Cognitive science states that the human memory system is made up of semantic and episodic memory systems. Semantic memory is concerned with general world knowledge, whereas episodic is concerned with the individual’s personal memory.
Inspired by this observation, a paper recently published on arXiv.org models an agent that has both semantic and episodic memory systems.
The researchers designed and released a challenging environment where an agent must learn how to encode, store and retrieve memories in order to maximize rewards. It has been shown that an agent with both memory systems answers questions more successfully than those using only one of the two memory systems.
It has also been shown that when an agent is trained with common sense, he or she outperforms someone who is not pre-trained. Furthermore, it has been demonstrated that when an agent cooperates with another agent or a human, it performs better.
Inspired by cognitive science theory, we explicitly model an agent with both semantic and episodic memory systems, and show that it is better than having just one of the two memory systems. To show this, we have designed and released our own challenging environment, “Room”, compatible with OpenAI Gym, where an agent must learn to properly encode, store and retrieve memories in order to maximize their rewards . The room environment allows for a hybrid intelligence setup where machines and humans can collaborate. We show that two agents cooperating with each other perform better than a single agent acting alone. We have open-sourced our code and model at this https URL.
Research Paper: Kim, T., Cochez, M., Francois-Lavet, V., Neerincx, M., and Vossen, P., “A Machine with Human-Like Memory Systems”, 2022. Link: https://arxiv.org /abs/2204.01611