Seminar on Computational Learning and Adaptation

 From Transient Patterns to Persistent Relational Structures
Episodic Memory formation via Cortico-hippocampal Interactions

Lokendra Shastri
International Computer Science Institute
Berkeley, CA

Abstract:

We readily remember events and situations in our daily lives and acquire memories of specific facts by watching a telecast, reading a newspaper, or participating in a dialogue. This one-shot mnemonic ability poses a challenge for computational neuroscience: How does the brain carry out this remarkable memorization task rapidly and effortlessly?

There is a broad consensus that the hippocampal system (HS), consisting of the hippocampal formation and neighboring cortical areas plays a critical role in the encoding and retrieval of such “episodic” memories. Furthermore, a great deal is known about the anatomy and electrophysiology of the HS, including quantitative data about cell counts and connectivity. But how the HS supports episodic memory function is not well understood.

I will describe a circuit-level computational model (SMRITI*) that demonstrates how a cortically expressed pattern of activity representing an event can be transformed rapidly (~500 msec.) into a robust memory trace by a neural structure analogous to the HS. The model has a large memory capacity (> 50,000 events), displays a high degree of pattern separation, and yet responds strongly to partial cues.

The model suggests that there exists a striking correspondence between the functional requirements of encoding episodic memories and the idiosyncratic architecture and local circuitry of the HS. The model also identifies constraints on the representation of relational knowledge and predicts memory deficits that would result from insult to specific HS subregions and cortical circuits projecting to the HS.
---------------------------------

*System for Memorizing Relational Instances from Transient Impulses

Return to the seminar schedule