A New Approach of Learning Based on Episodic Memory Model

Abstract

This paper presents a computational model of episodic memory that learns event in response to a continuous sensory input. The proposed model considered event (personal experience) as a collection of coactive activities, where it learns the activities in incremental manner (learns new activity without forgetting the old activities) with the help of fuzzy ART network and learns the event as a unique combination of different category field coactive activities, and also captures the occurred sequence of event in the form of sequence-dependent weights in an episode, which makes it more robust to recall with noisy cue. Also used Hebbian learning to make associations between coactive activities, which helps in pattern completion from the partial and noisy input. To validate the proposed model, an empirical study conducted, where the proposed episodic memory model is evaluated based on the recall accuracy using partial and erroneous cues. The analysis shows that the proposed model significantly associated with encoding and recalling events and episodes even with incomplete and noisy cues, and also our model is found to be more space efficient, and more robust in recalling with noisy cue in comparison with previous ART network-based episodic memory models.