Median Matrix Completion: from Embarrassment to Optimality

Authors: Weidong Liu, Xiaojun Mao, Raymond K. W. Wong

ICML 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Empirical results are also provided to confirm the effectiveness of the proposed method.We conducted a simulation study, under which we fixed the dimensions to n1 = n2 = 400. In each simulated data, the target matrix A was generated as UVT, where the entries of U Rn1 r and V Rn2 r were all drawn from the standard normal distributions N(0, 1) independently.
Researcher Affiliation Academia 1School of Mathematical Sciences and Mo E Key Lab of Artificial Intelligence Shanghai Jiao Tong University, Shanghai, 200240, China 2School of Data Science, Fudan University, Shanghai, 200433, China 3Department of Statistics, Texas A&M University, College Station, TX 77843, U.S.A.
Pseudocode Yes Algorithm 1 Distributed Least Absolute Deviation Matrix Completion
Open Source Code No The paper does not provide any statement or link regarding the availability of its source code.
Open Datasets Yes We tested various methods on the Movie Lens-100K1 dataset. This data set consists of 100,000 movie ratings provided by 943 viewers on 1682 movies. The ratings range from 1 to 5. [...] 1https://grouplens.org/datasets/movielens/100k/
Dataset Splits Yes all the tuning parameters λN,t in Algorithm 1 were chosen by validation. Namely, we minimized the absolute deviation loss evaluated on an independently generated validation sets with the same dimensions n1, n2.To evaluate the performance of different methods, we directly used the data splittings from the data provider, which splits the data into two sets. We refer them to as Raw A and Raw B.The tuning parameters for all the methods were chosen by 5-fold cross-validations.
Hardware Specification No No specific hardware details (e.g., GPU/CPU models, memory amounts) used for running experiments are provided in the paper.
Software Dependencies No The paper mentions algorithms and methods (e.g., ADMM) but does not specify any software packages or libraries with version numbers (e.g., Python, PyTorch, TensorFlow, scikit-learn versions).
Experiment Setup Yes As h (n1n2) 1/2a N, the bandwidth h was simply set to h = c2(n1n2) 1/2a N, and similarly, ht = c2(n1n2) 1/2a N,t where a N,t was defined by (3.2) with c2 = 0.1. ... If we compute e = b A(t) b A(t 1) 2 F / b A(t 1) 2 F and stop the algorithm once e 10 5, it typically only requires 4 5 iterations. ... For the proposed DLADMC, we fixed the iteration number to 7. ... The tuning parameters for all the methods were chosen by 5-fold cross-validations.