Paper reveals how input symmetries can distort AI representational similarity analyses

A new paper published on arXiv details how symmetries in network inputs can confound representational similarity analyses (RSMs). The research demonstrates that these stimulus symmetries can lead to functionally equivalent representations, yet produce different RSMs, reflecting distinct representational geometries. The study also shows that techniques like stochastic gradient descent or energetic regularization can generate sparse, drifting codes, resulting in fluctuating RSMs, and that these phenomena are observable even in networks trained on image data where symmetry is latent. AI

IMPACT Highlights potential pitfalls in analyzing AI model representations, suggesting a need for more robust methods.

RANK_REASON The cluster contains a research paper published on arXiv discussing a technical aspect of AI model analysis. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.LG →

• alphaXiv
• arXiv
• CatalyzeX Code Finder for Papers
• CORE Recommender
• DagsHub
• Gotit.pub
• Hugging Face
• Influence Flower
• Jacob A Zavatone-Veth
• Neurons and Cognition
• Quantitative Biology
• ScienceCast

• paper
• other

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →