Goal directed adaptive behavior in second-order neural networks: The MAXSON family of architectures

The paper presents a neural network architecture (MAXSON) based on second-o rder connections that can learn a multiple goal approach/avoid task using r einforcement from the environment. It also enables an agent to learn vicari ously, from the successes and failures of other agents. The paper shows tha t MAXSON can learn certain spatial navigation tasks much faster than tradit ional Q-learning, as well as learn goal directed behavior, increasing the a gent's chances of long-term survival. The paper shows that an extension of MAXSON (V-MAXSON) enables agents to learn vicariously, and this improves th e overall survivability of the agent population.