AI RESEARCH
Quotient Geometry, Effective Curvature, and Implicit Bias in Simple Shallow Neural Networks
arXiv CS.AI
•
ArXi:2603.21502v1 Announce Type: cross Overparameterized shallow neural networks admit substantial parameter redundancy: distinct parameter vectors may represent the same predictor due to hidden-unit permutations, rescalings, and related symmetries. As a result, geometric quantities computed directly in the ambient Euclidean parameter space can reflect artifacts of representation rather than intrinsic properties of the predictor.