Keynotes | K&Ts | GACs | Talks | Posters | Search

Poster D in Poster Session D: Wednesday, August 5, 2:00 – 3:45 pm, Kimmel Center, Shorin & Rosenthal Rooms

Task context shapes strategy for memory-guided decision in a neural network

Moufan Li¹, Kristopher T Jensen², Qihong Lu³, Marcelo G Mattar¹; ¹New York University, ²University College London, ³City University of Hong Kong

Presenter: Moufan Li

Episodic memory enables flexible decision-making by encoding specific past experiences and retrieving them when needed, yet the computational principles governing how memories are encoded and retrieved across different task contexts remain poorly understood. We developed a neural network model to investigate how task context shapes episodic memory processes. Our model was trained via reinforcement learning on a set of memory and decision tasks. The model learned to strategically encode memories and adaptively control retrieval time when solving novel decision problems. We found that encoding strategies depended on both the breadth of training tasks and whether the task context was known at encoding. Training on a wider range of memory tasks shifted the model toward encoding items as separate memories rather than encoding a summary of the whole sequence. When the task context was unknown at encoding, the model encoded raw stimulus information; when context was known in advance, it selectively encoded and retrieved task-relevant information. These findings provide a mechanistic account of how episodic memory can adaptively balance encoding specificity against retrieval flexibility to support decision-making under varying contextual demands.

Topic Area: Memory, Learning & Knowledge Structures