DOCSLIB.ORG

Efficient and Robust Multi-Task Learning in the Brain with Modular Task Primitives

Total Page：16

File Type：pdf, Size：1020Kb

Download full-text PDF Read full-text
Efficient and Robust Multi-Task Learning in the Brain with Modular Task Primitives Efficient and robust multi-task learning in the brain with modular task primitives. Christian D. Márton Guillaume Lajoiey Friedman Brain Institute Dep. of Mathematics and Statistics Icahn School of Medicine at Mount Sinai Université de Montréal New York, NY 10029 MILA - Québec AI Institute [email protected] [email protected] Kanaka Rajany Friedman Brain Institute Icahn School of Medicine at Mount Sinai New York, NY 10029 [email protected] Abstract In a real-world setting biological agents do not have infinite resources to learn new things. It is thus useful to recycle previously acquired knowledge in a way that allows for faster, less resource-intensive acquisition of multiple new skills. Neural networks in the brain are likely not entirely re-trained with new tasks, but how they leverage existing computations to learn new tasks is not well understood. In this work, we study this question in artificial neural networks trained on commonly used neuroscience paradigms. Building on recent work from the multi-task learning literature, we propose two ingredients: (1) network modularity, and (2) learning task primitives. Together, these ingredients form inductive biases we call structural and functional, respectively. Using a corpus of nine different tasks, we show that a modular network endowed with task primitives allows for learning multiple tasks well while keeping parameter counts, and updates, low. We also show that the skills acquired with our approach are more robust to a broad range of perturbations compared to those acquired with other multi-task learning strategies. This work offers a new perspective on achieving efficient multi-task learning in the brain, and makes predictions for novel neuroscience experiments in which targeted perturbations are employed to explore solution spaces. arXiv:2105.14108v1 [cs.AI] 28 May 2021 1 Introduction Task knowledge can be stored in the collective dynamics of large populations of neurons in the brain [37, 48, 64, 46, 27, 68, 10, 45, 58, 34, 57, 2, 8, 56]. This body of work has offered insight into the computational mechanisms underlying the execution of multiple different tasks across various domains. In these studies, however, animals and models are over-trained on specific tasks, at the expense of long training duration and high energy expenditure. Models are optimized for one specific task, and unlike networks of neurons in the brain, are not meant to be re-used across tasks. In addition to being resource-efficient, it is useful that acquired task representations be resistant to failures and perturbations. Biological organisms are remarkably resistant to lesions affecting large extents of the neural circuit [3, 12], and to neural cell death with aging [18, 63]. Current approaches Preprint. Under review. do not take this into account; often, representations acquired naively are highly susceptible to perturbations [60]. In the multi-task setting, where multiple tasks are learnt simultaneously [64] or sequentially [15, 69, 28], the amount of weight updates performed grows quickly with the number of tasks as the entire model is essentially being re-trained with each new task. Training time and energy cost increase rapidly and can become prohibitive. To overcome this, it is conceivable that the brain has adopted particular inductive biases to speed up learning over an evolutionary timescale [48]. Inductive biases can help guide and accelerate future learning by providing useful prior knowledge. This becomes particularly useful when resources (energy, time, neurons) to acquire new skills are limited, and the space of tasks that need to be acquired is not infinite. In fact, biological agents often inhabit particular ecological niches [9] which offer very particular constraints on the kinds of tasks that are useful and need to be acquired over a lifetime. Under these circumstances, useful priors or inductive biases can help decrease the amount of necessary weight updates. One way in which such biases may be expressed in the brain is in terms of neural architecture or structure [48]. Neural networks in the brain may come with a pre-configured wiring diagram, encoded in the genome [67]. Such structural pre-configurations can put useful constraints on the learning problem. One way in which this is expressed in the brain is modularity. The brain exhibits specializations over various scales [55, 62, 21, 36, 35], with visual information being primarily represented in the visual cortices, auditory information in the auditory cortices, and decision-making tasks encoded in circuits in the prefrontal cortex [37, 10, 34]. Further subdivisions within those specialized modules may arise in the form of particular cell clusters, or groupings in a graph-theoretic sense, processing particular kinds of information (such as cells mapping particular visual features or aspects of a task). From a multi-task perspective, modularity can offer an efficient solution to the problem of task interference and catastrophic forgetting