Expectile Matrix Factorization for Skewed Data Analysis

Authors: Rui Zhu, Di Niu, Linglong Kong, Zongpeng Li

AAAI 2017 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental For synthetic data with skewed noise and a real-world dataset containing web service response times, the proposed scheme achieves lower recovery errors than the existing matrix factorization method based on least squares in a wide range of settings.
Researcher Affiliation Academia Rui Zhu,1 Di Niu,1 Linglong Kong,2 Zongpeng Li3 1 Department of Electrical and Computer Engineering, University of Alberta, {rzhu3, dniu}@ualberta.ca 2 Department of Mathematical and Statistical Sciences, University of Alberta, lkong@ualberta.ca 3 Department of Computer Science, University of Calgary, zongpeng@ucalgary.ca
Pseudocode Yes Algorithm 1 Alternating minimization for expectile matrix factorization.
Open Source Code No The paper does not provide a link to open-source code or explicitly state that the code is publicly available.
Open Datasets Yes Experiments on Web Service Latency Estimation In these experiments, we aim to recover the web service response times between 339 users and 5825 web services (Zheng, Zhang, and Lyu 2014) distributed worldwide, under different sampling rates.
Dataset Splits No The paper mentions 'sampling rates' and evaluating on 'missing entries' but does not explicitly define or use a separate 'validation' set or specific train/validation/test splits.
Hardware Specification No The paper does not specify the hardware (e.g., CPU, GPU models, memory) used for the experiments.
Software Dependencies No The paper mentions 'standard quadratic program (QP) solvers' but does not list any specific software or library names with version numbers.
Experiment Setup Yes We randomly generate a 1000 × 1000 matrix M = XY T of rank k = 10... We observe some elements in the contaminated matrix and aim to recover the underlying true low-rank M under two sampling rates R = 0.05 and R = 0.1, respectively... The experiment is repeated for 10 times for each R. ... In Fig. 4, we plot the relative errors of recovering missing response times with EMF under different ω.