Generalization Demands Task-Appropriate Modular Neural Architectures

Ruiyi Zhang, Dora Angelaki, New York University, United States; Xaq Pitkow, Baylor College of Medicine, United States

Posters 1 Poster

Cognitive science

Pacific Ballroom H-O

Presentation Time:
Thu, 25 Aug, 19:30 - 21:30 Pacific Time (UTC -8)

Although both use reward-based learning, it is still unclear why deep reinforcement learning (RL) agents, in general, perform worse than the brain in novel out-of-distribution (OOD) tasks. Here, we propose one reason: generalization requires task-appropriate modular neural architectures like the brain; inferior generalization abilities result from using architectures without task-appropriate modules. To verify this hypothesis, we used a spatial navigation task to train RL agents with different neural architectures varying in modularity, then compared their generalization abilities in a novel task presenting novel sensorimotor mappings. We found that, although all agents could master the training task, only those with highly modular architectures separating computations of task variables learned a robust state representation, which supported generalization in the novel task. Our work exemplifies the rationale of the architectural modularity, i.e., supporting generalization.

Creative Commons License
This work is licensed under a Creative Commons Attribution 3.0 Unported License.
2022 Conference on Cognitive Computational Neuroscience
No resources available.
Session P-1
P-1.1: Learning Invariant Object Representations through Local Prediction Error Minimization in a Model of Generative Vision
Matthias Brucklacher, Sander M. Bohte, Jorge F. Mejias, Cyriel M. A. Pennartz, University of Amsterdam, Netherlands
P-1.2: Target similarity effects in Lag 1 Sparing of the Attentional Blink
Emmanuel Lebeau, Ian Charest, Université de Montréal, Canada
P-1.3: Flexible representations of abstract cognitive maps under dynamically changing contexts
Sarah Sweigart, Seongmin Park, Nam Nguyen, Charan Ranganath, Erie Boorman, University of California, Davis, United States
P-1.4: Linking human and artificial neural representations underlying face recognition: Insights from MEG and CNNs
Hamza Abdelhedi, Karim Jerbi, University of Montreal, Canada
P-1.5: Decision-making in dynamic, continuously evolving environments: quantifying the flexibility of human choice
Maria Ruesseler, Lilian Weber, Tom Marshall, Jill O'Reilly, Laurence Hunt, University of Oxford, United Kingdom
P-1.6: Predicting Individual Differences from Brain Responses to Music using Functional Network Centrality
Arihant Jain, Vinoo Alluri, IIIT Hyderabad, India; Elvira Brattico, Aarhus University, Denmark; Petri Toiviainen, University of Jyvaskyla, Finland
P-1.7: Models of confidence to facilitate engaging task designs
Vanessa Ceja, Yussuf Ezzeldine, Megan A. K. Peters, University of California, Irvine, United States
P-1.8: Hierarchical representations of naturalistic social interactions in the lateral visual pathway
Emalie McMahno, Michael Bonner, Leyla Isik, Johns Hopkins University, United States
P-1.9: Net2Brain: A Toolbox to compare artificial vision models with human brain responses
Domenic Bersch, Kshitij Dwivedi, Martina Vilas, Gemma Roig, Goethe Universität, Frankfurt am Main, Germany, Germany; Radoslaw M. Cichy, Freie Universität Berlin, Berlin, Germany, Germany
P-1.10: Efficiency of object recognition networks on an absolute scale
Richard Murray, Devin Kehoe, York University, Canada
P-1.11: Deep Learning Reveals Non-linear Relationships between EEG and fMRI Dynamics
Leandro Jacob, Laura Lewis, Boston University, United States
P-1.12: Exploring the Plasticity-Stability Trade-Off in Spiking Neural Networks
Nicholas Soures, Dhireesha Kudithipudi, University of Texas at San Antonio, United States
P-1.13: Common and distinct changes in brain activation patterns modulated by two different types of prediction errors
Leon Möhring, Jan Gläscher, University Medical Center Hamburg-Eppendorf, Germany
P-1.14: Detecting change points in neural population activity with contrastive metric learning
