Projection-free linear TD(0) achieves a robust O~(1/√T) convergence rate under Markovian noise; this is the first such guarantee without projections or curvature assumptions.
Finite sample analysis of two-timescale stochastic approximation with applications to reinforcement learning
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
-
A Robust $\widetilde{\mathcal{O}}(1/\sqrt{T})$ Rate for Unprojected TD Learning with Linear Function Approximation
Projection-free linear TD(0) achieves a robust O~(1/√T) convergence rate under Markovian noise; this is the first such guarantee without projections or curvature assumptions.