English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

We revisit on-average algorithmic stability of GD for trainingoverparameterised shallow neural networks and prove new generalisation andexcess risk bounds without the NTK or PL assumptions. In particular, we showoracle type bounds which reveal that the generalisation and excess risk of GDis controlled by an interpolating network with the shortest GD path frominitialisation (in a sense, an interpolating network with the smallest relativenorm). While this was known for kernelised interpolants, our proof appliesdirectly to networks trained by GD without intermediate kernelisation. At thesame time, by relaxing oracle inequalities developed here we recover existingNTK-based risk bounds in a straightforward way, which demonstrates that ouranalysis is tighter. Finally, unlike most of the NTK-based analyses we focus onregression with label noise and show that GD with early stopping is consistent.

Stability & Generalisation of Gradient Descent for Shallow Neural Networks without the Neural Tangent Kernel