Topology-Aware Revival for Efficient Sparse Training

2026

Read Topology-Aware Revival for Efficient Sparse Training (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2602-04166.txt.

Description

One revival step can rescue a pruned network. Meiling Jin and 4 co-authors propose Topology-Aware Revival for static sparse training in a 2026 paper. Static sparsity fixes a connection mask early, which is efficient but brittle, especially in reinforcement learning where the data distribution keeps shifting.

After pruning, their method spends a small reserve budget across layers by topology need, randomly reactivating a few cut connections per layer. The mask then stays fixed. Across continuous-control tasks, revival improved final returns substantially over static baselines and beat dynamic sparse training too. A single cheap correction outperforms constant rewiring.

Catalog record

YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE42 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2602-04166

Browse the full library · About the Tower of Babel