A compression statistic's observation leaves a fiber of compatible endpoint-risk tables, so only orders fixed across that fiber are identified, and stable scores can still select worse group-robust models.
Zico Kolter , title =
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.AI 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
When Compression Scores Cannot Decide: Information Boundaries for Group-Robust LLM Pruning
A compression statistic's observation leaves a fiber of compatible endpoint-risk tables, so only orders fixed across that fiber are identified, and stable scores can still select worse group-robust models.