Autoencoders are popular among neural-networkbased matrix completion models due to their ability to retrieve potential latent factors from the partially observed matrices. Nevertheless, when training data is scarce their performance is significantly degraded due to overfitting. In this paper, we mitigate overfitting with a data-dependent regularization technique that relies on the principles of multi-task learning. Specifically, we propose an autoencoder-based matrix completion model that performs prediction of the unknown matrix values as a main task, and manifold learning as an auxiliary task. The latter acts as an inductive bias, leading to solutions that generalize better. The proposed model outperforms the existing autoencoder-based models designed for matrix completion, achieving high reconstruction accuracy in well-known datasets.
Nguyen, MD, Tsiligianni, E, Calderbank, R & Deligiannis, N 2018, Regularizing autoencoder-based matrix completion models via manifold learning. in 2018 26th European Signal Processing Conference, EUSIPCO 2018: EUSIPCO. vol. 2018-September, 8553528, pp. 1880-1884. https://doi.org/10.23919/EUSIPCO.2018.8553528
Nguyen, M. D., Tsiligianni, E., Calderbank, R., & Deligiannis, N. (2018). Regularizing autoencoder-based matrix completion models via manifold learning. In 2018 26th European Signal Processing Conference, EUSIPCO 2018: EUSIPCO (Vol. 2018-September, pp. 1880-1884). Article 8553528 https://doi.org/10.23919/EUSIPCO.2018.8553528
@inproceedings{d2bd18a17c3a46848a28b689de32ba90,
title = "Regularizing autoencoder-based matrix completion models via manifold learning",
abstract = "Autoencoders are popular among neural-networkbased matrix completion models due to their ability to retrieve potential latent factors from the partially observed matrices. Nevertheless, when training data is scarce their performance is significantly degraded due to overfitting. In this paper, we mitigate overfitting with a data-dependent regularization technique that relies on the principles of multi-task learning. Specifically, we propose an autoencoder-based matrix completion model that performs prediction of the unknown matrix values as a main task, and manifold learning as an auxiliary task. The latter acts as an inductive bias, leading to solutions that generalize better. The proposed model outperforms the existing autoencoder-based models designed for matrix completion, achieving high reconstruction accuracy in well-known datasets.",
keywords = "Autoencoder, Deep neural network, Matrix completion, Multi-task learning, Regularization",
author = "Nguyen, \{Minh Duc\} and Evangelia Tsiligianni and Robert Calderbank and Nikolaos Deligiannis",
year = "2018",
month = nov,
day = "29",
doi = "10.23919/EUSIPCO.2018.8553528",
language = "English",
volume = "2018-September",
pages = "1880--1884",
booktitle = "2018 26th European Signal Processing Conference, EUSIPCO 2018",
}