P-1B.104

The role of object-centric representations, guided attention, and external memory on generalizing visual relations

Guillermo Puebla, National Center for Artificial Intelligence, Chile; Jeffrey Bowers, University of Bristol, United Kingdom

Session:
Posters 1B Poster

Track:
Cognitive science

Location:
Marquee

Presentation Time:
Thu, 24 Aug, 17:00 - 19:00 United Kingdom Time

Abstract:
Visual reasoning is a long-term goal of vision research. In the last decade, several works have attempted to apply deep neural networks (DNNs) to the task of learning visual relations from images, with modest results in terms of the generalization of the relations learned. In recent years, several innovations in DNNs have been developed in order to enable learning abstract relation from images. In this work, we systematically evaluate a series of DNNs that integrate mechanism such as slot attention, recurrently guided attention, and external memory, in the simplest possible visual reasoning task: deciding whether two objects are the same or different. We found that, although some models performed better than others in generalizing the same-different relation to specific types of images, no model was able to generalize this relation across the board. We conclude that abstract visual reasoning remains largely an unresolved challenge for DNNs.

Manuscript:
License:
Creative Commons License
This work is licensed under a Creative Commons Attribution 3.0 Unported License.
DOI:
10.32470/CCN.2023.1499-0
Publication:
2023 Conference on Cognitive Computational Neuroscience
Presentation
Discussion
Resources
No resources available.
Session P-1B
P-1B.61: Cortical dopamine enables deep reinforcement learning and leverages dopaminergic heterogeneity
Jack Lindsey, Ashok Litwin-Kumar, Columbia University, United States
P-1B.62: Illusion of control differentially affects outcome predictions in pathological and recreational gamblers
Frederike H. Petzschner, Brown University, United States; Saee Paliwal, Benevolent AI, United Kingdom; Gina Paolini, University of Zurich and ETH Zurich, Switzerland; Stephanie Olaiya, Chloe Zimmerman, Brown University, United States; Nicole Zahnd, Helen Schmidt, Katharina Wellstein, University of Zurich and ETH Zurich, Switzerland; Ines Bodmer, Franz Eidenbenz, Karinna Schärli, Till Siegrist, Zentrum für Spielsucht und andere Verhaltenssüchte, Switzerland; Klaas Enno Stephan, University of Zurich and ETH Zurich, Switzerland
P-1B.63: Confidence is detection-like in high-dimensional spaces
Wiktoria Luczak, Stephen Fleming, University College London, United Kingdom
P-1B.64: The Role of Mid-Frontal Oscillations in Value-biased Objective Decisions
Alice Vidal, Rubén Moreno-Bote, Salvador Soto-Faraco, universitat Pompeu Fabra, Spain
P-1B.65: Gazed and confused: Gaze reinstatement reveals effects of semantization
Rebekka Heinen, Nikolai Axmacher, Ruhr University Bochum, Germany
P-1B.66: Human prefrontal neurons encode economic risk and risk prediction error
Elliot Smith, Rhiannon Cowan, Tyler Davis, Shervin Rahimpour, university of utah, United States; Matthew Botvinick, deepmind, United Kingdom; Timothy Muller, Steven Kennerley, University College London, United Kingdom; John Rolston, Harvard University, United States
P-1B.67: Valence biases in reinforcement learning and autobiographical memory
Susan Benear, New York University, United States; Michael Evans, First Place for Youth, United States; Gail Rosenbaum, Geisinger Health, United States; Catherine Hartley, New York University, United States
P-1B.68: Spontaneous activity randomly samples its activity space
Jonas Elpelt, Maren Wehrheim, Matthias Kaschube, Frankfurt Institute for Advanced Studies, Germany
P-1B.69: Controllability Inferences Promote Adaptive Action Selection during Adolescence
Noam Goldway, Hillary Raab, New York University, United States; Careen Foord, Weill Cornell Medicine, United States; Catherine Hartley, New York University, United States
