consolidation and memory prior knowledge · 4/4/2019  · perirhinal cortex, and postrhinal cortex...

17
Prior Knowledge and Memory Consolidation Expanding Competitive Trace Theory Anna Smith

Upload: others

Post on 17-Jul-2020

4 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Consolidation and Memory Prior Knowledge · 4/4/2019  · perirhinal cortex, and postrhinal cortex ... receives input from the cortex and projects to CA3 sparse connectivity (granule

Prior Knowledge and Memory ConsolidationExpanding Competitive Trace Theory

Anna Smith

Page 2: Consolidation and Memory Prior Knowledge · 4/4/2019  · perirhinal cortex, and postrhinal cortex ... receives input from the cortex and projects to CA3 sparse connectivity (granule

1. Background in Memory Models2. Models of Consolidation 3. The Hippocampus4. Competitive Trace Theory5. My Project

Outline

Page 3: Consolidation and Memory Prior Knowledge · 4/4/2019  · perirhinal cortex, and postrhinal cortex ... receives input from the cortex and projects to CA3 sparse connectivity (granule

Memory is fundamental

inextricable, descriptive, predictive, hopefully cohesive

1

Page 4: Consolidation and Memory Prior Knowledge · 4/4/2019  · perirhinal cortex, and postrhinal cortex ... receives input from the cortex and projects to CA3 sparse connectivity (granule

Page 5: Consolidation and Memory Prior Knowledge · 4/4/2019  · perirhinal cortex, and postrhinal cortex ... receives input from the cortex and projects to CA3 sparse connectivity (granule

Search of Associative Memory

Associative network model

“Memory images” are either active or inactive

Can activate others based on weight of connections

Page 6: Consolidation and Memory Prior Knowledge · 4/4/2019  · perirhinal cortex, and postrhinal cortex ... receives input from the cortex and projects to CA3 sparse connectivity (granule

Substrate doesn’t matter?

2

April 6, 2017

Page 7: Consolidation and Memory Prior Knowledge · 4/4/2019  · perirhinal cortex, and postrhinal cortex ... receives input from the cortex and projects to CA3 sparse connectivity (granule

The Hippocampus

Bilateral limbic structures in the medial temporal lobe

Patient HM: damage to MTL → lost consolidation and working memory, but retained some LTM

Hippocampal Formation (HF): Dentate Gyrus (DG), CA1-CA2-CA3 complex, and subiculum

Parahippocampal Gyrus: presubiculum, entorhinal cortex (EC), perirhinal cortex, and postrhinal cortex

3

Page 8: Consolidation and Memory Prior Knowledge · 4/4/2019  · perirhinal cortex, and postrhinal cortex ... receives input from the cortex and projects to CA3 sparse connectivity (granule

Connectivity

EC: cortical input/output structure

◉ Perirhinal Ctx → EC→ HF : codes for spatial information (place cells)

◉ Postrhinal Ctx → EC → HF: codes for nonspatial

DG:

◉ receives input from the cortex and projects to CA3◉ sparse connectivity (granule cells)

CA3:

◉ Full of recurrent connections, forming its own network

CA1:

◉ Fewer neurons relative to CA3◉ Relays information from CA3 back to afferent regions in the

cortex via the EC

Page 9: Consolidation and Memory Prior Knowledge · 4/4/2019  · perirhinal cortex, and postrhinal cortex ... receives input from the cortex and projects to CA3 sparse connectivity (granule

Pattern Separation and Pattern Completion

DG: Pattern Separation

packages input and retains segregation

◉ Sensory input arrives from EC, dispersed onto granule cells

→ Redundancy reduced

◉ Output becomes orthogonalized (non-overlapping)

CA3: Pattern Completion

takes orthogonal input and compares it to stored patterns

◉ Partial activation of a strong CA3 network can activate the rest of the network

◉ Impoverished stimuli gravitate towards stable states (attractors)

CA1: strong enough CA3 representations elicit response in corresponding CA1 network

◉ More neurons relative to CA3 → creates redundancy and robustness (less sensitive to noise)

◉ Relays “memory traces” back to afferent regions

Trade-off between gist (pattern-completed) and detail (pattern-separated) memory

How?

