Fall 2016

Stochastic Control via Entropy Compression

Wednesday, Oct. 5, 2016 2:50 pm3:30 pm

Add to Calendar


Calvin Lab Auditorium

We consider an agent trying to bring a system to an acceptable state by repeated probabilistic action (stochastic control). Specifically, in each step the agent observes the flaws present in the current state, selects one of them, and addresses it by probabilistically moving to a new state, one where the addressed flaw is most likely absent, but where one or more new flaws may be present. Several recent works on algorithmizations of the Lov\'{a}sz Local Lemma have established sufficient conditions for such an agent to succeed. Motivated by the paradigm of Partially Observable Markov Decision Processes (POMDPs) we study whether such stochastic control is also possible in a noisy environment, where both the process of state-observation and the process of state-evolution are subject to adversarial perturbation (noise). The introduction of noise causes the tools developed for LLL algorithmization to break down since the key LLL ingredient, the sparsity of the causality (dependence) relationship, no longer holds. To overcome this challenge we develop a new analysis where entropy plays a central role, both to measure the rate at which progress towards an acceptable state is made and the rate at which the noise undoes this progress. The end result is a sufficient condition that allows a smooth tradeoff between the intensity of the noise and the amenability of the system, recovering an asymmetric LLL condition in the noiseless case. To our knowledge, this is the first tractability result for a nontrivial class of POMDPs under stochastic memoryless control.