← Search

Michael Krämer

2 accepted papers

2025

From Kernels to Features: A Multi-Scale Adaptive Theory of Feature Learning

ICML 2025poster

Feature learning in neural networks is crucial for their expressive power and inductive biases, motivating various theoretical approaches. Some approaches describe network behavior after training through a change in kernel scale from initialization, resulting in a generalization power comparable to…

Cited by 1SourcePDFScholar
2024

Critical feature learning in deep neural networks

ICML 2024poster

A key property of neural networks driving their success is their ability to learn features from data. Understanding feature learning from a theoretical viewpoint is an emerging field with many open questions. In this work we capture finite-width effects with a systematic theory of network kernels in…

Cited by 3SourcePDFScholar