in continual learning [23, 40, 49], while increasing robustness to perturbations. Another way in which inductive biases may be expressed in the brain, beyond structure, is in terms of function: the types of neural dynamics instantiated in particular groups of neurons, or neural ensembles. Rather than learn separate models for separate tasks, as many of the studies cited above do, or re-train the same circuit many times on different tasks [64, 15], core computational principles could be learned once and re-used across tasks. There is evidence that the brain implements stereotyped dynamics which can be used across tasks [13, 51]. Ensembles of neurons may be divided into distinct groups based on the nature of the computations they perform, forming functional modules [42, 55, 5, 44, 35]. In the motor domain, for instance, neural activity was found to exhibit shared structure across tasks [19]. Different neural ensembles can be found to implement particular aspects of a task [29, 64, 10] and theoretical work suggests the number of different neural sub-populations influences the set of dynamics a neural network is able to perform [4]. Thus, a network that implements a form of task primitives, or basis-computations, could offer dynamics useful for training multiple tasks. Putting these different strands together, in this work we investigate the advantage of combining structural and functional inductive biases in a recurrent neural network (RNN) model of modular task primitives. Structural specialization in the form of modules can help avoid task interference and increase robustness to failure of particular neurons, while core functional specialization in the form of task primitives can help substantially lower the costs of training. We show how our model can be used to facilitate resource-efficient learning in a multi-task setting. We also compare our approach to other training paradigms, and demonstrate higher robustness to perturbations. Altogether, our work shows that combining structural and functional inductive biases in a recurrent network allows an agent to learn many tasks well, at lower cost and higher robustness compared to other approaches. This draws a path towards cheaper multi-task solutions, and offers new hypotheses for how multi-task learning may be achieved in the brain. 2 Background The brain is highly recurrently connected [30], and recurrent neural networks have been successfully used to model dynamics in the brain [37, 48, 64, 46, 27, 68, 10, 45, 58, 34, 57, 2, 8, 56]. Reservoir 2 computing approaches, such as those proposed in the context of echo-state networks (ESNs) and liquid state machines (LSMs) [41, 26, 32], avoid learning the recurrent weights of a network (only training output weights) and thereby achieve reduced training time and lower computational cost compared to recurrent neural networks (RNNs) whose weights are all trained in an end-to-end fashion. This can be advantageous: when full state dynamics are available for training, reservoir networks exhibit higher predictive performance and lower generalisation error at significantly lower training cost, compared to recurrent networks fully trained with gradient descent using backpropagation through time (BPTT); in the face of partial observability, however, RNNs trained with BPTT perform better [61]. Also, naive implementations of reservoir networks show high susceptibility to local perturbations, even when implemented in a modular way [60]. Module training inputs Modular RNN Tasks trained Modules performing task primitives 1 0.5 M1 Instructions Fixation 0 Autoencode 1 0.5 M2 T1 T2 T3 T4 Inputs Input 1 0 Memory 1 Input 2 0 -0.5 M3 Inputs Memory 2 -1 Fix Stim Go Recurrent Weights Output Weights 0 50 100 150 200 Time (msec) Fig 1. Schematic of fully disconnected recurrent network with modular task primitives. Left: inputs for modules M1-M3. Middle: recurrent weight matrix of the modular network, with separate partitions for modules M1-M3 along the diagonal. Right: different tasks (T1-T9) trained into different output weights, leveraging the same modular network. Meanwhile, previous work in continual learning has largely focused on feedforward architectures [23, 66, 1, 40, 20, 65, 53, 52, 68, 28, 69, 31, 14, 50]. The results obtained in feedforward systems, however, are not readily extensible to continual learning in dynamical systems, where recurrent interactions may lead to unexpected behavior in the evolution of neural responses over time [54, 15]. Previously, modular approaches have also largely been explored in feedforward systems [59, 16, 24]. Recently, a method for continual learning has been proposed for dynamical systems [15]. The authors have shown that their method, dynamical non-interference, outperforms other gradient-based approaches [69, 28] in key applications. All of these approaches, however, require the entire network to be re-trained with each new task and thus come with a high training cost. A network with functionally specialized modules could provide dynamics that are useful across tasks and thus help mitigate this problem. In the use of modules, our approach is similar to memory-based approaches to continual learning which have also been applied in recurrently connected architectures [22, 11].
