PFB prunes training samples based on the probability density of their shallow-layer features, blocking pruned samples from deep-layer forward and backward passes, and reports lossless or better accuracy with up to 33% time savings.
Conceptual 12m: Pushing web-scale image-text pre- training to recognize long-tail visual concepts
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CV 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Partial Forward Blocking: A Novel Data Pruning Paradigm for Lossless Training Acceleration
PFB prunes training samples based on the probability density of their shallow-layer features, blocking pruned samples from deep-layer forward and backward passes, and reports lossless or better accuracy with up to 33% time savings.