Page 10: Consolidation and Memory Prior Knowledge · 4/4/2019  · perirhinal cortex, and postrhinal cortex ... receives input from the cortex and projects to CA3 sparse connectivity (granule

SMSCStandard Model of Systems Consolidation

◉ Hippocampus only necessary in initial transfer of information

◉ MTL no longer needed after “migration to the cortex”

Models of Consolidation

MTTMultiple Trace Theory

◉ Argues for continued interaction between cortex and MTL in encoding and retrieval

◉ Each instance of recalling or experiencing a stimulus creates a new memory trace

Role of hippocampus in resulting cortical memory traces?

Page 11: Consolidation and Memory Prior Knowledge · 4/4/2019  · perirhinal cortex, and postrhinal cortex ... receives input from the cortex and projects to CA3 sparse connectivity (granule

Substrate doesn’t matter?

2

April 6, 2017

Page 12: Consolidation and Memory Prior Knowledge · 4/4/2019  · perirhinal cortex, and postrhinal cortex ... receives input from the cortex and projects to CA3 sparse connectivity (granule

Competitive Trace Theory

4

◉ Hippocampus is an index◉ Unifies input, not the site of

memory storage◉ 2 processes after binding a

memory trace:○ Consolidation○ Decontextualization

◉ Orthogonal traces cause anti-Hebbian competition

→ more exposures → more gist-based

Reagh and Yassa (2014)Hypothesis: repeated exposures to a stimulus will be detrimental to detail memory

Findings: Confirmed hypothesis; subjects more likely to confuse similar lures and targets when stimuli are repeated three times vs once

Page 13: Consolidation and Memory Prior Knowledge · 4/4/2019  · perirhinal cortex, and postrhinal cortex ... receives input from the cortex and projects to CA3 sparse connectivity (granule

Finally… my comps:

The Effect of Cortical Schemas on Memory Trace Competition

5

A follow-up of Reagh and Yassa (2014) that seeks to expand CTT to account memory traces of novel stimuli.

Schemas: related cortical representations in an associative network◉ Allows for systematic search

during retrieval◉ Assists in consolidation;

allows for alteration of synaptic weights

Hypothesis: encoding nonschematic stimuli is more cognitively laborious, and recognition will take more exposures

butI expect to see a competitive effect after a threshold number of repetitions

Page 14: Consolidation and Memory Prior Knowledge · 4/4/2019  · perirhinal cortex, and postrhinal cortex ... receives input from the cortex and projects to CA3 sparse connectivity (granule

The TaskStudy Phase

60 distinct itemsPress a, s, or d for density rating

Test Phase

79 test items: 30 targets, 30 lures, 19 foils. Counterbalanced.

Target Bin 1 Lure(most difficult)

Bin 2 Lure(less difficult)

Novel Foil

Repeated 1x, 3x, or 6x

Page 15: Consolidation and Memory Prior Knowledge · 4/4/2019  · perirhinal cortex, and postrhinal cortex ... receives input from the cortex and projects to CA3 sparse connectivity (granule

The Task

Page 16: Consolidation and Memory Prior Knowledge · 4/4/2019  · perirhinal cortex, and postrhinal cortex ... receives input from the cortex and projects to CA3 sparse connectivity (granule

◉ Discrimination (correct lure rejections) strongest for 3-repetition, weak in 1 and 6

◉ Recognition (correct target acceptances) strongest for 6-repetition

◉ Both facilitated in Bin 2 lures

Predictions

schemas used in this task are weakly bound

Cortical inputs enter through EC → pattern separation in DG → pattern completion in CA3 → mental representation

Celtic knots don’t gravitate towards any strong attractor, so perhaps more detail is preserved compared to Reagh and Yassa (2014)

but orthogonalization still causes trace competition

Page 17: Consolidation and Memory Prior Knowledge · 4/4/2019  · perirhinal cortex, and postrhinal cortex ... receives input from the cortex and projects to CA3 sparse connectivity (granule

Acknowledgements

Thank you!

… to Michael Yassa, Zachariah Reagh, and the members of Yassa Lab

Larry Wichlinski, my comps adviser

My friends, for edits and speech practice

Anyone who participated in my pilot study of stimulus similarity