Load more

Recommended publications
  • Dynamic Compression and Expansion in a Classifying Recurrent Network
    bioRxiv preprint doi: https://doi.org/10.1101/564476; this version posted March 1, 2019. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under aCC-BY-NC-ND 4.0 International license. Dynamic compression and expansion in a classifying recurrent network Matthew Farrell1-2, Stefano Recanatesi1, Guillaume Lajoie3-4, and Eric Shea-Brown1-2 1Computational Neuroscience Center, University of Washington 2Department of Applied Mathematics, University of Washington 3Mila|Qu´ebec AI Institute 4Dept. of Mathematics and Statistics, Universit´ede Montr´eal Abstract Recordings of neural circuits in the brain reveal extraordinary dynamical richness and high variability. At the same time, dimensionality reduction techniques generally uncover low- dimensional structures underlying these dynamics when tasks are performed. In general, it is still an open question what determines the dimensionality of activity in neural circuits, and what the functional role of this dimensionality in task learning is. In this work we probe these issues using a recurrent artificial neural network (RNN) model trained by stochastic gradient descent to discriminate inputs. The RNN family of models has recently shown promise in reveal- ing principles behind brain function. Through simulations and mathematical analysis, we show how the dimensionality of RNN activity depends on the task parameters and evolves over time and over stages of learning. We find that common solutions produced by the network naturally compress dimensionality, while variability-inducing chaos can expand it. We show how chaotic networks balance these two factors to solve the discrimination task with high accuracy and good generalization properties.
    [Show full text]
  • Dynamics of Excitatory-Inhibitory Neuronal Networks With
    I (X;Y) = S(X) - S(X|Y) in c ≈ p + N r V(t) = V 0 + ∫ dτZ 1(τ)I(t-τ) P(N) = 1 V= R I N! λ N e -λ www.cosyne.org R j = R = P( Ψ, υ) + Mγ (Ψ, υ) σ n D +∑ j n k D k n MAIN MEETING Salt Lake City, UT Feb 27 - Mar 2 ................................................................................................................................................................................................................. Program Summary Thursday, 27 February 4:00 pm Registration opens 5:30 pm Welcome reception 6:20 pm Opening remarks 6:30 pm Session 1: Keynote Invited speaker: Thomas Jessell 7:30 pm Poster Session I Friday, 28 February 7:30 am Breakfast 8:30 am Session 2: Circuits I: From wiring to function Invited speaker: Thomas Mrsic-Flogel; 3 accepted talks 10:30 am Session 3: Circuits II: Population recording Invited speaker: Elad Schneidman; 3 accepted talks 12:00 pm Lunch break 2:00 pm Session 4: Circuits III: Network models 5 accepted talks 3:45 pm Session 5: Navigation: From phenomenon to mechanism Invited speakers: Nachum Ulanovsky, Jeffrey Magee; 1 accepted talk 5:30 pm Dinner break 7:30 pm Poster Session II Saturday, 1 March 7:30 am Breakfast 8:30 am Session 6: Behavior I: Dissecting innate movement Invited speaker: Hopi Hoekstra; 3 accepted talks 10:30 am Session 7: Behavior II: Motor learning Invited speaker: Rui Costa; 2 accepted talks 11:45 am Lunch break 2:00 pm Session 8: Behavior III: Motor performance Invited speaker: John Krakauer; 2 accepted talks 3:45 pm Session 9: Reward: Learning and prediction Invited speaker: Yael
    [Show full text]
  • COSYNE 2014 Workshops
    I (X;Y) = S(X) - S(X|Y) in c ≈ p + N r V(t) = V0 + ∫dτZ 1(τ)I(t-τ) P(N) = V= R I 1 N! λ N e -λ www.cosyne.org R j = γ R = P( Ψ, υ) + M(Ψ, υ) σ n D n +∑ j k D n k WORKSHOPS Snowbird, UT Mar 3 - 4 ................................................................................................................................................................................................................. COSYNE 2014 Workshops Snowbird, UT Mar 3-4, 2014 Organizers: Tatyana Sharpee Robert Froemke 1 COSYNE 2014 Workshops March 3 & 4, 2014 Snowbird, Utah Monday, March 3, 2014 Organizer(s) Location 1. Computational psychiatry – Day 1 Q. Huys Wasatch A T. Maia 2. Information sampling in behavioral optimization – B. Averbeck Wasatch B Day 1 R.C. Wilson M. R. Nassar 3. Rogue states: Altered Dynamics of neural activity in C. O’Donnell Magpie A brain disorders T. Sejnowski 4. Scalable models for high dimensional neural data I. M. Park Superior A E. Archer J. Pillow 5. Homeostasis and self-regulation of developing J. Gjorgjieva White Pine circuits: From single neurons to networks M. Hennig 6. Theories of mammalian perception: Open and closed E. Ahissar Magpie B loop modes of brain-world interactions E. Assa 7. Noise correlations in the cortex: Quantification, J. Fiser Superior B origins, and functional significance M. Lengyel A. Pouget 8. Excitatory and inhibitory synaptic conductances: M. Lankarany Maybird Functional roles and inference methods T. Toyoizumi Workshop Co-Chairs Email Cell Robert Froemke, NYU [email protected] 510-703-5702 Tatyana Sharpee, Salk [email protected] 858-610-7424 Maps of Snowbird are at the end of this booklet (page 38).
