Improved No-Regret Algorithms for Stochastic Shortest Path with Linear MDP

TL;DR

This paper introduces two new no-regret algorithms for stochastic shortest path problems with linear MDPs, improving computational efficiency and regret bounds over previous methods, with one achieving horizon-free regret.

Contribution

The paper presents two novel algorithms for linear MDP SSPs, one computationally efficient with improved regret bounds, and another horizon-free with near-optimal regret, advancing the state-of-the-art.

Findings

Efficient algorithm achieves $ ilde{O}( ext{poly}(d,B_{ ext{max}},T_{ ext{hit}}) imes ext{sqrt}(K))$ regret.

Modified algorithm attains logarithmic regret under certain conditions.

Second algorithm achieves horizon-free regret $ ilde{O}(d^{3.5} B_{ ext{max}} imes ext{sqrt}(K))$, nearly matching lower bounds.

Abstract

We introduce two new no-regret algorithms for the stochastic shortest path (SSP) problem with a linear MDP that significantly improve over the only existing results of (Vial et al., 2021). Our first algorithm is computationally efficient and achieves a regret bound $\widetilde{O}\left(\sqrt{d^3B_{\star}^2T_{\star} K}\right)$, where $d$ is the dimension of the feature space, $B_{\star}$ and $T_{\star}$ are upper bounds of the expected costs and hitting time of the optimal policy respectively, and $K$ is the number of episodes. The same algorithm with a slight modification also achieves logarithmic regret of order $O\left(\frac{d^3B_{\star}^4}{c_{\min}^2\text{gap}_{\min}}\ln^5\frac{dB_{\star} K}{c_{\min}} \right)$, where $\text{gap}_{\min}$ is the minimum sub-optimality gap and $c_{\min}$ is the minimum cost over all state-action pairs. Our result is obtained by developing a simpler and improved analysis for the finite-horizon approximation of (Cohen et al., 2021) with a smaller approximation error, which might be of independent interest. On the other hand, using variance-aware confidence sets in a global optimization problem, our second algorithm is computationally inefficient but achieves the first "horizon-free" regret bound $\widetilde{O}(d^{3.5}B_{\star}\sqrt{K})$ with no polynomial dependency on $T_{\star}$ or $1/c_{\min}$, almost matching the $\Omega(dB_{\star}\sqrt{K})$ lower bound from (Min et al., 2021).