United States Department of Transportation
i
Noisy Blackbox Optimization Using Multi-Fidelity Queries: A Tree Search Approach
-
2020-08-01
[PDF-727.14 KB]
Details:
-
Alternative Title:Sensing and Communications in V2V and V2I Settings [Project Title, from cover]
-
Creators:
-
Corporate Creators:
-
Corporate Contributors:
-
Subject/TRT Terms:
-
Publication/ Report Number:
-
Resource Type:
-
Geographical Coverage:
-
Corporate Publisher:
-
Abstract:We study the problem of black-box optimization of a noisy function in the presence of low-cost approximations or fidelities, which is motivated by problems like hyper-parameter tuning. In hyper-parameter tuning evaluating the black-box function at a point involves training a learning algorithm on a large dataset at a particular hyper-parameter and evaluating the validation error. Even a single such evaluation can be prohibitively expensive. Therefore, it is beneficial to use low cost approximations, like training the learning algorithm on a sub-sampled version of the whole data-set. These low-cost approximations/ fidelities can however provide a biased and noisy estimate of the function value. In this work, we combine structured state-space exploration through hierarchical partitioning with querying these partitions at multiple fidelities, and develop a multi-fidelities bandit based tree-search algorithm for noisy blackbox optimization. We derive simple regret guarantees for our algorithm and validate its performance on real and synthetic datasets.
-
Format:
-
Funding:
-
Collection(s):
-
Main Document Checksum:
-
Download URL:
-
File Type:
