We study minimax optimal reinforcement learning in episodic factored Markov decision processes (FMDPs), which are MDPs with conditionally independent transition components.

The machines gain possibly different utilities by processing different jobs, and all jobs assigned to the same machine should be processed without overlap.

Multi-agent interacting systems are prevalent in the world, from purely physical systems to complicated social dynamic systems. In many applications, effective understanding of the situation and accurate trajectory prediction of interactive agents play a significant role in downstream tasks, such as decision making and planning.

In what follows, we comment on certain key aspects of our proposed algorithms.

Setting up a well-designed reward function has been challenging for many reinforcement learning applications.

Here we prove the variational bound on the informativeness loss term (second term in Eq. (4)) that Recall that the speaker's belief states, Therefore, any other decoder would lend an upper bound on the informativeness loss term. In this case, the speaker's belief states are given by While Eq. (A.1) follows from [ Eq. (A.2) is equivalent to assuming that the listener's The main paper is available at https://openreview.net/pdf?id=O5arhQvBdH. Therefore, we treat it here as a discrete set. We therefore aim to bias our agents toward these systems. One way of achieving that is by regularizing the entropy of the speaker's communication vectors.

Policy optimization, i.e. algorithms that learn to make sequential decisions by local search on the agent's policy directly, is a widely used class of algorithms in reinforcement learning [