Meenakshi Khosla, UCSD
Nancy Kanwisher, MIT
Stefano Fusi, Columbia
Nikolaus Kriegeskorte, Columbia
Tatiana Engel, Princeton
Marcus Benna, UCSD
David Klindt, CSHL
Carsen Stringer, HHMI Janeli
Short description
How do brains and artificial networks represent information? The structure of the neural code has been described using diverse frameworks, often creating confusion about their meanings and relationships. Key perspectives include interpretable tuning or category selectivity (vs mixed selectivity), systematic tuning, modularity, clustering, disentangled coding, sparse coding and
distributed coding (vs localized coding). While some of these concepts emphasize structured, interpretable, or systematic organization, others focus on population-level representations that defy simple partitioning. Single terms (like “modularity”) have been used to refer to multiple fundamentally different concepts, and multiple terms have been used to refer to identical or
overlapping ideas. This conceptual ambiguity hinders cross-study comparisons, synthesis of theories, and the design of experiments based on a common vocabulary. To add to this confusion, it is also unclear how these concepts interact with or shape the geometry and dimensionality of the activity space.
We propose a collaborative initiative that aims to clarify these perspectives, develop a consensus terminology, agree on what would count as evidence for each kind of coding scheme, agree on at least one specific data set that embodies each of the coding schemes (including what aspect of those data reveal that coding scheme), understand the conditions under which different forms of coding can coexist or conflict, understand how they relate to underlying properties of the neural code such as its dimensionality and geometry, and seek underlying computational principles that explain why brains and artificial networks code information the way they do.
Schedule of Events
4:00 pm - 5:00 pm Introduction into the debate (15 mins)
5:00 pm - 6:30 pm ` Main discussion
[Room A1.02]
5:00 – 5:05 — Welcome & Goals
Format overview (debates + panel), mention live polling.
5:05 – 5:10 — Key Definitions of topics relevant to debates
Mixed selectivity, disentangled coding, geometry, tuning, coordinate system, variables etc.
5:10 – 5:25 — Debate 1: Geometry is all you need vs. Tuning matters
6 min per side, 3 min audience Q&A + poll. (Pro-geometry: Stefano, David, Pro-tuning: Nancy, Meenakshi)
5:25 – 5:40 — Debate 2: Is there a “right” way to choose variables?
Short statements (Tatiana, David, Niko, Marcus), 3 min Q&A + poll.
5:40 – 5:45 — Third Issue (Carsen): Are classical coding schemes outdated for primary sensory areas?
Brief statement + poll.
5:45 – 6:25 — Panel Discussion & Audience Qs
All speakers respond to themes and poll results.
6:25 – 6:30 — Closing
Key takeaways + next steps.
Meenakshi Khosla
Nancy Kanwisher
Stefano Fusi
Nikolaus Kriegeskorte
Tatiana Engel
Marcus Benna
David Klindt
Carsen Stringer
Download the proposal PDF using the button above.
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.