Carolina Urzay, Nauman Ahad, Mehdi Azabou, Geethika Atmakuri, Eva L. Dyer, Georgia Institute of Technology, United States; Aidan Schneider, Keith B. Hengen, Washington University in St. Louis, United States
P-1.15: Manipulated decoy desirability modulates phantom decoy effect
Luis Alvarez, Daniel Acosta-Kane, Angela Yu, University of California, San Diego, United States
P-1.16: Neuro-Nav: A Library for Neurally-Plausible Reinforcement Learning
Arthur Juliani, Ida Momennejad, Microsoft Research, United States; Samuel Barnett, Princeton University, United States; Brandon Davis, Massachusetts Institute of Technology, United States; Margaret Sereno, University of Oregon, United States
P-1.17: Spiking Neural Networks for Predictive Coding with a Feedforward Gist Pathway
Kwangjun Lee, Jorge Mejias, Cyriel Pennartz, University of Amsterdam, Netherlands; Shirin Dora, Loughborough University, United Kingdom; Sander Bohte, Centrum Wiskunde & Informatica, Netherlands
P-1.18: Trial-by-trial Bayesian integration with attentional switching, rather than non-Bayesian switching heuristics, underlie perceptual estimation
Tamás Kovács, Central European University, Hungary; Máté Lengyel, University of Cambridge; Central European University, Hungary
P-1.19: Understanding Learning Trajectories With Infinite Hidden Markov Models
Sebastian Bruijns, Peter Dayan, Max Planck Institute for Biological Cybernetics, Germany; The International Brain Laboratory, The International Brain Laboratory, Germany
P-1.20: A goal-driven Deep Reinforcement Learning Model Predicts Neural Representations Related to Human Visuomotor Control
JONGHYUK LIM, SUNGBEEN PARK, Sungshin Kim, Hanyang University, Korea (South)
P-1.21: Opportunistic Experiments on a Large-Scale Survey of Diverse Artificial Vision Models in Prediction of 7T Human fMRI Data
Colin Conwell, Jacob Prince, George Alvarez, Talia Konkle, Havard University, United States; Kendrick Kay, University of Minnesota, United States
P-1.22: Distinguishing Neural Time-series Patterns based on Reservoir-derived Error
Muhammad Furqan Afzal, Christian David Marton, Erin L. Rich, Kanaka Rajan, Icahn School of Medicine at Mount Sinai, United States
P-1.23: Isolating Motor Learning Mechanisms in Embodied Virtual Reality
Federico Nardi, Mabel Ziman, Shlomi Haar, A. Aldo Faisal, Imperial College London, United Kingdom
P-1.24: Construal Set Selection and Rigidity in Planning
Mark Ho, Jonathan Cohen, Thomas Griffiths, Princeton Univeresity, United States
P-1.25: Common Encoding Axes for Face Selectivity and Non-face Objects in Macaque Face Cells
Kasper Vinken, Margaret Livingstone, Harvard Medical School, United States; Talia Konkle, Harvard University, United States
P-1.26: Variance-Invariance-Covariance Regularization with Local Self-Supervised Learning Improves Hippocampus Segmentation with Fewer Labels
Kassymzhomart Kunanbayev, Donggon Jang, Jeongwon Lee, Dae-Shik Kim, KAIST, Korea (South)
P-1.27: Accurate implementation of computational neuroscience models through neural ODEs
Sabine Muzellec, CerCO CNRS; Brown University, France; Mathieu Chalvidal, CerCO CNRS; Brown University; ANITI, France; Thomas Serre, Brown University; ANITI, United States; Rufin VanRullen, CerCO CNRS; ANITI, France
P-1.28: Navigation representations during active navigation are predominantly goal-directed
Tianjiao Zhang, Jack Gallant, University of California, Berkeley, United States
P-1.29: Modeling Risk and Reward Expectation and Surprise using Optimal Learning Rates in Human Neuronal Populations to assess Impulsive Choice
Rhiannon Cowan, Tyler Davis, Bornali Kundu, John Rolston, Elliot Smith, University of Utah, United States
P-1.30: A contextual encoding model for human ECoG responses to a spoken narrative
Kristijan Armeni, Christopher Honey, Johns Hopkins University, United States; Tal Linzen, New York University, United States
P-1.32: Social Inference from Relational Visual Information: An Investigation with Graph Neural Network Models
Manasi Malik, Leyla Isik, Johns Hopkins University, United States
P-1.33: “Attentional fingerprints” in conceptual space: Reliable, individuating patterns of visual attention revealed using a natural language model
