Pairwise Elimination with Instance-Dependent Guarantees for Bandits with Cost Subsidy

Authors: Ishank Juneja, Carlee Joe-Wong, Osman Yagan

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

Reproducibility Variable Result LLM Response
Research Type Experimental Finally, experiments are conducted using the Movie Lens 25M and Goodreads datasets for both PE and PE-CS revealing the effectiveness of PE and the superior balance between performance and reliability offered by PE-CS compared to baselines from the literature.
Researcher Affiliation Academia Ishank Juneja, Carlee Joe-Wong & Osman Yagan Department of Electrical and Computer Engineering Carnegie Mellon University Pittsburgh, PA 15213, USA EMAIL
Pseudocode Yes Algorithm 1: Pairwise Elimination (PE) for a known reference arm ℓ Algorithm 2: Pairwise Elimination for Cost Subsidy Problem (PE-CS). Algorithm 3: Cost-Subsidized Explore-Then-Commit (ETC-CS) Algorithm 4: Cost-Subsidized Thompson Sampling with Beta Priors (TS-CS) Algorithm 5: Cost-Subsidized UCB (UCB-CS) Algorithm 6: FIXED THRESHOLD UCB (FT-UCB)
Open Source Code Yes Code available at https://github.com/ishank-juneja/bandits-with-costs
Open Datasets Yes Finally, experiments are conducted using the Movie Lens 25M and Goodreads datasets for both PE and PE-CS revealing the effectiveness of PE and the superior balance between performance and reliability offered by PE-CS compared to baselines from the literature. F. Maxwell Harper and Joseph A. Konstan. The movielens datasets: History and context. ACM Trans. Interact. Intell. Syst., 5(4), dec 2015. ISSN 2160-6455. doi: 10.1145/2827872. URL https://doi.org/10.1145/2827872. Mengting Wan and Julian Mc Auley. Item recommendation on monotonic behavior chains. In Proceedings of the 12th ACM conference on recommender systems, pp. 86 94, 2018.
Dataset Splits No The paper describes how bandit instances are constructed from the Movie Lens and Goodreads datasets (e.g., 20 arms for Movie Lens, 8 for Goodreads) and how rewards and costs are assigned. However, it does not provide specific training/test/validation splits for these datasets in the traditional sense of partitioning the data for distinct training, validation, and testing phases. The experiments involve sequential sampling over a time horizon T (e.g., T=5M).
Hardware Specification No The paper does not provide specific hardware details such as GPU models, CPU types, or memory specifications used for running the experiments.
Software Dependencies No The paper does not mention any specific software dependencies with version numbers (e.g., Python, PyTorch, CUDA versions) that would be needed to replicate the experiments.
Experiment Setup Yes Data points represent terminal regret at T =5M and each data point represents the outcome from an experiment. There are 25 such independent runs for each algorithm. For every genre we first obtain the mean 5-point scale rating of all books or movies tagged with that genre and then divide this rating by 5 so that it lies between 0 and 1. We then treat this fractional rating as the expected reward return from that genre.