Dynamic Algorithm Selection (DAS) is closely related to the problem of online acquisition function selection in Bayesian Optimization. In both settings, the objective is to adaptively choose a strategy (algorithm or acquisition function) that maximizes performance under uncertainty and limited evaluation budgets.
Our goal is to conduct a focused literature review in this area to identify:
- transferable methodologies,
- experimental protocols,
- and theoretical insights that can be adapted to the DAS setting.
I propose the following initial set of papers as a starting point:
- https://openreview.net/pdf?id=EPKmSgXvRe
- https://arxiv.org/pdf/2211.01455 + https://github.com/automl/pi_is_back
- https://arxiv.org/pdf/1406.4625
- https://proceedings.mlr.press/v37/hernandez-lobatob15.pdf
- https://arxiv.org/pdf/1009.5419
The objective of this task is to:
- Carefully read and analyze the listed papers,
- Prepare concise summaries and structured notes,
- Extract key ideas/results relevant to Dynamic Algorithm Selection,
- Add comments in this issue.
Summaries may be written in English or Polish, depending on preference.
Feel free to:
- propose and include relevant new papers,
- suggest novel connections between DAS and acquisition function selection,
- highlight potential experimental designs or benchmarks transferable to our setting.
Dynamic Algorithm Selection (DAS) is closely related to the problem of online acquisition function selection in Bayesian Optimization. In both settings, the objective is to adaptively choose a strategy (algorithm or acquisition function) that maximizes performance under uncertainty and limited evaluation budgets.
Our goal is to conduct a focused literature review in this area to identify:
I propose the following initial set of papers as a starting point:
The objective of this task is to:
Summaries may be written in English or Polish, depending on preference.
Feel free to: