2023 GAC 2

Comparing artificial and biological networks: are we limited by tools, hypotheses or data?

Organizers & Speakers at CCN 2023: 

Meenakshi Khosla, Massachusetts Institute of Technology

Apurva Ratan Murty, Massachusetts Institute of Technology

Tal Golan, Ben-Gurion University of the Negev

Jenelle Feather, Flatiron Institute

Katherine Hermann, Google Research

Aran Nayebi, Massachusetts Institute of Technology

Jessica Thompson, Oxford University

Alex Williams, New York University

Rosa Cao, Stanford University

Radoslaw Martin Cichy, Freie Universitat Berlin

Bradley Love, University College London

Kalanit Grill-Spector, Stanford University

SueYeon Chung, New York University

Other contributors: 

James DiCarlo, Massachusetts Institute of Technology

Abstract:

Deep neural networks have emerged as powerful models of computations underlying sensory and high-level cognitive processing. This synergistic study of computational models (AI) and neuroscience with comparative analyses is now widely used to understand the fundamental processes of cognition. Tremendous progress towards this goal has been fueled by major dataset collection and benchmark efforts. While this progress is exciting, recent developments suggest that we might be hitting a wall with our current hypotheses, tools and data for model differentiation. In modern DNNs, seemingly wide disparities in architectural mechanisms have little consequence on alignment with biological networks based on our current comparative analyses tools and data. Moreover, current models and tools cannot help distinguish the computational logic of different brain areas within a domain. This raises the question: Is this apparent model equivalence a scientifically important result, or is something amiss in terms of our current hypotheses, tools or data used for model differentiation? Our main goal in this GAC is to identify the major challenges that have hindered adjudication between competing representational models and to discuss potential solutions for making progress. As a community, we aim to agree on the tools and data that should be employed for comparative analyses. 

>>> the workshop recording can be found here <<<

Instructions for CCN community members: 

1. Download the proposal PDF using the button above. 

2. Use the Disqus box below to leave comments and reviews, and discuss the submission. You'll need to create a Disqus account or log into Disqus using a google account or similar in order to leave a comment/review.