P-1B.70: Pre-acquired functional connectivity predicts choice inconsistency
Asaf Madar, Tel Aviv University, Israel; Vered Kurtz-David, New York University, United States; Adam Hakim, Dino Levy, Ido Tavor, Tel Aviv University, Israel
P-1B.71: NeuralPlayground: A Standardised Environment for Evaluating Models of Hippocampus and Entorhinal Cortex
Rodrigo Carrasco-Davis, Clementine Domine, Luke Hollingsworth, Caswell Barry, Andrew Saxe, UCL, United Kingdom
P-1B.72: Retinotopy improves the categorisation and localisation of visual objects in CNNs
Jean-Nicolas JÉRÉMIE, Emmanuel DAUCÉ, Laurent PERRINET, Centre national de la recherche scientifique/Aix-Marseille Université, France
P-1B.73: Connecting hippocampal representations to predictive auxiliary tasks in deep RL
Ching Fang, Columbia University, United States; Kimberly Stachenfeld, DeepMind, United States
P-1B.74: BVCs without borders: reinterpreting boundary vector cells as trajectories in the successor representation
Lauren Bennett, Will de Cothi, Caswell Barry, University College London, United Kingdom
P-1B.75: Computational principles of predictive representation in the human brain
Anna Leshinskaya, Charan Ranganath, Erie Boorman, UC Davis, United States
P-1B.76: A process model for human problem solving in puzzles
Jeroen Olieslagers, Zahy Bnaya, Wei Ji Ma, New York University, United States
P-1B.77: General mechanisms of task engagement in the primate frontal cortex
Jan Grohn, Nima Khalighinejad, Caroline Jahn, Alessandro Bongioanni, Urs Schuffelgen, Jerome Sallet, Matthew Rushworth, Nils Kolling, Oxford, United Kingdom
P-1B.78: Population-Specific MRI Brain Template Construction for Chinese Older Adults
Feng Sang, Yaojing Chen, Zhanjun Zhang, Beijing Normal University, China
P-1B.79: Decoding Hypnotic Experience from Raw EEG using a Multi-Output Auto-Encoder
Yeganeh Farahzadi, Eötvös Loránd University, Hungary; Morteza Ansarinia, University of Luxembourg, Luxembourg; Zoltán Kekecs, Eötvös Loránd University, Hungary
P-1B.80: Goal-conditioned world models: Adaptive computation over multi-granular generative models explains human scene perception
Mario Belledonne, Chloë Geller, Ilker Yildirim, Yale University, United States
P-1B.81: Learning of Cognitive Control during Task Switching in Recurrent Neural Networks
Shengjie Xu, Tom Verguts, Senne Braem, Ghent University, Belgium
P-1B.82: Stimulus expectation shapes somatosensory perception
Carina Forster, Martin Grund, Eleni Panagoulas, Tilman Stephani, Esra Al, Arno Villringer, Max Planck Institute for Human and Cognitive Brain Sciences, Germany
P-1B.83: Striatal Dopamine Reflects Long-term Learning Trajectories
Samuel Liebana Garcia, Aeron Laffere, Chiara Toschi, Louisa Schilling, Peter Zatka-Haas, Rafal Bogacz, University of Oxford, United Kingdom; Andrew Saxe, University College London, United Kingdom; Armin Lak, University of Oxford, United Kingdom
P-1B.84: Distinct roles of human hippocampus and medial frontal cortex in solving credit assignment problem
Ali Mahmoodi, Shuyi Luo, Caroline Harbison, University of Oxford, United Kingdom; Payam Piray, University of Southern California, United States; Matthew Rushworth, University of Oxford, United Kingdom
P-1B.85: Long-Term Credit Assignment in Humans Critically Depends on Sequential Structuring of Events
Sienna Bruinsma, Frederike Petzschner, Matthew Nassar, Brown University, United States
P-1B.86: Keep moving: sensorimotor integration of fixational eye-movements yields human-like superresolution in recurrent neural networks
Adrien Doerig, Osnabrück University, Germany; Kirubeswaran O.R., Indian Institute of Science Education and Research Pune, India; Tim Kietzmann, Osnabrück University, Germany
P-1B.87: The neurocomputational basis of human simple choice when overtrained as done in nonhuman primate paradigms
Zeynep Enkavi, Caltech, United States; Gaia Lombardi, Todd Hare, University of Zurich, Switzerland; Antonio Rangel, Caltech, United States
