000			01935nam a22002417a 4500
008			230803b |||||||| |||| 00| 0 eng d
041			_aen
082			_a600 _bNAM
100			_aSaxena, Naman
245			_aAverage reward actor-critic with deterministic policy search
260			_aBangalore : _bIISc , _c2023 .
300			_aviii, 143p. _bcol. ill. ; _c29.1 cm * 20.5 cm _ee-Thesis _g3.477Mb
500			_ainclude bibliographic reference and index
502			_aMTech (Res); 2023; Computer science and automation
520			_aThe average reward criterion is relatively less studied as most existing works in the Reinforcement Learning literature consider the discounted reward criterion. There are few recent works that present on-policy average reward actor-critic algorithms, but average reward off-policy actor-critic is relatively less explored. In this work, we present both on-policy and off-policy deterministic policy gradient theorems for the average reward performance criterion. Using these theorems, we also present an Average Reward Off-Policy Deep Deterministic Policy Gradient (ARO-DDPG) Algorithm. We first show asymptotic convergence analysis using the ODE-based method. Subsequently, we provide a finite time analysis of the resulting stochastic approximation scheme with linear function approximator and obtain an $\epsilon$-optimal stationary policy with a sample complexity of $\Omega(\epsilon^{-2.5})$. We compare the average reward performance of our proposed ARO-DDPG algorithm and observe better empirical performance compared to state-of-the-art on-policy average reward actor-critic algorithms over MuJoCo-based environments.
650			_aReinforcement Learning
650			_aActor-Critic Algorithm
650			_aStochastic Approximation
700			_aKolathaya, Shishir N Y advised
700			_aBhatnagar, Shalabh advised
856			_uhttps://etd.iisc.ac.in/handle/2005/6175
942			_cT
999			_c429608 _d429608