CCN 2026 | Tutorial: Adaptive, degenerate, and yet comparable?

Value of the Information & Strength of the Argument

The tutorial provides a rigorous and accessible introduction to the theoretical underpinnings of representational comparisons. The value lies in its clear demonstration, through simple linear networks, of the dissociation between function and representation, which is a fundamental challenge for the field. The argumentation is solid, building from basic multiplication to matrix transformations and Gram matrices, and is supported by interactive visualizations that allow participants to manipulate parameters and observe the effects. The presenters effectively convey the counterintuitive idea that representational geometry is not uniquely determined by function, and they provide a theoretical framework for understanding when representational comparisons are meaningful. The tutorial successfully bridges theory and practice, offering a compelling case for studying representational plurality.

Scientific Rigor, Source Quality, Title Accuracy

The tutorial is based on a peer-reviewed paper (referenced in the description) and presents original theoretical results. The sources are of high quality, and the content is consistent with current literature in computational neuroscience and machine learning. The title accurately reflects the content, and the tutorial is well-structured. The presenters do not cite external sources during the talk, but the paper provides a solid foundation. The interactive figures are well-designed and enhance understanding. Overall, the scientific rigor is high, and the tutorial is a valuable resource for researchers in the field.

The title accurately reflects the content: the tutorial addresses the challenges of representational comparisons in adaptive and degenerate neural networks, and the subtitle 'Rethinking representational comparisons' is well matched.

8/10

The tutorial is grounded in a peer-reviewed paper and presents rigorous mathematical derivations and interactive demonstrations. The content is consistent with current theoretical neuroscience and machine learning literature. The presentation is clear and the interactive figures enhance understanding. Minor limitations include the lack of direct citations to external sources during the talk and the reliance on simplified linear models, but the theoretical framework is solid.

Markers derived by PSI from the transcript: the creator did not define chapters.

• 00:00Introduction and overview of the tutorial structure
• 05:00Introduction to linear networks and the concept of function vs. representation
• 10:00Demonstration of parameter symmetries and equivalent sets in weight space
• 15:00Interactive figure showing dissociation of function and representation
• 20:00Extension to matrix transformations and representational geometry
• 25:00Introduction of Gram matrix and its dependence on parameter transformations
• 30:00Discussion of assumptions underlying representational comparisons
• 35:00Transition to nonlinear networks and deeper implications
• 40:00Case study: simulating a human experiment to identify representational plurality
• 45:00Conclusion and take-home messages on representational comparisons

• CCN 2026 Tutorial: Adaptive, degenerate, and yet comparable? — Official tutorial page with materials and references.

• Not all solutions are created equal: analytical dissociation of functional and representational similarity in deep linear neural networks — The paper referenced in the tutorial, providing the theoretical basis for the presented results.