A memory-based account of retrospective revaluation

Area of Science:

• Cognitive Psychology
• Computational Neuroscience
• Machine Learning

Background:

• Human memory research often models learning and recall separately.
• Retrospective revaluation, where prior learning influences later judgments, presents a challenge for existing memory models.
• Associative learning requires understanding how events are linked and updated in memory.

Purpose of the Study:

• To adapt the instance-based human memory model, Minerva 2, for simulating retrospective revaluation.
• To investigate how memory traces and expectancy-encoding contribute to complex associative learning.
• To computationally model phenomena such as backward blocking and recovery from blocking.

Main Methods:

• Utilizing the Minerva 2 instance model, which stores individual trial events as separate memory traces.
• Implementing a parallel probe mechanism where memory contacts all traces simultaneously.
• Modeling learning as cued-recall and encoding via differential encoding of unexpected probe features (expectancy-encoding).

Main Results:

• The adapted Minerva 2 model successfully simulated three key examples of retrospective revaluation.
• Demonstrated backward blocking, recovery from blocking, and backward conditioned inhibition within the model.
• Showcased the model's ability to integrate human memory principles with complex associative learning.

Conclusions:

• Instance-based memory models, like Minerva 2, can effectively simulate retrospective revaluation.
• Expectancy-encoding provides a mechanism for updating memory representations based on prediction errors.
• This work bridges computational models of memory with theories of associative learning.