Keynote Lecture: Anna Schapiro - CCN 2025

Value of the Information & Strength of the Argument

The talk provides substantial value by integrating computational modeling with empirical data to propose a novel mechanistic account of hippocampal function. The argumentation is solid, building on established theory (CLS) and addressing key questions with testable predictions. The speaker clearly explains the logic and evidence, making a compelling case for the role of the monosynaptic pathway in gradual learning. The presentation is well-structured, with clear transitions between ideas and appropriate use of examples.

Scientific Rigor, Source Quality, Title Accuracy

The scientific rigor is high, as the talk is based on peer-reviewed research and established theories. The speaker cites relevant literature and presents original data. The title accurately reflects the content, focusing on learning representations of specifics and generalities over time. The talk is appropriate for a specialized audience, and the speaker does not overstate conclusions, acknowledging limitations and open questions.

The title accurately reflects the content, which focuses on learning representations of specifics and generalities over time.

8/10

The talk presents original research from a leading lab, grounded in established theory (CLS) and supported by computational models and empirical data. The speaker is a recognized expert, and the content is consistent with current scientific understanding. However, as a conference keynote, it may not undergo the same peer-review scrutiny as a journal article.

Markers derived by PSI from the transcript: the creator did not define chapters.

• 02:00Introduction and overview of the question: how do we learn across timescales?
• 05:00Complementary learning systems theory and the role of hippocampus in rapid learning and replay.
• 10:00Proposal that hippocampus is not homogeneous, with distinct pathways for different learning functions.
• 15:00Computational model SEAHORSE and its two pathways (trisynaptic and monosynaptic).
• 20:00Evidence for statistical learning in hippocampus and necessity of hippocampus for this learning.
• 25:00Application of model to category learning tasks, showing trade-off between pathways.
• 30:00New project on mixture of experts with prefrontal cortex for adaptive use of representations.
• 35:00Associative inference task and predictions for different learning strategies.
• 40:00Behavioral results showing interleaved learning leads to overlapping representations and false alarms.
• 45:00Discussion of implications and future directions.

• Complementary learning systems theory — Referenced as foundational theory for the talk.
• SEAHORSE model — Computational model of hippocampus developed by the lab.
• REMERGE model — Alternative model for associative inference.

• Complementary learning systems theory — The talk builds on this theory and extends it.
• Hippocampal subfields — The talk discusses distinct roles of CA1, CA3, and dentate gyrus.

• Alternative models of hippocampal function — Other models may emphasize different mechanisms for inference and learning.