    [Show full text]
  • Predictive Coding in Balanced Neural Networks with Noise, Chaos and Delays
    Predictive coding in balanced neural networks with noise, chaos and delays Jonathan Kadmon Jonathan Timcheck Surya Ganguli Department of Applied Physics Department of Physics Department of Applied Physics Stanford University,CA Stanford University,CA Stanford University, CA [email protected] Abstract Biological neural networks face a formidable task: performing reliable compu- tations in the face of intrinsic stochasticity in individual neurons, imprecisely specified synaptic connectivity, and nonnegligible delays in synaptic transmission. A common approach to combatting such biological heterogeneity involves aver- aging over large redundantp networks of N neurons resulting in coding errors that decrease classically as 1= N. Recent work demonstrated a novel mechanism whereby recurrent spiking networks could efficiently encode dynamic stimuli, achieving a superclassical scaling in which coding errors decrease as 1=N. This specific mechanism involved two key ideas: predictive coding, and a tight balance, or cancellation between strong feedforward inputs and strong recurrent feedback. However, the theoretical principles governing the efficacy of balanced predictive coding and its robustness to noise, synaptic weight heterogeneity and communica- tion delays remain poorly understood. To discover such principles, we introduce an analytically tractable model of balanced predictive coding, in which the de- gree of balance and the degree of weight disorder can be dissociated unlike in previous balanced network models, and we develop a mean field theory of coding accuracy. Overall, our work provides and solves a general theoretical framework for dissecting the differential contributions neural noise, synaptic disorder, chaos, synaptic delays, and balance to the fidelity of predictive neural codes, reveals the fundamental role that balance plays in achieving superclassical scaling, and unifies previously disparate models in theoretical neuroscience.
    [Show full text]
  • COSYNE 2012 Workshops February 27 & 28, 2012 Snowbird, Utah
    COSYNE 2012 Workshops February 27 & 28, 2012 Snowbird, Utah Monday, February 27 Organizer(s) Location 1. Coding and Computation in visual short-term memory. W.J. Ma Superior A 2. Neuromodulation: beyond the wiring diagram, M. Higley Wasatch B adding functional flexibility to neural circuits. 3. Sensorimotor processes reflected in spatiotemporal Z. Kilpatrick Superior B of neuronal activity. J-Y. Wu 4. Characterizing neural responses to structured K. Rajan Wasatch A and naturalistic stimuli. W.Bialek 5. Neurophysiological and computational mechanisms of D. Freedman Magpie A categorization. XJ. Wang 6. Perception and decision making in rodents S. Jaramillio Magpie B A. Zador 7. Is it time for theory in olfaction? V.Murphy Maybird N. Uchida G. Otazu C. Poo Workshop Co-Chairs Email Cell Brent Doiron, Pitt [email protected] 412-576-5237 Jess Cardin, Yale [email protected] 267-235-0462 Maps of Snowbird are at the end of this booklet (page 32). 1 COSYNE 2012 Workshops February 27 & 28, 2012 Snowbird, Utah Tuesday, February 28 Organizer(s) Location 1. Understanding heterogeneous cortical activity: S. Ardid Wasatch A the quest for structure and randomness. A. Bernacchia T. Engel 2. Humans, neurons, and machines: how can N. Majaj Wasatch B psychophysics, physiology, and modeling collaborate E. Issa to ask better questions in biological vision J. DiCarlo 3. Inhibitory synaptic plasticity T. Vogels Magpie A H Sprekeler R. Fromeke 4. Functions of identified microcircuits A. Hasenstaub Superior B V. Sohal 5. Promise and peril: genetics approaches for systems K. Nagal Superior A neuroscience revisited. D. Schoppik 6. Perception and decision making in rodents S.
