Goal-Directed Inductive Matrix Completion | Center for Big Data Analytics

Abstract: Matrix completion (MC) with additional information has found wide applicability in several machine learning applications. Among algorithms for solving such problems, Inductive Matrix Completion(IMC) has drawn a considerable amount of attention, not only for its well established theoretical guarantees but also for its superior performance in various real-world applications; however, IMC based methods usually place very strong constraints on the quality of the features(side information) to ensure accurate recovery, which might not be met in practice. In this paper, we propose Goal-directed Inductive Matrix Completion(GIMC) to learn a nonlinear mapping of the features so that they satisfy the required properties. A key distinction between GIMC and IMC is that the feature mapping is learnt in a supervised manner, deviating from the traditional approach of unsupervised feature learning followed by model training. Experiments on using our model for several popular machine learning applications including multi-label learning, multiclass classification, and semi-supervised clustering, establish the superiority of the method we propose.

Download: pdf, slides

Citation

Goal-Directed Inductive Matrix Completion (pdf, slides, software)
S. Si, K. Chiang, C. Hsieh, N. Rao, I. Dhillon.
In ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (KDD), pp. 1165–1174, August 2016. (Oral)

Bibtex:
@inproceedings{si2016goaldirec, author = "Si Si AND Kai-Yang Chiang AND Cho-Jui Hsieh AND Nikhil Rao AND Inderjit S. Dhillon", title = "Goal-Directed Inductive Matrix Completion", booktitle = "ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (KDD)", page = "1165—-1174", year = "2016", month = "aug", abstract = "Matrix completion (MC) with additional information has found wide applicability in several machine learning applications. Among algorithms for solving such problems, Inductive Matrix Completion(IMC) has drawn a considerable amount of attention, not only for its well established theoretical guarantees but also for its superior performance in various real-world applications; however, IMC based methods usually place very strong constraints on the quality of the features(side information) to ensure accurate recovery, which might not be met in practice. In this paper, we propose Goal-directed Inductive Matrix Completion(GIMC) to learn a nonlinear mapping of the features so that they satisfy the required properties. A key distinction between GIMC and IMC is that the feature mapping is learnt in a supervised manner, deviating from the traditional approach of unsupervised feature learning followed by model training. Experiments on using our model for several popular machine learning applications including multi-label learning, multiclass classification, and semi-supervised clustering, establish the superiority of the method we propose." }