For mixable and exp-concave losses, a continuous fixed-share exponential-weights method attains dynamic regret of order O(d log T (1 + T^{1/3} P_T^{2/3})), improving the known d^{10/3} factor to d.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Non-stationary Online Learning for Curved Losses: Improved Dynamic Regret via Mixability
For mixable and exp-concave losses, a continuous fixed-share exponential-weights method attains dynamic regret of order O(d log T (1 + T^{1/3} P_T^{2/3})), improving the known d^{10/3} factor to d.