Adversarially-Trained Nonnegative Matrix Factorization

Ting Cai (1) , Vincent y F Tan (1) , Cédric Févotte (2)

1 NUS - National University of Singapore
2 IRIT-SC - Signal et Communications

We consider an adversarially-trained version of the nonnegative matrix factorization, a popular latent dimensionality reduction technique. In our formulation, an attacker adds an arbitrary matrix of bounded norm to the given data matrix. We design efficient algorithms inspired by adversarial training to optimize for dictionary and coefficient matrices with enhanced generalization abilities. Extensive simulations on synthetic and benchmark datasets demonstrate the superior predictive performance on matrix completion tasks of our proposed method compared to state-of-the-art competitors, including other variants of adversarial nonnegative matrix factorization.

Mots clés: Adversarial Training; Non-negative Matrix Factorization; Matrix Completion

Domaines: Traitement du signal et de l'image [eess.SP]; Machine Learning [stat.ML]

Ting Cai, Vincent y F Tan, Cédric Févotte. Adversarially-Trained Nonnegative Matrix Factorization. IEEE Signal Processing Letters, 2021, 28, pp.1415-1419. ⟨10.1109/LSP.2021.3092231⟩. ⟨hal-03431514⟩