Fixed-Confidence Multiple Change Point Identification under Bandit Feedback

Authors: Joseph Lazzaro, Ciara Pike-Burke

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

Reproducibility Variable Result LLM Response
Research Type Experimental We support our theoretical findings with experimental results in synthetic environments demonstrating the efficiency of our method. [...] In Section 6 we conduct experiments in synthetic environments to illustrate MCPI is optimal and that it outperforms existing related works in Clustering Bandits (Yang et al., 2022).
Researcher Affiliation Academia 1Department of Mathematics, Imperial College London, London, England. Correspondence to: Joseph Lazzaro <EMAIL>.
Pseudocode Yes Algorithm 1 Single Change Point Identification (CPI) [...] Algorithm 2 Multiple Change Point Identification (MCPI)
Open Source Code No The paper does not contain any explicit statements about open-sourcing their code, nor does it provide links to a code repository.
Open Datasets No To complement our theoretical results we conduct experiments to test our proposed algorithms in synthetic environments. [...] We simulate our methods in environments with more than one change point. For example, in Figure 3 we consider and environment v2 with K = 19 actions and N = 2 change points with mean rewards µv2 = (2, 2, 2, 2, 2, 2, 4, 4, 4, 4, 4, 4, 4, 0, 0, 0, 0, 0, 0).
Dataset Splits No The experiments are conducted on 'synthetic environments' by running the algorithm multiple times (e.g., '100 runs' or '1000 times') to average results. This type of sequential data collection does not typically involve predefined training/test/validation splits like those found in supervised learning.
Hardware Specification No The paper does not provide specific details about the hardware (e.g., GPU, CPU models, memory) used to run the experiments.
Software Dependencies No The paper does not list any specific software dependencies or their version numbers (e.g., programming languages, libraries, frameworks, or operating systems) used for implementing or running the experiments.
Experiment Setup Yes In Figure 2, we consider an environment v1 with only one change point with means µ = (2, 2, 2, 2, 2, 2, 1, 1, 1). In this case, we run our proposed policy MCPI (setting N = 1) and a range of choices for δ. At each of these choices for δ we run MCPI 100 times and plot the average stopping time. [...] We again plot the average stopping time when running the MPCI algorithm inputting N = 2 100 times at different values for log(1/δ).