Research map / Optimization / Implicit bias and generalization

Understanding Double Descent through Universal Compression

NeurIPS 2026 · Optimization › Implicit bias and generalization · approach: Edge-of-stability dynamics

Read the paperCheck your idea against 108,967 papers

AI-extracted summary. Written by AI from the paper's abstract only. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Studies the double-descent phenomenon in deep learning through the lens of universal compression, offering an explanation for why test loss peaks near the interpolation threshold and then descends again.

Problem

Double descent, where test loss peaks near the interpolation threshold and then falls again as model size grows, is not well understood for modern deep learning models, despite being explained in toy classical models.

Approach

Frames double descent via universal compression (information-theoretic compressibility of the learned model/data), offering an explanation for modern deep learning beyond toy classical models. Full details unavailable; based only on the abstract.

Experimental setup

Metrics: test loss

Builds on: bias-variance tradeoff

Closest papers by meaning

• Unraveling the Enigma of Double Descent: An In-depth Analysis through the Lens of Learned Feature Space (ICLR 2024 · similarity 0.91)
• DSD²: Can We Dodge Sparse Double Descent and Compress the Neural Network Worry-Free? (AAAI 2024 · similarity 0.89)
• On Double Descent in Reinforcement Learning with LSTD and Random Features (ICLR 2024 · similarity 0.85)
• Unified Scaling Laws for Compressed Representations (NeurIPS 2025 · similarity 0.85)
• Does Compression Imply Generalization? A Minimum Description Length Perspective (NeurIPS 2026 · similarity 0.84)
• Asymmetric Scaling Laws from Sparse Features (NeurIPS 2026 · similarity 0.84)