Reproduce “Belief dynamics extraction” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Arun Kumar, Zhengwei Wu, Xaq Pitkow, Paul Schrater
The authors infer candidate latent states from foraging behavior using a continuous-time switching semi-Markov jump process whose transition rates depend on actions. They first test it on a near-optimal simulated forager with known beliefs, where inferred states correspond to parts of belief dynamics. In macaque behavior, they identify states interpreted as expectant waiting, not direct observations of private beliefs.
Key insights
- The data-driven SMJP approach mitigates model-mismatch biases, offering a more direct inference of internal states compared to inverse model-based Reinforcement Learning.
- Incorporating action-dependent state transitions and non-exponential dwell times is crucial for accurately capturing the complex, continuous-time belief dynamics of animals.
- Uniformization enables the learning of continuous-time semi-Markov processes from discrete observations by transforming them into equivalent discrete-time chains, making EM-based inference tractable.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.