BSI ranks singular-vector bases for LLM low-rank compression by estimating expected task loss increase via second-order Taylor expansion of the loss and an efficient Hessian-diagonal estimator, outperforming magnitude-based baselines on math reasoning benchmarks.
Block-diagonal Hessian- free Optimization for Training Neural Networks
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Importance-Guided Basis Selection for Low-Rank Decomposition of Large Language Models
BSI ranks singular-vector bases for LLM low-rank compression by estimating expected task loss increase via second-order Taylor expansion of the loss and an efficient Hessian-diagonal estimator, outperforming magnitude-based baselines on math reasoning benchmarks.