S. Khodadadeh, S. Zehtabian, S. Vahidian, W. Wang, B. Lin, and L. Bölöni

Unsupervised Meta-Learning through Latent-Space Interpolation in Generative Models

Cite as:

S. Khodadadeh, S. Zehtabian, S. Vahidian, W. Wang, B. Lin, and L. Bölöni . Unsupervised Meta-Learning through Latent-Space Interpolation in Generative Models. In to be presented at the International Conference on Learning Representations (ICLR-2021), July 2021.

Download:

(unavailable)

Abstract:

Several recently proposed unsupervised meta-learning approaches rely on synthetic meta-tasks created using techniques such as random selection, clustering and/or augmentation. In this work, we describe a novel approach that generates meta-tasks using generative models. The proposed family of algorithms generate pairs of in-class and out-of-class samples from the latent space in a principled way, allowing us to create synthetic classes forming the training and validation data of a meta-task. We find that the proposed approach, LAtent Space Interpolation Unsupervised Meta-learning (LASIUM), outperforms or is competitive with current unsupervised learning baselines on few-shot classification tasks on the most widely used benchmark datasets.

BibTeX:

@inproceedings{Khodadadeh-2021-ICLR,
  author = "S. Khodadadeh and S. Zehtabian and S. Vahidian and W. Wang and B. Lin and L. B{\"o}l{\"o}ni ",
  title = "Unsupervised Meta-Learning through Latent-Space Interpolation in Generative Models",
  booktitle = "to be presented at the International Conference on Learning Representations (ICLR-2021)",
  year = "2021",
  month = "July",
  xxxacceptance = "29\%",
  abstract = {
  Several recently proposed unsupervised meta-learning approaches rely on synthetic meta-tasks created using techniques such as random selection, clustering and/or augmentation. In this work, we describe a novel approach that generates meta-tasks using generative models. The proposed family of algorithms generate pairs of in-class and out-of-class samples from the latent space in a principled way, allowing us to create synthetic classes forming the training and validation data of a meta-task. We find that the proposed approach, LAtent Space Interpolation Unsupervised Meta-learning (LASIUM), outperforms or is competitive with current unsupervised learning baselines on few-shot classification tasks on the most widely used benchmark datasets.
  },
}

Generated by bib2html.pl (written by Patrick Riley, Lotzi Boloni ) on Fri Jan 29, 2021 20:15:22