A new framework, DSO, unifies policy gradients, natural gradients, Hessians, and proximal methods by optimizing an autonomous Markov chain rather than an MDP with actions.
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
-
Dynamical System Optimization
A new framework, DSO, unifies policy gradients, natural gradients, Hessians, and proximal methods by optimizing an autonomous Markov chain rather than an MDP with actions.