Tracking Most Significant Shifts in Infinite-Armed Bandits

Authors: Joe Suk, Jung-Hun Kim

ICML 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We validate our findings via experiments on synthetic data, showing our procedures perform better than the previous art for rotting infinite-armed bandits. Here, we demonstrate the performance of Algorithms 1 and 2 on synthetic datasets.
Researcher Affiliation Academia 1Columbia University 2CREST, ENSAE Paris. Correspondence to: Joe Suk <EMAIL>.
Pseudocode Yes Algorithm 1: Blackbox Non-Stationary Algorithm Algorithm 2: Restarting Subsampling Elimination
Open Source Code No The paper does not provide any explicit statement about releasing source code or a link to a code repository for the methodology described in this paper.
Open Datasets No Here, we demonstrate the performance of Algorithms 1 and 2 on synthetic datasets. For the Base-Alg in Algorithm 1 we use UCB (Auer et al., 2002). ... Comparing Total Variation-based Regret Bounds. To ensure a fair comparison between algorithms, we consider a rotting scenario where the mean reward of each selected arm decreases at a rate of O(1/t) at time t. In this environment, for all our algorithms, it can be shown that L = Ω(T), V = O(1).
Dataset Splits No The paper mentions using "synthetic datasets" but does not provide any information about how these datasets were generated or split (e.g., training, validation, test splits) for the experiments.
Hardware Specification No The paper does not provide any specific hardware details such as GPU/CPU models, processor types, or memory used for running the experiments.
Software Dependencies No For the Base-Alg in Algorithm 1 we use UCB (Auer et al., 2002). ... As benchmarks, we implement SSUCB (Bayati et al., 2020) and a variant of it using a sliding window with size T (SSUCB-SW), and AUCBT-ASW (Kim et al., 2024). The paper names specific algorithms but does not provide version numbers for any software, libraries, or environments used in the implementation.
Experiment Setup No The paper describes the scenarios for the synthetic datasets (e.g., "rotting scenario where the mean reward of each selected arm decreases at a rate of O(1/t)", "piecewise stationary setting"). However, it does not provide specific experimental setup details such as hyperparameter values (e.g., learning rates, batch sizes, specific UCB parameters), model initialization, or training configurations.