Amanda J. Haskins, Katherine O. Packard, Caroline E. Robertson, Dartmouth College, United States
P-1.34: Measuring Behavioral Arbitration of the Successor Representation
Ari Kahn, Nathaniel Daw, Princeton University, United States
P-1.35: Modelling inter-animal variability
Javier Sagastuy-Brena, Imran Thobani, Aran Nayebi, Rosa Cao, Dan Yamins, Stanford University, United States
P-1.36: Contextual Influences on the Perception of Motion and Depth
Zhe-Xin Xu, Greg DeAngelis, University of Rochester, United States
P-1.37: Spontaneous Learning of Face Identity in Expression-Trained Deep Nets
Emily Schwartz, Stefano Anzellotti, Boston College, United States; Kathryn O'Nell, Dartmouth College, United States; Rebecca Saxe, Massachusetts Institute of Technology, United States
P-1.38: Distinct prefrontal networks for semantic integration and articulatory planning
Leyao Yu, Nikolai Chapochnikov, Adeen Flinker, New York University, United States
P-1.39: Multivariate Representation of Sustained Visual Content in a No-Report Paradigm
Gal Vishne, Edden M. Gerber, Leon Y. Deouell, Hebrew University of Jerusalem, Israel; Robert T. Knight, University of California Berkeley, United States
P-1.40: Contextual Representation Ensembling
Tyler Tomita, Johns Hopkins University, United States
P-1.42: A population receptive field modeling framework of sensory suppression in human visual cortex
Eline Kupers, Insub Kim, Kalanit Grill-Spector, Stanford University, United States
P-1.43: Approximate Bayesian Inference captures differential effects of value confidence on obligatory and voluntary choices
Joonhwa Kim, Romy Frömer, Xiamin Leng, Amitai Shenhav, Brown University, United States
P-1.44: A Large and Rich EEG Dataset for Modeling Human Visual Object Recognition
Alessandro Gifford, Radoslaw Cichy, Freie Universität Berlin, Germany; Kshitij Dwivedi, Gemma Roig, Goethe Universität Frankfurt, Germany
P-1.45: Predictive Coding in Auditory Cortical Neurons of Songbirds
Srihita Rudraraju, Brad Theilman, Michael Turvey, Timothy Gentner, University of California San Diego, United States
P-1.46: The Cortical Representation of Linguistic Structures at Different Timescales is Shared between Reading and Listening
Catherine Chen, Tom Dupré la Tour, Jack Gallant, Daniel Klein, Fatma Deniz, UC Berkeley, United States
P-1.47: Evidence that noise in human visual cortex encodes naturalistic visual representations
Thomas Naselaris, Thomas Gebhart, Ghislain St-Yves, Kendrick Kay, University of Minnesota, United States
P-1.48: Manipulating and Measuring Variation in DNN Representations
Jason Chow, Thomas Palmeri, Vanderbilt University, United States
P-1.49: Phonemic representation of narrative speech in human cerebral cortex
Xue Gong, Frederic Theunissen, University of California, Berkeley, United States; Alexander Huth, University of Texas, Austin, United States
P-1.50: Unsupervised learning of translucent material appearance using StyleGAN
Chenxi Liao, Bei Xiao, American University, United States; Masataka Sawayama, Inria, France
P-1.51: Generalization Demands Task-Appropriate Modular Neural Architectures
Ruiyi Zhang, Dora Angelaki, New York University, United States; Xaq Pitkow, Baylor College of Medicine, United States
P-1.52: Task-Dependent Incremental Binding Explained by Cortico-Thalamo-Cortical Interactions – A Neuro-Dynamical Model of Mental Contour Tracing
Daniel Schmid, Heiko Neumann, Ulm University, Germany
P-1.53: A Counterfactual Model of Causal Judgments in Double Prevention
Kevin O'Neill, Duke University, United States; Tadeg Quillien, University of Edinburgh, United Kingdom; Paul Henne, Lake Forest College, United States
P-1.54: Informative associations between feature, spatial, and category selectivity in human visual cortex
Margaret M. Henderson, Michael J. Tarr, Leila Wehbe, Carnegie Mellon University, United States
P-1.55: Dissociation Between The Use of Implicit and Explicit Priors in Bayesian Perceptual Inference
Caroline Bévalot, Atomic Energy Commission, National Institute of Health and Medical Research, University Paris-Saclay & Sorbonne, France; Florent Meyniel, Atomic Energy Commission, National Institute of Health and Medical Research, University Paris-Saclay, France
