Research
Amortization moves the cost of inference and design into an offline training phase, so each new problem is answered immediately, fast enough to run in the loop.
Research interests
- Amortized Inference
- Networks trained once that infer for new problems in a single pass.
- Bayesian Experimental Design
- Choosing the next experiment adaptively, by how much it is expected to reveal.
- Bayesian Optimization
- Sample-efficient optimization of expensive black-box objectives.
News
- 2026.09 Our paper “Amortized Bayesian Experimental Design with In-Context Knowledge Conditioning” has been accepted by NeurIPS 2026!
- 2026.06 Started my Ph.D. in the PML Group at Aalto University, supervised by Prof. Samuel Kaski.
Education
-
Ph.D. in Computer Science
Aalto University -
M.Sc. in Business Analytics
Aalto University -
M.Sc. in Computer Science
University of Helsinki -
B.Sc. in Computer Science and Technology
Soochow University