Derives explicit scaling law for risk in sketched linear contrastive learning w.r.t. sketch dimension M, sample size N, and optimization horizon under paired Gaussian and power-law assumptions.
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.LG 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Sketched Linear Contrastive Learning: Approximation, Optimization, and Statistical Scaling
Derives explicit scaling law for risk in sketched linear contrastive learning w.r.t. sketch dimension M, sample size N, and optimization horizon under paired Gaussian and power-law assumptions.