EURO 2024 Copenhagen
Abstract Submission

EURO-Online login

2155. Limited memory bundle DC algorithm for sparse pairwise kernel learning

Invited abstract in session TD-28: Advancements of OR-analytics in statistics, machine learning and data science 7, stream Advancements of OR-analytics in statistics, machine learning and data science.

Tuesday, 14:30-16:00
Room: 065 (building: 208)

Authors (first author is the speaker)

1. Napsu Karmitsa
Department of Computing, University of Turku
2. Kaisa Joki
Department of Mathematics and Statistics, University of Turku
3. Antti Airola
University of Turku
4. Tapio Pahikkala
Department of Future Technologies, University of Turku, Finland

Abstract

Pairwise learning is a specialized form of supervised learning that focuses on predicting outcomes for pairs of objects. Data with pairwise observations naturally arise, for instance, in recommender systems, information retrieval, drug-target interaction (DTI) prediction, and link prediction in social networks. Here, we formulate the pairwise learning problem as a difference of convex (DC) optimization problem using the Kronecker product kernel, L1- and L0-regularizations, and various possible nonsmooth loss functions. Our aim is to create an efficient learning algorithm, SparsePKL, that produces accurate predictions with the desired sparsity level. In particular, we aim to solve the realistic form of the DTI problem called zero-shot learning, which corresponds to predicting labels for drug-target pairs where neither the drug nor the target is encountered during the training phase. In addition, we propose a novel limited memory bundle DC algorithm (LMB-DCA) for general large-scale nonsmooth DC optimization and apply it as an underlying solver in the SparsePKL. The performance of the SparsePKL algorithm is studied in seven real-world drug-target interaction data, and the results are compared with those of the state-of-the-art methods in pairwise learning. Further, we evaluate the LMB-DCA as a standalone optimization method by comparing it with the well-known DCA.

Keywords

Status: accepted

Back to the list of papers