| Paper: | PS-2B.43 | ||
| Session: | Poster Session 2B | ||
| Location: | H Fläche 1.OG | ||
| Session Time: | Sunday, September 15, 17:15 - 20:15 | ||
| Presentation Time: | Sunday, September 15, 17:15 - 20:15 | ||
| Presentation: | Poster | ||
| Publication: | 2019 Conference on Cognitive Computational Neuroscience, 13-16 September 2019, Berlin, Germany | ||
| Paper Title: | Optimal Timing for Episodic Retrieval and Encoding for Event Understanding | ||
| Manuscript: | Click here to view manuscript | ||
| License: |  This work is licensed under a Creative Commons Attribution 3.0 Unported License. |
||
| DOI: | https://doi.org/10.32470/CCN.2019.1072-0 | ||
| Authors: | Qihong Lu, Zi Ying Fan, Uri Hasson, Kenneth Norman, Princeton University, United States | ||
| Abstract: | When should an intelligent agent encode and retrieve episodic memories? In this work, we use a memory-augmented neural network to study how episodic memory can be most effectively deployed in the service of event understanding. Events are generated from underlying situation models and situations sometimes re-occur, making it useful to have an episodic memory system that can store and retrieve these situation models. For retrieval, our model learned to wait adaptively to accumulate information to ensure accurate retrieval of the target memory. Additionally, model variants that stored episodic memories at event boundaries (but not mid-event) had better subsequent recall performance. This latter result provides a normative explanation of the finding (from human fMRI) that the hippocampus is differentially engaged at event boundaries. | ||