Same-accuracy pruned vision-language prefixes can retain very different amounts of traceable OCR support (PosECR 0.620 vs 0.270 for Target vs Random at 30% budget), so pruning evaluation should report spatial provenance and cost.
Advances in Neural Information Processing Systems , volume =
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CV 1years
2026 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Beyond Accuracy: Auditing Spatial Provenance in Visual Token Pruning for OCR-Critical MLLM Inference
Same-accuracy pruned vision-language prefixes can retain very different amounts of traceable OCR support (PosECR 0.620 vs 0.270 for Target vs Random at 30% budget), so pruning evaluation should report spatial provenance and cost.