P-1B.88: Toward computational accounts of reaction times from recurrent neural network models of vision
Lore Goetschalckx, Lakshmi Govindarajan, Alekh Ashok, Thomas Serre, Brown University, United States
P-1B.89: Object-enhanced and Object-centered Representations Across Primate Ventral Visual Cortex
Tahereh Toosi, Nikolaus Kriegeskorte, Elias Issa, Columbia University, United States
P-1B.90: Generalization in an Evidence Accumulation Task
Quan Do, Michael Hasselmo, Benjamin Scott, Boston University, United States
P-1B.91: Animal, Plant, or Mineral: Disentangling Object Concepts from Visual Features
Sophia Shatek, Thomas Carlson, The University of Sydney, Australia
P-1B.92: The Cost of Adjusting Cognitive Control: A Dynamical Systems Approach
Ivan Grahek, Xiamin Leng, Sebastian Musslick, Amitai Shenhav, Brown University, United States
P-1B.93: Understanding spatial neglect: A Bayesian perspective
Tianwei Gong, Bonan Zhao, Robert D. McIntosh, Christopher G. Lucas, University of Edinburgh, United Kingdom
P-1B.94: Cross-modal semantic activation investigated using multivariate approaches applied to MEG data
Yulia Bezsudnova, Andrew Quinn, Syanah Wynn, Ole Jensen, University of Birmingham, United Kingdom
P-1B.95: Successor Representation captures dynamics of structure learning during context-dependent decision-making
Daniel Kimmel, Stefano Fusi, C. Daniel Salzman, Daphna Shohamy, Columbia University, United States
P-1B.96: Using Evidence Pulses and EEG to Determine the Reference Frame for Post-decision Evidence Accumulation
John Grogan, Trinity College Dublin, Ireland; Hannah McDermott, Free University Berlin, Germany; Redmond O'Connell, Trinity College Dublin, Ireland
P-1B.97: Transfer of abstract structural knowledge aids new concept learning in humans and artificial neural networks
Robert Mok, Danyal Akarca, Alexander Anwyl-Irvine, John Duncan, University of Cambridge, United Kingdom; Bradley Love, University College London, United Kingdom
P-1B.98: Adaptive cognitive map: Flexibly scaled representation of abstract concepts in human mPFC
Hai-Tao Wu, Qingtian Mi, Lusha Zhu, Peking University, China
P-1B.99: Dynamic mean-field model of frontal eye-fields performing sequential visual target selection in naturalistic stimuli
Vaishnavi Narayanan, Rainer Goebel, Mario Senden, Maastricht University, Netherlands
P-1B.100: Introducing the MindSet benchmark for comparing DNNs to human vision
Valerio Biscione, Don Yin, Gaurav Malhotra, Marin Dujmović, Milton Montero, Guillermo Puebla, Federioc Adolfi, Christian Tsvetkov, University of Bristol, United Kingdom; Benjamin Evans, Sussex University, United Kingdom; Jeffrey Bowers, University of Bristol, United Kingdom
P-1B.101: People who worry more explore more
Kristin Witte, Max Planck Institute for Biological Cybernetics, Germany; Toby Wise, King's College London, United Kingdom; Quentin Huys, University College London, United Kingdom; Eric Schulz, Max Planck Institute for Biological Cybernetics, Germany
P-1B.102: Developing a Neural Network Model Generating Handwriting Motor Sequences Within Realistic Spatiotemporal Scales and Its Biological Implications
Sungjae Cho, Taegon Kim, Korea Institute of Science and Technology, Korea (South)
P-1B.103: Feedforward Neural Networks can capture Human-like Perceptual and Behavioral Signatures of Contour Integration
Fenil R. Doshi, Talia Konkle, George A. Alvarez, Harvard University, United States
P-1B.104: The role of object-centric representations, guided attention, and external memory on generalizing visual relations
Guillermo Puebla, National Center for Artificial Intelligence, Chile; Jeffrey Bowers, University of Bristol, United Kingdom
P-1B.105: Strategic Control of Episodic Memory Through Post-Gating
Cody Dong, Qihong Lu, Kenneth Norman, Princeton University, United States
P-1B.106: Pre-Training on High-Quality Natural Image Data Reduces DCNN Texture Bias
