Ti-Fen Pan

Reward function compression facilitates goal-dependent reinforcement learning Reinforcement learning agents learn from rewards, but humans can uniquely assign value to novel, abstract outcomes in a goal-dependent manner. However, this flexibility is cognitively costly, making l...

📢 New preprint!
How do humans learn from arbitrary, abstract goals? We show that, when goal spaces can be compressed, costly working-memory processes give way to internalized reward functions, enabling efficient goal-dependent reinforcement learning. @annecollins.bsky.social arxiv.org/abs/2509.06810

✅ Works with tractable & intractable models
✅ Handles continuous & discrete latent spaces
✅ Applicable to real-world datasets.

We tested our method on various cognitive models, including reinforcement learning models, Bayesian models, and GLM-HMM. A collective effort with @drjingjing.bsky.social @wdt.bsky.social @annecollins.bsky.social‬

Latent variable sequence identification for cognitive models with neural network estimators - Behavior Research Methods Extracting time-varying latent variables from computational cognitive models plays a key role in uncovering the dynamic cognitive processes that drive behaviors. However, existing methods are limited ...

New paper out in Behavioral Research Methods! We introduce a simulation-based method using RNNs to infer trial-varying latent variables from computational cognitive models.
Link: doi.org/10.3758/s134...
#ComputationalCognitiveModeling #SBI