Robust Lipschitz Bandits to Adversarial Corruptions

Authors: Yue Kang, Cho-Jui Hsieh, Thomas Chun Man Lee

NeurIPS 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental In this section, we show by simulations that our proposed RMEL and Bo B Robust Zooming algorithm outperform the classic Zooming algorithm in the presence of adversarial corruptions. Average cumulative regrets over 20 repetitions are reported in Figure 1.
Researcher Affiliation Collaboration Yue Kang Department of Statistics University of California, Davis Davis, CA 95616 yuekang@ucdavis.edu Cho-Jui Hsieh Google and Department of Computer Science, UCLA Los Angeles, CA chohsieh@cs.ucla.edu Thomas C. M. Lee Department of Statistics University of California, Davis Davis, CA 95616 tcmlee@ucdavis.edu
Pseudocode Yes Algorithm 1 Robust Zooming Algorithm
Open Source Code No The paper does not provide an explicit statement about releasing the source code for their methodology, nor does it provide a link to a code repository.
Open Datasets No The paper conducts simulations using synthetic mean functions (triangle, sine, two dim) and does not refer to publicly available datasets or provide access information for any data used.
Dataset Splits No The paper conducts simulations for a bandit problem, which involves sequential interaction rather than fixed dataset splits for training, validation, and testing. Therefore, it does not provide specific dataset split information.
Hardware Specification No The paper does not provide specific hardware details such as GPU/CPU models, processor types, or memory amounts used for running its experiments.
Software Dependencies No The paper mentions “Pyxab a python library” as a reference, but it does not provide specific software dependency details with version numbers (e.g., Python 3.x, PyTorch 1.x) for its own implementation.
Experiment Setup Yes We set the time horizon T = 50, 000 (60, 000) for the metric space with d = 1 (2) and the false probability rate δ = 0.01. The random noise at each round is sampled IID from N(0, 0.01). Specifically, we set C = 0 for the non-corrupted case, C = 3, 000 for the moderate-corrupted case and C = 4, 500 for the strong-corrupted case.