Belief dynamics extraction.

Area of Science:

• Neuroscience
• Computational Biology
• Animal Behavior

Background:

• Animal behavior is influenced by internal states, not just external observations.
• Understanding these internal states is key to deciphering the neural underpinnings of behavior.
• Current methods like inverse model-based Reinforcement Learning face challenges with parameterization and model mismatch.

Purpose of the Study:

• To develop a data-driven approach for inferring latent internal states from observed animal behavior.
• To create a computational model that captures complex behavioral dynamics, including non-exponential waiting times and action-dependent state transitions.
• To validate the model's effectiveness in both simulated and real-world animal behavior tasks.

Main Methods:

• Utilized a partially observable switching semi-Markov process to model animal behavior.
• Applied the model to data from a simulated optimal agent in a foraging task.
• Tested the model on behavioral data from monkeys performing a foraging task.

Main Results:

• The model successfully inferred latent dynamics that corresponded with the belief dynamics of a simulated agent.
• Analysis of monkey foraging behavior revealed distinct latent states, including periods of "expectant waiting."
• The data-driven model effectively captures non-Markovian features crucial for animal behavior.

Conclusions:

• The developed partially observable switching semi-Markov process provides a robust method for inferring latent cognitive states from behavioral data.
• This approach offers a valuable tool for neuroscience research, enabling the measurement of neural representations of inferred states.
• The model's ability to capture complex behavioral patterns opens new avenues for understanding the neural basis of cognition and decision-making.