Collaborative research · Active inference
CPOMDP.
Building agents.
Understanding their inference.
Alongside my PhD, I collaborate on CPOMDP, developing and evaluating artificial agents within the active inference framework.
This work explores how agents infer what is happening around them, and how we can evaluate the models and approximations behind that inference.
A question behind the work
If an agent gets something wrong,
where does the error come from?
An agent’s model of its environment and the way it performs inference can each contribute to error. Our current work examines how to evaluate those contributions separately.
Current work
A paper
in preparation.
Manuscript in preparation
Evaluating models and approximate inference
I am co-authoring the project’s second paper. It examines how predictive-model mismatch and errors in approximate inference can be distinguished and evaluated under controlled conditions.
The work uses simulation to investigate when a diagnostic has an exact reference, and what comparisons remain possible when that reference is unavailable.
This is ongoing research. The project description will develop alongside the manuscript and the wider agent work.
Shared interests
A connection through
perception and inference.
My PhD and CPOMDP approach related questions from different directions. The PhD uses behavioural experiments to study reward and expectation in human cognition. CPOMDP develops and evaluates computational agents.
Both inform my interest in how information is interpreted and prioritised. The precise relationship between the projects will develop as the PhD progresses.
Explore the PhD research