    [Show full text]
  • Minian: an Open-Source Miniscope Analysis Pipeline
    bioRxiv preprint doi: https://doi.org/10.1101/2021.05.03.442492; this version posted May 4, 2021. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under aCC-BY-NC-ND 4.0 International license. Title: Minian: An open-source miniscope analysis pipeline Authors: Zhe Dong1, William Mau1, Yu (Susie) Feng1, Zachary T. Pennington1, Lingxuan Chen1, YosiF Zaki1, Kanaka RaJan1, Tristan Shuman1, Daniel Aharoni*2, Denise J. Cai*1 *Corresponding Authors 1Nash Family Department oF Neuroscience, Icahn School oF Medicine at Mount Sinai 2Department oF Neurology, David GeFFen School oF Medicine, University oF CaliFornia bioRxiv preprint doi: https://doi.org/10.1101/2021.05.03.442492; this version posted May 4, 2021. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under aCC-BY-NC-ND 4.0 International license. Abstract Miniature microscopes have gained considerable traction For in vivo calcium imaging in Freely behaving animals. However, extracting calcium signals From raw videos is a computationally complex problem and remains a bottleneck For many researchers utilizing single-photon in vivo calcium imaging. Despite the existence oF many powerFul analysis packages designed to detect and extract calcium dynamics, most have either key parameters that are hard-coded or insuFFicient step-by-step guidance and validations to help the users choose the best parameters.
    [Show full text]
  • Neuronal Ensembles Wednesday, May 5Th, 2021
    Neuronal Ensembles Wednesday, May 5th, 2021 Hosted by the NeuroTechnology Center at Columbia University and the Norwegian University of Science and Technology in collaboration with the Kavli Foundation Organized by Rafael Yuste and Emre Yaksi US Eastern Time European Time EDT (UTC -4) CEDT (UTC +2) Introduction 09:00 – 09:05 AM 15:00 – 15:05 Rafael Yuste (Kavli Institute for Brain Science) and Emre Yaksi (Kavli Institute for Systems Neuroscience) Introductory Remarks 09:05 – 09:15 AM 15:05 – 15:15 John Hopfield (Princeton University) Metaphor 09:15 – 09:50 AM 15:15 – 15:50 Moshe Abeles (Bar Ilan University) Opening Keynote Session 1 Development (Moderated by Rafael Yuste) 09:50 – 10:20 AM 15:50 – 16:20 Rosa Cossart (Inmed, France) From cortical ensembles to hippocampal assemblies 10:20 – 10:50 AM 16:20 – 16:50 Emre Yaksi (Kavli Institute for Systems Neuroscience) Function and development of habenular circuits in zebrafish brain 10:50 – 11:00 AM 16:50 – 17:00 Break Session 2 Sensory/Motor to Higher Order Computations (Moderated by Emre Yaksi) 11:00 – 11:30 AM 17:00 – 17:30 Edvard Moser (Kavli Institute for Systems Neuroscience) Network dynamics of entorhinal grid cells 11:30 – 12:00 PM 17:30 – 18:00 Gilles Laurent (Max Planck) Non-REM and REM sleep: mechanisms, dynamics, and evolution 12:00 – 12:30 PM 18:00 – 18:30 György Buzsáki (New York University) Neuronal Ensembles: a reader-dependent viewpoint 12:30 – 12:45 PM 18:30 – 18:45 Break US Eastern Time European Time EDT (UTC -4) CEDT (UTC +2) Session 3 Optogenetics (Moderated by Emre Yaksi)
    [Show full text]
  • 11:30Am EST Joshua Gordon, NIMH Director Neural Mechanisms Of
