined and stacked across superficially different tasks that share similar latent dynamics. Such computational reuse supports zero-shot generalization and predicts selective correlations of inference parameters between tasks. We confirm these key predictions in human behavior, identifying learned and reusable world-model components as a candidate computational principle for flexible cognition.

neuroscience↗

bioRxiv2026

PDZD-8 interacts with LGG-2/LC3 to promote autolysosome formation

LGG-2/LC3 and LGG-1/GABARAP are members of the Atg8 ubiquitin-like protein family, playing central roles in autophagy. Through a yeast two-hybrid screen, we identified the endoplasmic reticulum protein PDZD-8 as a novel interactor of the C. elegans autophagy protein LGG-2/LC3. We demonstrate that PDZD-8 binds specifically to LGG-2 and not LGG-1 via a canonical LC3-interacting region (LIR) and colocalizes with LGG-2-positive autophagosomes in vivo. Loss of PDZD-8 leads to a transient accumulation of enlarged LGG-2-positive autophagosomes that remain closely associated with lysosomes. Genetic disruption of core autophagy genes suppresses the pdzd-8 mutant phenotype, indicating that this phenotype is dependent on autophagy. Moreover, loss of rab-7, which blocks autophagosome-lysosome fusion, also suppresses the pdzd-8 phenotype, suggesting that the RAB-7-dependent fusion step is epistatic to PDZD-8 and that PDZD-8 acts downstream of autophagosome-lysosome maturation. Microscopy approaches (TEM, CLEM, and FIB-SEM) reveal an accumulation of fused autophagosome-lysosome structures, pointing to a transient block in autophagic flux in pdzd-8 mutants. Furthermore, PDZD-8 lacking either the SMP lipid-transfer domain or the LIR motif phenocopies the pdzd-8 null mutant, underscoring the importance of both domains. Our findings support a model in which PDZD-8, in conjunction with LGG-2/LC3, mediates ER-autophagosome/autolysosome contacts that are critical for subsequent autolysosome completion. Significance StatementAutophagy is a conserved cellular degradation pathway essential for maintaining homeostasis and responding to stress. Although the molecular machinery driving autophagosome formation and fusion with lysosomes is relatively well-characterized, the contribution of membrane contact sites to autophagic flux remains poorly understood. Here, we identify PDZD-8, an endoplasmic reticulum (ER) protein, as a novel interactor of the Caenorhabditis elegans LC3/Atg8 homolog LGG-2. We show that PDZD-8 regulates the completion of autolysosome maturation by bridging ER-autophagosome/autolysosome contacts. Loss of PDZD-8 results in the accumulation of enlarged autophagosomes adjacent to lysosomes and a transient block in autophagic flux, a phenotype that depends on the RAB-7-mediated tethering step. These findings reveal a previously unrecognized role for PDZD-8 in coordinating ER-autophagosome/autolysosome dynamics during autophagy.

cell biology↗

bioRxiv2024

Exploiting correlations across trials and behavioral sessions to improve neural decoding

Traditional neural decoders model the relationship between neural activity and behavior within individual trials of a single experimental session, neglecting correlations across trials and sessions. However, animals exhibit similar neural activities when performing the same behavioral task, and their behaviors are influenced by past experiences from previous trials. To exploit these informative correlations in large datasets, we introduce two complementary models: a multi-session reduced-rank regression model that shares similar behaviorally-relevant statistical structure in neural activity across sessions to improve decoding, and a multi-session state-space model that shares similar behavioral statistical structure across trials and sessions. Applied across 433 sessions spanning 270 brain regions in the International Brain Laboratory public mouse Neuropixels dataset, our decoders demonstrate improved decoding accuracy for four distinct behaviors compared to traditional approaches. These results generalize across additional datasets, species, and behavioral tasks. Unlike existing deep learning approaches, our models are interpretable and efficient, uncovering low-dimensional representations that predict animal decisions, quantifying single-neuron contributions to decoding behaviors, and identifying different activation timescales of neural activity across the brain. Code: https://github.com/yzhang511/neural_decoding.

neuroscience↗

bioRxiv2024

Weber's law and natural inference

For most extensive sensory variables such as speed or numerosity, the discrimination thresholds of human subjects are proportional to the value around which the discrimination is performed, a scaling known as Webers law. Many theories have been proposed for this law, which all rely on the assumption that neurons are noisy. By contrast, we argue here that noisy neurons are not required to explain Webers law. Instead, we propose that it is the unavoidable consequence of the statistics of natural sensory inputs. In natural environments, sensory measurements are typically scaled by global variables such as contrast in vision or loudness in audition. These global scaling parameters induce positive correlations among measurements which in turn lead to Webers scaling. This theory makes testable experimental predictions and accounts for the fact that tuning curves to speed and numerosity in vivo are approximately log normal.

