Introduces logarithmic smoothing algorithms for sequential off-policy learning that match batch SOTA and outperform prior methods under sequential updates.
Once it is trained, we use an inverse temperature parameter α on its score to interpolate between a uniform policy α = 0 and a trained policy α = 1
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
stat.ML 1years
2025 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Sequential Off-Policy Learning with Logarithmic Smoothing
Introduces logarithmic smoothing algorithms for sequential off-policy learning that match batch SOTA and outperform prior methods under sequential updates.