Our latest work on understanding the behavior of bounded agents in more naturalistic tasks accepted at #ICLR2025: Inverse decision-making using neural amortized Bayesian actors, with @dominikstrb.bsky.social @tobnie.bsky.social and @jan-peters.bsky.social based on @tobnie.bsky.social MSc thesis - ThreadSky

About ThreadSky

c-rothkopf.bsky.social • 25 days ago

Our latest work on understanding the behavior of bounded agents in more naturalistic tasks accepted at #ICLR2025: Inverse decision-making using neural amortized Bayesian actors, with @dominikstrb.bsky.social @tobnie.bsky.social and @jan-peters.bsky.social based on @tobnie.bsky.social MSc thesis

Comments

c-rothkopf.bsky.social•25 days ago

Bayesian observer & actor models are a powerful framework that explains behavioral biases and variability in perception, sensorimotor control, and more by linking priors, uncertainties, and costs

c-rothkopf.bsky.social•25 days ago

But, when we extend these models to naturalistic tasks with continuous actions and complex cost-functions, solving the Bayesian decision-making problem quickly becomes analytically intractable

c-rothkopf.bsky.social•25 days ago

Inverse decision-making (inferring model parameters from behavioral data) is even tougher. Typically, one simplifies models (e.g., using Gaussian distributions or quadratic costs) or uses approximations. But naturalistic goal-directed behavior may involve much more complex, structured cost functions

c-rothkopf.bsky.social•25 days ago

Here we amortize the Bayesian actor using a neural network trained unsupervised on a broad range of parameter settings. This allows for efficient gradient-based Bayesian inference of the model’s parameters within probabilistic programming

c-rothkopf.bsky.social•25 days ago

On synthetic data, our method’s inferred posterior distributions closely match those from analytical solutions—when available—and recover ground truth distributions even when no analytical solution exists

c-rothkopf.bsky.social•25 days ago

This approach also enables principled model comparison and helps disentangle factors that lead to unidentifiability between priors and costs—a common issue in behavioral modeling

Posting Rules

Be respectful to others
No spam or self-promotion
Stay on topic
Follow Bluesky's terms of service

Comments

Posting Rules

Reply