the strategic allocation of working memory and episodic
TRANSCRIPT
I. Background
References & acknowledgments1- Einstein, G. O., McDaniel, M. A. et al. (2005). Multiple processes in prospective memory retrieval: factors determining monitoring versus spontaneous retrieval. JEPG.2- Usher, M., & McClelland, J. L. (2001). The time course of perceptual choice: the leaky, competing accumulator model. Psychological Review.3- Brewer et al. (2010). Individual differences in PM: Evidence for multiple processes supporting cue detection. Memory and Cognition.This work was supported by the John Templeton Foundation. [email protected]
The successful realization of future plans, prospective memory or PM, requires the agent to maintain and retrieve a goal for execution at a future time. PM poses a memory problem for periods during which the agent is occupied with other ongoing tasks (OG) while being responsive to target events that demand goal execution. We suggest a mechanistic account of how working memory (WM) and episodic memory (EM) strategies are integrated to strike the right balance between maintenance and retrieval when solving varieties of PM problem.
b
Our mechanistic model combines WM & EM strategies to solve the prospective memory problem, & shows human-like regulation of planned action while perfomring ongoing tasks. Representations & dynamics derived from the model can be compared to patterns & dynamics of fMRI data from PM paradigms to test our proposed mechanism.
The strategic allocation of working memory and episodic memory in prospective remembering: A neural network model
Ida Momennejad*, Momchil Tomov*, Kenneth A. Norman, Jonathan D. CohenPrinceton Neuroscience Institute, Princeton University
OG task: Category matchPM task: Syllable match
II. Behavioral paradigm
III. Neural network modelExp1. Focality X Emphasis
Focal PM: OG task’s stimulus features are same as PM target’s Non-focal PM: Attention to different features for OG stimuli vs. PM targetPM emphasis: Priority of PM vs. OG (e.g. PM more rewarding)
Exp 3. 1 target vs. 6 targets
IV. Simulation of major behavioral phenomena
Strengthening EMimproves PM
and compensatesfor low WM
Exp 2. Focal vs. nonfocal costs over time
Brewer et al. 2010
Block order: * non-PM (baseline OG)* PM * non-PM (aftereffects)
VI. Commission errors
Prediction
WM EM
Conclusions
V. WM capacity & strategic WM-EM balance
1st P
M R
TO
G R
TPM
hitr
ate
Human Simulation- Low WM- High WM
Task
3 R
TA
ccur
acy
- Target: correct task 3 - Non-Target: correct - Target: commission error
Commission errors: PM response is made outsidePM context. We suggest (i) strong encoding of PM
context or (ii) strong EM target-task association can trigger a bottom-up reaction to a former PM target.Over time, context activation & hence CEs diminish.
Task
3 R
T
Acc
urac
yLow PM context & Low EM link
Einstein et al. 2005
Simulation: high EM
PM h
itrat
e
Task-setC
ontext
WM control (PFC-parietal) Input-output mappingsemantic/automatic network
Episodic control (hippocampus)
Cycle
Act
ivat
ion
WM control network: dynamics of LCAs (2)
EM
Task
3 R
T
Task
3 R
T
Human Simulation
Simulation Human Simulation
Human Simulation
OG
RT
(s)
OG
cos
ts (m
s)
Exp 5. After-effectsafter PM task is over,slower RT to a former target during 3rd task
Exp 4. Individual differences in OG RT costs reflect low cost vs. high cost strategies (n=104) O
G R
T
Block #Block #
Human
Einstein et al. 2005
OG
RT
(s)
PM h
itrat
e
Focal Non-focal
Emphasis Emphasis
PMinstructionTarget: tor
ANIMAL cat
VEHICLE dog
BUILDING table
ANIMAL tortoise
SUBJECTmath
Correct resposnes: YES NO NO PM YES
(n=24)
i. High PM context ii. High EM association
No PM PM No PM PM