animal behavior and cognition↗

bioRxiv2024

Reticulon dependent ER-phagy mediates adaptation to heat stress in Caenorhabditis elegans

The selective degradation of ER by autophagy, named ER-phagy, promotes the recovery of ER homeostasis after a stress. Depending on the ER stress, different types of ER-phagy involve various selective autophagy receptors. In this study, we report a macroER-phagy induced by the fragmentation of tubular ER in response to acute heat stress. We identified a novel ER-phagy receptor encoded by the reticulon long isoform RET-1d. RET-1d is mainly expressed in the nervous system and the epidermis and colocalize with the ubiquitin-like autophagy protein LGG-1/GABARAP during heat stress induced autophagy. Two LIR motifs in the long intrinsically disordered region of RET-1d mediate its interaction with LGG-1 protein. The specific depletion of RET-1d isoform resulted in a delay in autophagosome biogenesis and a decrease in the capacity of animals to adapt to heat stress. Our data revealed a RET-1d dependent ER-phagy mechanism that takes place in neurons and epidermis and participates to the adaptation of C. elegans to heat stress.

cell biology↗

bioRxiv2024

Acquiring musculoskeletal skills with curriculum-based reinforcement learning

Efficient musculoskeletal simulators and powerful learning algorithms provide computational tools to tackle the grand challenge of understanding biological motor control. Our winning solution for the inaugural NeurIPS MyoChallenge leverages an approach mirroring human skill learning. Using a novel curriculum learning approach, we trained a recurrent neural network to control a realistic model of the human hand with 39 muscles to rotate two Baoding balls in the palm of the hand. In agreement with data from human subjects, the policy uncovers a small number of kinematic synergies even though it is not explicitly biased towards low-dimensional solutions. However, by selectively inactivating parts of the control signal, we found that more dimensions contribute to the task performance than suggested by traditional synergy analysis. Overall, our work illustrates the emerging possibilities at the interface of muscu-loskeletal physics engines, reinforcement learning and neuro-science to advance our understanding of biological motor control.

neuroscience↗

bioRxiv2023

Multi-timescale reinforcement learning in the brain

To thrive in complex environments, animals and artificial agents must learn to act adaptively to maximize fitness and rewards. Such adaptive behavior can be learned through reinforcement learning1, a class of algorithms that has been successful at training artificial agents2-6 and at characterizing the firing of dopamine neurons in the midbrain7-9. In classical reinforcement learning, agents discount future rewards exponentially according to a single time scale, controlled by the discount factor. Here, we explore the presence of multiple timescales in biological reinforcement learning. We first show that reinforcement agents learning at a multitude of timescales possess distinct computational benefits. Next, we report that dopamine neurons in mice performing two behavioral tasks encode reward prediction error with a diversity of discount time constants. Our model explains the heterogeneity of temporal discounting in both cue-evoked transient responses and slower timescale fluctuations known as dopamine ramps. Crucially, the measured discount factor of individual neurons is correlated across the two tasks suggesting that it is a cell-specific property. Together, our results provide a new paradigm to understand functional heterogeneity in dopamine neurons, a mechanistic basis for the empirical observation that humans and animals use non-exponential discounts in many situations 10-14, and open new avenues for the design of more efficient reinforcement learning algorithms.

neuroscience↗

bioRxiv2023

A Brain-Wide Map of Neural Activity during Complex Behaviour

A key challenge in neuroscience is understanding how neurons in hundreds of interconnected brain regions integrate sensory inputs with prior expectations to initiate movements and thereby make decisions. It is difficult to meet this challenge if different laboratories apply different analyses to different recordings in different regions during different behaviours. Here, we report a comprehensive set of recordings from 621733 neurons across 139 mice in 12 labs performing a decision-making task with sensory, motor, and cognitive components, obtained with 699 Neuropixels probe insertions covering 279 brain areas in the left forebrain and midbrain and the right hindbrain and cerebellum. We provide an initial appraisal of this brain-wide map, assessing how neural activity encoded key task variables. Representations of visual stimuli appeared transiently in classical visual areas after stimulus onset and then spread to ramp-like activity in a collection of mid- and hindbrain regions that also encoded choices. Neural responses correlated with motor action almost everywhere in the brain. Responses to reward delivery and consumption versus negative feedback were also widespread. This publicly available dataset represents an unprecedented resource for understanding how computations distributed across and within brain areas drive behaviour.

neuroscience↗

bioRxiv2023

Computational principles of adaptive multisensory combination in the Drosophila larva

