CI-VAE: a Generative Deep Learning Model for Class-Specific Data Interpolation

Mohsen Nabian, Zahra Eftekhari, Chi Wah Wong

Primary Area: unsupervised, self-supervised, semi-supervised, and supervised representation learning

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.

Keywords: Deep Learning, Generative Models, Class-Specific Interpolation, Variational Auto-Encoder

Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2024/AuthorGuide.

TL;DR: This paper introduces a generative deep learning frame work for class-specific data interpolation and inter-class data traversal.

Abstract: We propose a new variant of Variational Autoencoder (VAE), Class-Informed VAE (CI-VAE), that enables interpolation between arbitrary pairs of observations of the same class. CI-VAE combines the general VAE architecture with a linear discriminator layer on the latent space to enforce the construction of a latent space where observations from different classes are linearly separable. This allows for robust latent-space linear traversal and data generation between two arbitrary observations of the same class, which has potential applications in science and engineering. One specific application is to enhance understanding of the biological processes involving the development of diseases or cancer from healthy cells. We demonstrate the effectiveness of CI-VAE on the MNIST dataset of handwritten digits, showing that it significantly improves class-specific linear traversal and data augmentation compared to VAE while maintaining comparable reconstruction error. We also apply CI-VAE to a study of colon cancer single-cell genomics data, showing that interpolation between normal cells and tumor cells using CI-VAE may enhance our understanding of the mechanism of cancer development.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors' identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Submission Number: 4324