← Back to feed

Escaping Parameter Space: Tight Generalization Bounds via Representation Quality

L5 · ResearcherResearcharXiv· 10/16/2026

Advances theoretical understanding of neural network generalization with practical diagnostic tools for ML researchers.

AI Summary

Researchers derive tight generalization bounds for neural networks using representation quality metrics instead of traditional parameter-space complexity measures, achieving non-vacuous certificates on CIFAR-100 for the first time.

Excerpt

We derive a tight generalization bound for deterministic neural We derive a tight generalization bound for deterministic neural networks based on their last hidden layer representations. By evaluating representation quality with a parameter-free $k$-nearest-neighbor ($k$-NN) classifier, our certificate avoids traditional parameter-space complexity measures, KL divergences to a posterior, weight norms, and compression. The bound is tight: trained from scratch, it reaches an 8.0 percentage point g

Read Original
0 upvotes · 0 downvotes · 1 min read

Related Articles

L5 · ResearcherResearcharXiv
Self-Building Labs

Researchers introduce self-building laboratories that autonomously design, configure, and operate experimental setups from high-level scientific goals.

L5 · ResearcherResearcharXiv
AI-Driven Reproducibility and Analytical Robustness in Scientific Discovery

Researchers developed LLM agents that automate scientific reproducibility checks, successfully reproducing 77.4% of claims in medical studies and flagging 9.5% for expert review.

L5 · ResearcherResearcharXiv
DiscoveryRSI: Recursively Learning to Discover Valuable Hypotheses

Researchers introduce DiscoveryRSI, a framework that recursively learns to generate valuable scientific hypotheses through multi-agent evolutionary search and self-improvement mechanisms.

L5 · ResearcherResearcharXiv
ENFR: Evidence Networks with Flow-Matched References for Amortized Simulation-Based Model Inference

Researchers introduce ENFR, a novel method using flow-matched references for amortized evidence estimation in simulation-based Bayesian model inference.

L5 · ResearcherResearcharXiv
Learning without relearning physical symmetry: an SO(3)-equivariant convolutional network (CNN) on the sphere

Researchers developed an SO(3)-equivariant CNN for spherical data that preserves physical symmetries, reducing simulation requirements and improving accuracy in cosmology applications.