Default Feature Representations of the Cognitive Map

bioRxiv · 12 Nov 2026 · Preprint · Open access

Bazarjani, Piray

The abstract is read at the publisher; the summary is JClub's.

Default Feature Representations (DFR) separates predictive maps into shared features and an environment operator, enabling efficient replanning after structural changes with temporal-difference learning.

Why it matters: Predictive maps can be sensitive to local environmental edits, because far-ahead predictions depend on many states the agent never directly experiences.
What they did: DFR uses fixed features plus an operator; DFR-MB derives a closed-form operator when the structural change is known, while DFR-TD learns the operator from sampled transitions via a temporal-difference rule.
The result: DFR-TD converges to the same solution as DFR-MB, learns the perturbed map from samples, plans comparably to the model-based method, and outperforms successor-representation baselines on replanning.

AI summary of the abstract · may contain errors · not medical advice · read the paper.

Open access
2 cites

Free. We’ll email you when a new paper cites this one or any other paper you follow on JClub — at most one email a day. Every email has a one-tap unsubscribe.

Appellian Neural Networks
bioRxiv · 27 Oct 2026 · Preprint
AI summary · may contain errors
Appellian Neural Network (ANNet) unifies inverse and forward dynamics for a double pendulum by learning Appell’s acceleration energy, enabling accurate real-time simulations for unseen trajectories.
- Open access
Open
Share
Post on X Post on Bluesky Share on LinkedIn
https://jclub.com.au/a/10.64898/2026.06.03.729998
Open in JClub
Dual reinforcement-learning network modules for modeling decision-making with multiple strategies
bioRxiv · 26 Oct 2026 · Preprint
AI summary · may contain errors
A hybrid deep reinforcement learning model reproduces mixed human strategies in a two-step task and matches mouse perceptual decision-making by switching model-free versus inference-based behavior by condition.
- Open access
Open
Share
Post on X Post on Bluesky Share on LinkedIn
https://jclub.com.au/a/10.64898/2026.03.07.709953
Open in JClub
The hippocampus as an epistemic forager: When curiosity and reward jointly steer exploration and hippocampal replay
bioRxiv · 14 Oct 2026 · Preprint
AI summary · may contain errors
The Epistemic Replay Algorithm (ERA) links expected reward and expected information gain to prioritize hippocampal replay, matching key replay biases toward reward sites and decision points.
- Open access
Open
Share
Post on X Post on Bluesky Share on LinkedIn
https://jclub.com.au/a/10.1101/2025.10.31.685837
Open in JClub

Journals this month

All journals →

Topics moving, week to 13 Nov 2026

All topics →

By field

All journals by field →

Default Feature Representations of the Cognitive Map

Related in JClub

Appellian Neural Networks

Dual reinforcement-learning network modules for modeling decision-making with multiple strategies

The hippocampus as an epistemic forager: When curiosity and reward jointly steer exploration and hippocampal replay

Browse JClub