Many sensory systems have evolved to optimally combine signals from multiple sensory modalities to improve perception. While theories have been proposed to explain how this process is accomplished through probabilistic inference using large neural populations in vertebrates, how animals with dramatically smaller nervous systems such as the Drosophila melanogaster larva achieve multisensory combination remains elusive. Here, we systematically characterize larval navigation in different configurations of odor and temperature gradients with optogenetically-controlled noise. Using a data-driven agent-based model, we find that larvae adapt to the reliability of individual sensory signals, and in some cases minimize the variance of the combined signal. Besides firmly establishing that probabilistic inference directs natural orientation behaviors in the Drosophila larva, our results indicate that the exact mechanism underlying the combination of sensory information may be modality-dependent. By underscoring that probabilistic inference is inherent to insect nervous systems, our work opens the way for studying its neural implementation.

neuroscience↗

bioRxiv2022

A Neural Model of Task Compositionality with Natural Language Instructions

AO_SCPLOWBSTRACTC_SCPLOWOne of humans most fundamental cognitive feats is the ability to interpret linguistic instructions in order to perform novel tasks without any explicit experience with the task. Yet, the computations that the brain might use to accomplish such a feat remains poorly understood. Here we use the latest advances in Natural Language Processing to create a neural model of generalization based on linguistic instructions. Models are trained on a set of commonly studied psychophysical tasks, and receive instructions embedded by a pre-trained language model. Our best models can perform a previously unseen task with a performance of 83% correct on average based solely on linguistic instructions (i.e. 0-shot learning). We found that language scaffolds sensorimotor representations such that activity for interrelated tasks share a common geometry with the semantic representations of instructions, allowing language to cue the proper composition of practiced skills in unseen settings. Finally, we show how this model can generate a linguistic description of a novel task it has identified using only motor feedback, which can subsequently guide a partner model to perform the task. Our models offer several experimentally testable predictions outlining how linguistic information must be represented in order to facilitate flexible and general cognition in the human brain.

neuroscience↗

bioRxiv2021

Morphological determinants of bite force capacity in insects: a biomechanical analysis of polymorphous leaf-cutter ants

The extraordinary success of social insects is partially based on division of labour, i. e. individuals exclusively or preferentially perform specific tasks. Task-preference may correlate with morphological adaptations so implying task-specialisation, but the extent of such specialisation can be difficult to determine. Here, we demonstrate how the physical foundation of some tasks can be leveraged to quantitatively link morphology and performance. We study the allometry of bite force capacity in Atta vollenweideri leaf-cutter ants, polymorphic insects in which the mechanical processing of plant material is a key aspect of the behavioural portfolio. Through a morphometric analysis of tomographic scans, we show that the bite force capacity of the heaviest colony workers is twice as large as predicted by isometry. This disproportionate boost is predominantly achieved through increased investment in muscle volume; geometrical parameters such as mechanical advantage, fibre length or pennation angle are likely constrained by the need to maintain a constant mandibular opening range. We analyse this preference for an increase in size-specific muscle volume and the adaptations in internal and external head anatomy required to accommodate it with simple geometric and physical models, so providing a quantitative understanding of the functional anatomy of the musculoskeletal bite apparatus in insects.

evolutionary biology↗

bioRxiv2020

Mice alternate between discrete strategies during perceptual decision-making

Classical models of perceptual decision-making assume that subjects use a single, consistent strategy to form decisions, or that strategies evolve slowly over time. Here we present new analyses suggesting that this common view is incorrect. We analyzed data from mouse and human decision-making experiments and found that choice behavior relies on an interplay between multiple interleaved strategies. These strategies, characterized by states in a hidden Markov model, persist for tens to hundreds of trials before switching, and may alternate multiple times within a session. The identified mouse decision-making strategies were highly consistent across individuals and comprised a single "engaged" state, in which decisions relied heavily on the sensory stimulus, and several biased states in which errors frequently occurred. These results provide a powerful alternate explanation for "lapses" often observed in rodent psychophysical experiments, and suggest that standard measures of performance mask the presence of dramatic changes in strategy across trials.

neuroscience↗

bioRxiv2020

Internally generated population activity in cortical networks hinders information transmission

How neuronal variability impacts neuronal codes is a central question in systems neuroscience, often with complex and model dependent answers. Most population models are parametric, with a tacitly assumed structure of neuronal tuning and population-wide variability. While these models provide key insights, they purposely divorce any mechanistic relationship between trial average and trial variable neuronal activity. By contrast, circuit based models produce activity with response statistics that are reflection of the underlying circuit structure, and thus any relations between trial averaged and trial variable activity are emergent rather than assumed. In this work, we study information transfer in networks of spatially ordered spiking neuron models with strong excitatory and inhibitory interactions, capable of producing rich population-wide neuronal variability. Motivated by work in the visual system we embed a columnar stimulus orientation map in the network and measure the population estimation of an orientated input. We show that the spatial structure of feedforward and recurrent connectivity are critical determinants for population code performance. In particular, when network wiring supports stable firing rate activity then with a sufficiently large number of decoded neurons all available stimulus information is transmitted. However, if the inhibitory projections place network activity in a pattern forming regime then the population-wide dynamics compromise information flow. In total, network connectivity determines both the stimulus tuning as well as internally generated population-wide fluctuations and thereby dictates population code performance in complicated ways where modeling efforts provide essential understanding.

