The Itakura-Saito loss, derived from Bregman divergence, learns risk-averse value functions that match the exponential-utility Bellman equations and trains more stably than exponential MSE in the tested benchmarks.
A sensitivity formula for risk-sensitive cost and the actor–critic algorithm
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
citation-role summary
background 1
citation-polarity summary
fields
cs.LG 1years
2025 1verdicts
CONDITIONAL 1roles
background 1polarities
background 1representative citing papers
citing papers explorer
-
Risk-Averse Reinforcement Learning with Itakura-Saito Loss
The Itakura-Saito loss, derived from Bregman divergence, learns risk-averse value functions that match the exponential-utility Bellman equations and trains more stably than exponential MSE in the tested benchmarks.