Trainable Weight Averaging: Efficient Training by Optimizing Historical Solutions

Tao Li, Zhehao Huang, Qinghua Tao, Yingwen Wu, X. Huang
2/12/2026

No official code found. Best implementation: nblt/TWA (28 stars, MIT, last commit 2025-02-11).

Semantic Scholar

Code Implementations(1)

nblt/TWACommunity63%
MIT

[ICLR 2023] Trainable Weight Averaging: Efficient Training by Optimizing Historical Solutions

283Shell, PythonMay 17, 20221 years agoMIT

Similar Papers

MetricEmbedding: Accelerate Metric Nearness by Tropical Inner Product

100% match2026

Red PANDA: Disambiguating Image Anomaly Detection by Removing Nuisance Factors

99% match2026

Looking Inward: Language Models Can Learn About Themselves by Introspection

99% match2026

AutoOS: Make Your OS More Powerful by Exploiting Large Language Models

99% match1 repo142026

Nonlinear Sequence Embedding by Monotone Variational Inequality

97% match1 repo52026

Improving Flow Matching by Aligning Flow Divergence

97% match1 repo82026

Cite this paper

@article{li2026trainable,
  title  = {Trainable Weight Averaging: Efficient Training by Optimizing Historical Solutions},
  author = {Tao Li and Zhehao Huang and Qinghua Tao and Yingwen Wu and X. Huang},
  year   = {2026},
  url    = {https://api.semanticscholar.org/CorpusID:035113896965c71a2ae46d770055d0fa2db448a9},
  journal = {ICLR 2023 2023}
}

Add a badge to your README

Shows whether official code is linked for this paper.

Code of Paper badge
[![Code of Paper](https://codeofpaper.com/badge/035113896965c71a2ae46d770055d0fa2db448a9.svg)](https://codeofpaper.com/paper/035113896965c71a2ae46d770055d0fa2db448a9)

Discussion