P-1.56: Testing the Effect of Visual Depth on the Perception of Faces in an Online Study
Simon M. Hofmann, Max Planck Institute for Human Cognitive and Brain Sciences, Germany; Abhay Koushik, Université de Paris, France; Felix Klotzsche, Vadim Nikulin, Arno Villringer, Michael Gaebler, Max Planck Institute for Human Cognitive & Brain Sciences, Germany
P-1.57: Humans imperfectly recruit reward systems as they learn to achieve novel goals
Gaia Molinaro, Anne Collins, University of California, Berkeley, United States
P-1.58: A Multivariate Point Process Model for Neural Spike Trains
Reza Ramezan, Meixi Chen, Martin Lysy, Paul Marriott, University of Waterloo, Canada
P-1.59: Goals distort the representation of space
Paul Muhle-Karbe, Hannah Sheahan, Christopher Summerfield, University of Oxford, United Kingdom; Giovanni Pezzulo, National Research Council of Italy, Italy; Hugo Spiers, University College London, United Kingdom; Samson Chien, Nicolas Schuck, Max Planck Institute for Human Development, Germany
P-1.60: Relevance, uncertainty, and expectations affect categorization
Janaki Sheth, Jared Collina, Konrad Kording, Yale Cohen, Maria Geffen, University Of Pennsylvania, United States
P-1.61: Do multimodal neural networks better explain human visual representations than vision-only networks?
Bhavin Choksi, Rufin VanRullen, Leila Reddy, Centre national de la recherche scientifique, France
P-1.62: A heuristic rule explains human perception of predictive structure in naturalistic sequences
Audrey Sederberg, University of Minnesota, United States; Biyu He, NYU Grossman School of Medicine, United States
P-1.63: CogEnv: A Reinforcement Learning Environment for Cognitive Tests
Morteza Ansarinia, Brice Clocher, Aurélien Defossez, Emmanuel Schmück, Pedro Cardoso-Leite, University of Luxembourg, Luxembourg
P-1.64: Looking into the past: Eye-tracking mental simulation in physical inference
Aaron Beller, Scott Linderman, Tobias Gerstenberg, Stanford University, United States; Yingchen Xu, University College London, United Kingdom
P-1.65: Estimates of cognitive processes in decision making from neural signals by an interpretable neural network model
Qinhua Jenny Sun, Khuong Vo, Michael Nunez, Joachim Vandekerckhove, Ramesh Srinivasan, University of California, Irvine, United States
P-1.66: The Role of Agency in Memory for Narratives
Xian Li, Savannah Born, Janice Chen, Johns Hopkins University, United States; Buddhika Bellana, York University, Canada
P-1.67: Model connectivity: overcoming the limitations of functional connectivity by leveraging the power of the voxelwise encoding model framework
Emily Meschke, Matteo Visconti di Oleggio Castello, Jack Gallant, University of California, Berkeley, United States
P-1.68: Bayesian Modeling of Language-Evoked Event-Related Potentials
Davide Turco, Conor Houghton, University of Bristol, United Kingdom
P-1.69: Emergence of Direction Selectivity and Motion Strength in Dot Motion Task Through Deep Reinforcement Learning Networks
Dolton Fernandes, Pramod Kaushik, Raju Bapi, International Institute of Information Technology, Hyderabad, India; Bhargav T Nallapu, Albert Einstein College of Medicine, United States
P-1.70: The medial temporal lobe enables visual perception not possible 'at a glance'
tyler bonnen, Daniel Yamins, Anthony Wagner, Stanford University, United States
P-1.71: Revealing the Feature Dimensions Driving Similarity Judgements of Natural Scenes
Peter Brotherwood, Ian Charest, Université de Montréal, Canada; Andrey Barsky, Jasper Van Den Bosch, University of Birmingham, United Kingdom; Kendrick Kay, University of Minnesota, United States
P-1.72: Optimizing fidelity of uncertainty representation in distributional codes
Mehrdad Salmasi, Maneesh Sahani, University College London, United Kingdom
P-1.73: Using Deep Learning tools for fitting Reinforcement Learning Models
Milena Rmus, Jimmy Xia, Jasmine Collins, Anne Collins, UC Berkeley, United States
P-1.74: Perceptography: using machine learning to peek into the subjective experience
Elia Shahbazi, Timothy Ma, Arash Afraz, National Institutes of Health (NIH), United States