Modified Error Bounds for Matrix Completion and Application to RL

In matrix completion under noisy measurements, most available results assume that there is an a priori bound on the Frobenius norm of noise, and derive bounds residual error. this letter, we obtain “component-wise” error, based a similar noise. This among first such results. As in our earlier paper, choose locations samples to correspond edge set Ramanujan bigraph. One recent application deriving nearly optimal action-value functions Reinforcement Learning (RL), which illustrated through examples. The presented here are only sufficient conditions. Then illustrate numerical simulations considerable room for improvement conditions derived here. problem future research.