Efficient Low-Rank Matrix Estimation, Experimental Design, and Arm-Set-Dependent Low-Rank Bandits

Authors: Kyoungseok Jang, Chicheng Zhang, Kwang-Sung Jun

ICML 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We now evaluate the empirical performance of Low Pop Art and our proposed experimental design to validate our improvement.
Researcher Affiliation Academia 1Dipartimento di Informatica, Università degli Studi di Milano, Milan, MI, Italy 2Department of Computer Science, University of Arizona, Tucson, AZ, United States.
Pseudocode Yes Algorithm 1 Low Pop Art
Open Source Code Yes please check https://github.com/jajajang/Low Pop Art for the code.
Open Datasets Yes We used the Movielens dataset (movielens-old 100k) to try the algorithm on a real-world dataset.
Dataset Splits No No explicit mention of specific training, validation, or test dataset splits or cross-validation methodology.
Hardware Specification Yes Computation resource: Apple M2 Pro, 16GB memory.
Software Dependencies No Software like Scikit-learn and cvxpy are mentioned, but no specific version numbers are provided.
Experiment Setup Yes For all experiments, we set ground truth Θ = uv where , u Unif(Sd1 1) and v Unif(Sd2 1) and we sample Θ before each experiment starts. The noise of the reward ηt N(0, 1). All plots are generated by averaging over 60 number of random instances.