neuroscience↗

bioRxiv2019

Strong information-limiting correlations in early visual areas

If the brain processes incoming data efficiently, information should degrade little between early and later neural processing stages, and so information in early stages should match behavioral performance. For instance, if there is enough information in a visual cortical area to determine the orientation of a grating to within 1 degree, and the code is simple enough to be read out by downstream circuits, then animals should be able to achieve that performance behaviourally. Despite over 30 years of research, it is still not known how efficient the brain is. For tasks involving a large number of neurons, the amount of information encoded by neural circuits is limited by differential correlations. Therefore, determining how much information is encoded requires quantifying the strength of differential correlations. Detecting them, however, is difficult. We report here a new method, which requires on the order of 100s of neurons and trials. This method relies on computing the alignment of the neural stimulus encoding direction, f', with the eigenvectors of the noise covariance matrix,{Sigma} . In the presence of strong differential correlations, f' must be spanned by a small number of the eigenvectors with largest eigenvalues. Using simulations with a leaky-integrate-and-fire neuron model of the LGN-V1 circuit, we confirmed that this method can indeed detect differential correlations consistent with those that would limit orientation discrimination thresholds to 0.5-3 degrees. We applied this technique to V1 recordings in awake monkeys and found signatures of differential correlations, consistent with a discrimination threshold of 0.47-1.20 degrees, which is not far from typical discrimination thresholds (1-2 deg). These results suggest that, at least in macaque monkeys, V1 contains about as much information as is seen in behaviour, implying that downstream circuits are efficient at extracting the information available in V1.

neuroscience↗

bioRxiv2019

Optimal policy for multi-alternative decisions

Every-day decisions frequently require choosing among multiple alternatives. Yet, the optimal policy for such decisions is unknown. Here we derive the normative policy for general multi-alternative decisions. This strategy requires evidence accumulation to nonlinear, time-dependent bounds, that trigger choices. A geometric symmetry in those boundaries allows the optimal strategy to be implemented by a simple neural circuit involving a normalization with fixed decision bounds and an urgency signal. The model captures several key features of the response of decision-making neurons as well as the increase in reaction time as a function of the number of alternatives, known as Hicks law. In addition, we show that, in the presence of divisive normalization and internal variability, our model can account for several so called irrational behaviors such as the similarity effect as well as the violation of both the independent irrelevant alternative principle and the regularity principle.

neuroscience↗

bioRxiv2018

The impact of learning on perceptual decisions and its implication for speed-accuracy tradeoffs

In standard models of perceptual decision-making, noisy sensory evidence is considered to be the primary source of choice errors and the accumulation of evidence needed to overcome this noise gives rise to speed-accuracy tradeoffs. Here, we investigated how the history of recent choices and their outcomes interacts with these processes using a combination of theory and experiment. We found that the speed and accuracy of performance of rats on olfactory decision tasks could be best explained by a Bayesian model that combines reinforcement-based learning with accumulation of uncertain sensory evidence. This model predicted the specific pattern of trial history effects that were found in the data. The results suggest that learning is a critical factor contributing to speed-accuracy tradeoffs in decision-making and that task history effects are not simply biases but rather the signatures of an optimal learning strategy.

neuroscience↗

bioRxiv2018

Neural Correlates of Optimal Multisensory Decision Making

Perceptual decisions are often based on multiple sensory inputs whose reliabilities rapidly vary over time, yet little is known about how our brain integrates these inputs to optimize behavior. Here we show multisensory evidence with time-varying reliability can be accumulated near optimally, in a Bayesian sense, by simply taking time-invariant linear combinations of neural activity across time and modalities, as long as the neural code for the sensory inputs is close to an invariant linear probabilistic population code (ilPPC). Recordings in the lateral intraparietal area (LIP) while macaques optimally performed a vestibular-visual multisensory decision-making task revealed that LIP population activity reflects an integration process consistent with the ilPPC theory. Moreover, LIP accumulates momentary evidence proportional to vestibular acceleration and visual velocity which are encoded in sensory areas with a close approximation to ilPPCs. Together, these results provide a remarkably simple and biologically plausible solution to optimal multisensory decision making.

neuroscience↗