In value-based reinforcement learning methods such as deep Q-learning, function approximation errors are known to lead to overestimated value estimates and suboptimal policies. We show that this problem persists in an actor-critic setting and propose novel mechanisms to minimize its effects on both the actor and critic. Our algorithm takes the minimum value between a pair of critics to restrict...

This paper presents a novel incremental algorithm that combines Q-learning, a well-known dynamic programming-based reinforcement learning method, with the TD() return estimation process, which is typically used in actor-critic learning, another well-known dynamic programming-based reinforcement learning method. The parameter is used to distribute credit throughout sequences of actions, leading ...

A knowledge based segmentation critic algorithm to enhance recognition of courtesy amounts on bank checks is proposed in this paper. This algorithm extracts the context from the handwritten material and uses a syntax parser based on a deterministic finite automaton to provide adequate feedback to enhance recognition. The segmentation critic presented is capable of handling a number of commonly ...

This paper reviews efforts to design an environment for authoring group performance support system (GPSS) agents that can interact with remote internet resources, applications, and users. We review the architecture of the GPSS environment and show how its tutor/critic elements do and don’t map into an idealized specification extracted from the agent oriented programming and inter-agent communic...