PROJECT TITLE :

SR-EM: Episodic Memory Aware of Semantic Relations Based on Hierarchical Clustering Resonance Network

ABSTRACT:

In order to provide an appropriate level of service to a user, an intelligent robot needs to have episodic memory. This type of memory allows the robot to retrieve a series of events for a service task that it has learned from previous experiences. On the basis of adaptive resonance theory (ART) networks, a variety of episodic memories have been designed. These episodic memories can learn new tasks incrementally without forgetting the tasks that they have already learned in the past. The traditional ART-based episodic memories, on the other hand, lack the ability to adapt to different environments. They are unable to make adaptive use of the retrieved task episode within the context of the working environment. In addition, a user must repeatedly issue the same command in order to receive multiple instances of the same type of service within the same context. In this article, a novel hierarchical clustering resonance network (HCRN) is proposed as a solution to these limitations. This network has a high clustering performance on multimodal data and is able to compute the semantic relations that exist between learned clusters. In order to intelligently carry out the task at hand, a semantic relation-aware episodic memory (SR-EM) has been designed with the help of HCRN. This memory can modify the retrieved task episode in accordance with the current working environment. The results of experimental simulations show that HCRN performs significantly better than the traditional ART when it comes to clustering performance on multimodal data. In addition, the viability of the suggested SR-EM is validated by running robot simulations for both of the aforementioned scenarios.