    Time Session Moderator Presentation Title Session Type Session Speakers Dr. Rachel Wilson, Martin Family Professor of Basic 10:30 - 11:30am EST Joshua Gordon, NIMH Director Neural Mechanisms of Navigation Behavior Plenary Keynote Research in the Field of Neurobiology, Harvard Medical School Grace Hwang, Johns Hopkins University Applied How Can Dynamical Systems Neuroscience Konrad P. Kording, Joseph D. Monaco, Kanaka Rajan, Xaq 11:30am - 1:00pm EST Scientific Symposium Physics Laboratory Reciprocally Advance Machine Learning? Pitkow, Brad Pfeiffer, Nathaniel D. Daw Developing and Distributing Novel Electrode Cynthia Chestek, Xie Chong, Tim Harris, Allison Yorita, Eric 11:30am - 1:00pm EST Cynthia Chestek, University of Michigan Scientific Symposium Technologies Yttri, Loren Frank, Rahul Panat Claudia Angeli, Phil Starr, Nanthia Suthana, Michelle 1:30 - 3:00pm EST Greg Worrell, Mayo Clinic Advances in Neurotechnologies for Human Research Scientific Symposium Armenta Salas, Harrison Walker, Winston Chiong Expanding Species Diversity in Neuroscience Cory Miller, Zoe Donaldson, Angeles Salles, Andres 1:30 - 3:00pm EST Zoe Donaldson, University of Colorado Boulder Scientific Symposium Research Bendesky, Paul Katz, Galit Pelled, Karen David, Cynthia Moss John Ngai, Director of NIH BRAIN Initiative AND 15 of our 30 BRAIN Initiative Trainee Travel Awardees will Michelle Jones-London, Chief, Office of Programs Monday, 2020 June 1, 3:30 - 4:30pm EST Trainee Award Highlight Talks Trainee Highlight Talk speak; the names can be found on the Presenter
    [Show full text]
  • How to Study the Neural Mechanisms of Multiple Tasks
    Available online at www.sciencedirect.com ScienceDirect How to study the neural mechanisms of multiple tasks Guangyu Robert Yang, Michael W Cole and Kanaka Rajan Most biological and artificial neural systems are capable of computational work has been devoted to understand completing multiple tasks. However, the neural mechanism by the neural mechanisms behind individual tasks. which multiple tasks are accomplished within the same system is largely unclear. We start by discussing how different tasks Although neural systems are usually studied with one task can be related, and methods to generate large sets of inter- at a time, these systems are usually capable of performing related tasks to study how neural networks and animals many different tasks, and there are many reasons to study perform multiple tasks. We then argue that there are how a neural system can accomplish this. Studying mul- mechanisms that emphasize either specialization or flexibility. tiple tasks can serve as a powerful constraint to both We will review two such neural mechanisms underlying multiple biological and artificial neural systems (Figure 1a). For tasks at the neuronal level (modularity and mixed selectivity), one given task, there are often several alternative models and discuss how different mechanisms can emerge depending that describe existing experimental results similarly well. on training methods in neural networks. The space of potential solutions can be reduced by the requirement of solving multiple tasks. Address Zuckerman Mind Brain Behavior Institute, Columbia University, Center Experiments can uncover neural representation or mech- for Molecular and Behavioral Neuroscience, Rutgers University-Newark, anisms that appear sub-optimal for a single task.
