Homomorphic Matrix Completion

Authors: Xiao-Yang Liu, Zechu (Steven) Li, Xiaodong Wang

NeurIPS 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Finally, on synthetic data and real-world data, we show that both homomorphic nuclear-norm minimization and alternating minimization algorithms achieve accurate recoveries on cyphertexts, verifying the homomorphism property.
Researcher Affiliation Academia 1Department of Electrical Engineering, Columbia University, New York, 2Department of Computer Science, Columbia University, New York,
Pseudocode Yes Algorithm 1 Homomorphic matrix completion at the cloud server Algorithm 2 Homomorphic matrix completion at node j, for j = 1, ..., n2
Open Source Code No The paper states in the 'Broader Impact Statement' that code is included, but does not provide a direct link or explicit statement in the main body of the paper for the code related to this specific work.
Open Datasets Yes The real-world datasets include two benchmark datasets for recommendation systems, namely the Movie Lens10M (Top 400)3 and Netflix (Top 400) datasets. The Movie Lens dataset contains ratings of 400 most rated movies made by approximately 7, 000 users, and the Netflix dataset contains ratings of 400 most rated movies made by approximately 480 thousand users. 3https://movielens.org/
Dataset Splits No The paper does not explicitly state training, validation, or test dataset splits.
Hardware Specification No The paper does not provide specific hardware details (e.g., GPU/CPU models, memory) used for running its experiments.
Software Dependencies No The paper does not list specific software dependencies with version numbers.
Experiment Setup Yes We set k = 10 in Alg. 1 and Alg. 2. For the newly introduced compared algorithm FW, we set the privacy parameter = 2 log(1/δ) and δ = 10 6. For the NN and AM algorithms, the setting is the same in Section 6.2.