MVPA EEG, Memory Reinstatement, and Credit Assignment
Date:
Gave a research talk to the Knight Lab at UC Berkeley focused on how multivariate EEG decoding techniques can be applied to investigate memory reinstatement and credit assignment in decision-making contexts. This talk built on previous work comparing univariate and multivariate ERP approaches but shifted emphasis toward learning dynamics and cognitive control.
Key sections included:
- A review of decoding methods (SVM, cross-validated Mahalanobis) and their application to ERP data
- Evidence that rare events and memory-relevant signals can be tracked more sensitively using MVPA
- A two-armed bandit paradigm demonstrating how outcome feedback may retroactively influence representations not actively held in working memory
This presentation was part of an ongoing discussion about how decoding tools can clarify the mechanisms underlying learning, memory, and decision-making.