Polynomial Optimization Methods for Matrix Factorization

Authors: Po-Wei Wang, Chun-Liang Li, J. Kolter

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

Reproducibility Variable Result LLM Response
Research Type Experimental On three benchmark datasets we show the method substantially improves convergence speed over state-of-the-art approaches, while also attaining lower objective value.
Researcher Affiliation Academia Po-Wei Wang Machine Learning Department Carnegie Mellon University Pittsburgh, PA 15213 poweiw@cs.cmu.edu Chun-Liang Li Machine Learning Department Carnegie Mellon University Pittsburgh, PA 15213 chunlial@cs.cmu.edu J. Zico Kolter School of Computer Science Carnegie Mellon University Pittsburgh, PA 15213 zkolter@cs.cmu.edu
Pseudocode Yes Algorithm 1: Poly MF-CD and Algorithm 2: Poly MF-SS are provided.
Open Source Code No The paper mentions using existing open-source packages like LIBMF and CCD++, but it does not state that the authors' own implementation code for the proposed methods (Poly MF-CD, Poly MF-SS) is publicly available.
Open Datasets Yes We evaluate the Poly MF algorithm (both the coordinate descent and subspace search approaches) on several benchmark problems in recommender systems. In particular, we consider three datasets: Movielens10m, Netflix (Bell and Koren 2007), and Yahoo Music (Dror et al. 2012).
Dataset Splits No Table 1 provides statistics for Movielens10m, Netflix, and Yahoo-Music, including '|S|' (implied training size) and '|Stest|' (test set size). However, there is no explicit mention or size given for a separate 'validation' dataset split.
Hardware Specification Yes We conduct the experiments on a Intel Core i7-4790 machine with 32 GB memory.
Software Dependencies No The paper mentions using 'LIBMF' and 'CCD++' packages for comparison and as a basis for their method, but it does not specify version numbers for these or any other software libraries or programming environments used in their experiments.
Experiment Setup Yes We choose regularization parameter λ as shown in Table 1, and focus our analysis on the optimization performance compared to the following algorithms:... if func_decr < 10-8max_func_decr then break