    [Show full text]
  • Arxiv:1907.08549V2 [Q-Bio.NC] 4 Dec 2019 1 Introduction
    Universality and individuality in neural dynamics across large populations of recurrent networks Niru Maheswaranathan∗ Alex H. Williams∗ Matthew D. Golub Google Brain, Google Inc. Stanford University Stanford University Mountain View, CA Stanford, CA Stanford, CA [email protected] [email protected] [email protected] Surya Ganguli David Sussilloy Stanford University and Google Brain Google Brain, Google Inc. Stanford, CA and Mountain View, CA Mountain View, CA [email protected] [email protected] Abstract Task-based modeling with recurrent neural networks (RNNs) has emerged as a popular way to infer the computational function of different brain regions. These models are quantitatively assessed by comparing the low-dimensional neural rep- resentations of the model with the brain, for example using canonical correlation analysis (CCA). However, the nature of the detailed neurobiological inferences one can draw from such efforts remains elusive. For example, to what extent does training neural networks to solve common tasks uniquely determine the network dynamics, independent of modeling architectural choices? Or alternatively, are the learned dynamics highly sensitive to different model choices? Knowing the answer to these questions has strong implications for whether and how we should use task-based RNN modeling to understand brain dynamics. To address these foundational questions, we study populations of thousands of networks, with com- monly used RNN architectures, trained to solve neuroscientifically motivated tasks and characterize
    [Show full text]
  • Spike-Timing-Dependent Ensemble Encoding by Non-Classically
    RESEARCH ARTICLE Spike-timing-dependent ensemble encoding by non-classically responsive cortical neurons Michele N Insanally1,2,3,4,5, Ioana Carcea1,2,3,4,5, Rachel E Field1,2,3,4,5, Chris C Rodgers6,7, Brian DePasquale8, Kanaka Rajan9,10, Michael R DeWeese11,12, Badr F Albanna13, Robert C Froemke1,2,4,5,14* 1Skirball Institute for Biomolecular Medicine, New York University School of Medicine, New York, United States; 2Neuroscience Institute, New York University School of Medicine, New York, United States; 3Department of Otolaryngology, New York University School of Medicine, New York, United States; 4Department of Neuroscience and Physiology, New York University School of Medicine, New York, United States; 5Center for Neural Science, New York University, New York, United States; 6Department of Neuroscience, Columbia University, New York, United States; 7Kavli Institute of Brain Science, Columbia University, New York, United States; 8Princeton Neuroscience Institute, Princeton University, Princeton, United States; 9Department of Neuroscience, Icahn School of Medicine at Mount Sinai, New York, United States; 10Friedman Brain Institute, Icahn School of Medicine at Mount Sinai, New York, United States; 11Helen Wills Neuroscience Institute, University of California, Berkeley, Berkeley, United States; 12Department of Physics, University of California, Berkeley, Berkeley, United States; 13Department of Natural Sciences, Fordham University, New York, United States; 14Howard Hughes Medical Institute, New York University School of Medicine, New York, United States *For correspondence: Abstract Neurons recorded in behaving animals often do not discernibly respond to sensory [email protected] input and are not overtly task-modulated. These non-classically responsive neurons are difficult to interpret and are typically neglected from analysis, confounding attempts to connect neural activity Competing interests: The to perception and behavior.
    [Show full text]
  • Multi-Region 'Network of Networks' Recurrent Neural Network Models of Adaptive and Maladaptive Learning
    Title: Multi-region 'Network of Networks' Recurrent Neural Network Models of Adaptive and Maladaptive Learning PI(s) Grant: Kanaka Rajan Institution(s): Icahn School of Medicine at Mount Sinai Grant Number: 1R01EB028166-01 Title of Grant: Theories, Models, and Methods Abstract Authors: Mathew Perich Aaron Andalman Eugene Carter Other authors in the Deisseroth lab: Burns, V. M., Lovett-Barron, M., Broxton, M., Poole, M., Yang, S. J., Grosenick, L., Lerner, T. N., Chen, R., Benster, T., Mourrain, P., Levoy, M., Karl Deisseroth* Kanaka Rajan* ​ *co-corresponding authors ​ Abstract Text: Based on imaging in over 10,000 neurons across more than 15 regions imaged simultaneously in larval zebrafish during a novel behavioral challenge paradigm, we built and analyzed a recurrent neural network model constrained by the data from the outset. The model focused on activity in the habenula and raphe, identifying a significant and specific change in intra-habenular connectivity as a result of the behavioral challenge paradigm. We are extending this work to a whole-brain level multi-region recurrent neural network model and its accompanying theory of brain-wide control of adaptive to maladaptive state transitions. We consider the complex multi-dimensional interactions in the whole brain and relate them causally to behavior of the intact organism. We therefore present a new, more robust, scalable class of circuit models – multi-region neural network models, with support from a BRAIN Initiative/NIH Theories, Models, and Methods grant (1R01EB028166-01). These models are constrained by experimental data at two levels simultaneously: (a) large-scale neural dynamics – cellular resolution activity imaged from all 15 larval zebrafish brain regions – and (b) the behavior of the organism – tail movements tracked real-time in a range of behavioral states.
    [Show full text]