Niklas Müller, Iris Groen, Steven Scholte, University of Amsterdam, Netherlands
P-1B.107: Early Visual Microcircuits Encoding Internal Models of Future Perceptual Information During Navigation
Yulia Lazarova, Angus Paton, Lucy Petro, Lars Muckli, University of Glasgow, United Kingdom
P-1B.108: Cerebellar-driven cortical dynamics enable cognitive task acquisition and consolidation
Joseph Pemberton, Paul Chadderton, Rui Ponte Costa, University of Bristol, United Kingdom
P-1B.109: Fluctuating responses evoked by multiple objects: a general feature of visual representations
Meredith Schmehl, Duke University, United States; Valeria Caruso, University of Michigan, United States; Shawn Willett, University of Pittsburgh, United States; Yunran Chen, Na Young Jun, Jeff Mohl, Duke University, United States; Douglas Ruff, Marlene Cohen, University of Chicago, United States; Akinori Ebihara, Winrich Freiwald, The Rockefeller University, United States; Surya Tokdar, Jennifer Groh, Duke University, United States
P-1B.110: Environmental motion presented ahead of self-motion modulates the heading direction estimation
Jongmin Moon, Liana Saftari, Oh-Sang Kwon, Ulsan National Institute of Science and Technology, Korea (South)
P-1B.111: The Courtois NeuroMod project: quality assessment of the initial data release (2020)
Julie Boyle, Basile Pinsard, CRIUGM, Canada; Valentina Borghesani, University of Geneva, Switzerland; Francois Paugam, University of Montreal, Canada; Elizabeth DuPre, Standford, United States; Pierre Bellec, University of Montreal, Canada
P-1B.112: Learning Dynamics of Semantic Knowledge in Humans and Neural Networks
Jirko Rubruck, University of Oxford, Oxford, UK, United Kingdom; Andrew Saxe, University College London & CIFAR Azrieli Global Scholars Program, United Kingdom; Christopher Summerfield, University of Oxford, United Kingdom
P-1B.113: The role of subgoals in hierarchical reinforcement learning
Milena Rmus, UC Berkeley, United States; Maria Eckstein, DeepMind, United Kingdom; Anne Collins, UC Berkeley, United States
P-1B.114: Neural Mechanisms of Mental Simulation in Primate Frontal Cortex
Aran Nayebi, Rishi Rajalingham, Mehrdad Jazayeri, Guangyu Yang, Massachusetts Institute of Technology, United States
P-1B.115: Playing to win or playing to learn? Human performance in a social card game task
Alexandra Witt, Joel Vasama, University of Tuebingen, Germany; Natalia Vélez, Harvard University, United States; Charley Wu, University of Tuebingen, Germany
P-1B.116: In dynamic social environments humans adaptively switch between reducing policy and epistemic uncertainty
Amrita Lamba, Michael Frank, Oriel FeldmanHall, Brown University, United States
P-1B.117: Identifying a Shared Source of Age-related Decline in Working Memory and Decision-making
Jade S. Duffy, Trinity College Dublin, Ireland; Hannah McDermott, Freie University, Germany; Robert Whelan, Redmond G. O'Connell, Trinity College Dublin, Ireland; Peter R. Murphy, Maynooth University, Ireland
P-1B.118: Miniature cognitive architectures: Modeling the honey bee's mushroom bodies in delay and trace conditioning
Andrea Alamia, CerCo - CNRS, France; Catherine Macri, CRCA- CNRS, France; Nathan Gaubil, CerCo- CNRS, France; Marco Paoli, Martin Giurfa, CRCA- CNRS, France
P-1B.119: Human Multi-Task Learning: the Why and What
Tobias Ludwig, Max Planck Institute Biological Cybernetics, Dept. of Neural Dynamics and MEG, Hertie Institute for Clinical Brain Research, Centre of Integrative Neuroscience, University of Tübingen, Germany; Markus Siegel, Dept. of Neural Dynamics and MEG, Hertie Institute for Clinical Brain Research, Centre of Integrative Neuroscience, Germany; Eric Schulz, Max Planck Institute Biological Cybernetics, Germany
P-1B.120: Stable, individuating patterns of visual attention in abstract conceptual feature space revealed using natural language model
Amanda J. Haskins, Katherine O. Packard, Caroline E. Robertson, Dartmouth College, United States