Speaker
Description
To explore the behaviour of expensive black-box functions, such as machine learning model evaluations or physical simulations, it is often useful to fit a surrogate regression model to a sequence of evaluated points. Choosing these points adaptively, rather than relying on a pre-specified design, is advantageous because it places greater emphasis on regions of the configuration space where the regression model is most uncertain. This adaptive, sequential selection of evaluation points is the core idea of active learning for computer experiments.
celecx is a new package in the mlr3 ecosystem that extends its functionality for black-box optimization and surrogate modelling toward active learning. It offers both a high-level user interface with sensible defaults and modular components that can be configured for specialized use cases. A particular focus of celecx is on diagnostics for assessing the quality of the surrogate fit and extrapolating the performance that can be achieved, and at what cost, through additional evaluations.
If you used AI tools or services to support the preparation of this submission, please state the name and reason for using each of them.
gpt-5.4: improvement of language and style of the abstract
gpt-5.2, gpt-5.2-codex, claude opus 4.5: support coding the package
Additional Material or Paper
https://github.com/mlr-org/celecx
| Keywords: Please list up to 5 keywords to help us find the right session for your contribution. | active learning, computer experiments, surrogate modelling, mlr3 |
|---|---|
| Virtual Option | This submission is for onsite presentation only |
| Material License | CC-BY 4.0 |
| Video Recording | Video sharing is fine |
| The author(s) agree(s) to take responsibility and be accountable for the contents of the submission and is/are authorized to present it. | Confirm |
| Interested in serving as reviewer? | martin.binder@stat.uni-muenchen.de |