Abstract:MOTIVATION:While generative models have shown great success in sampling high-dimensional samples conditional on low-dimensional descriptors (stroke thickness in MNIST, hair color in CelebA, speaker identity in WaveNet), their generation out-of-distribution poses fundamental problems due to the difficulty of learning compact joint distribution across conditions. The canonical example of the conditional variational autoencoder (CVAE), for instance, does not explicitly relate conditions during training and, hence, has no explicit incentive of learning such a compact representation.RESULTS:We overcome the limitation of the CVAE by matching distributions across conditions using maximum mean discrepancy in the decoder layer that follows the bottleneck. This introduces a strong regularization both for reconstructing samples within the same condition and for transforming samples across conditions, resulting in much improved generalization. As this amount to solving a style-transfer problem, we refer to the model as transfer VAE (trVAE). Benchmarking trVAE on high-dimensional image and single-cell RNA-seq, we demonstrate higher robustness and higher accuracy than existing approaches. We also show qualitatively improved predictions by tackling previously problematic minority classes and multiple conditions in the context of cellular perturbation response to treatment and disease based on high-dimensional single-cell gene expression data. For generic tasks, we improve Pearson correlations of high-dimensional estimated means and variances with their ground truths from 0.89 to 0.97 and 0.75 to 0.87, respectively. We further demonstrate that trVAE learns cell-type-specific responses after perturbation and improves the prediction of most cell-type-specific genes by 65%.AVAILABILITY AND IMPLEMENTATION:The trVAE implementation is available via github.com/theislab/trvae. The results of this article can be reproduced via github.com/theislab/trvae_reproducibility.

RMVAE: one-class classification via divergence regularization and maximization mutual information

RSC-VAE: Recoding Semantic Consistency Based VAE for One-Class Novelty Detection

Regularizing Variational Autoencoder with Diversity and Uncertainty Awareness

Conditional out-of-sample generation for unpaired data using trVAE

Consistency Regularization for Variational Auto-Encoders

FVAE: a regularized variational autoencoder using the Fisher criterion

Contrastive Variational Autoencoder Enhances Salient Features

A robust variational autoencoder using beta divergence

Conditional Out-of-distribution Generation for Unpaired Data Using Transfer VAE.

Improve variational autoEncoder with auxiliary softmax multiclassifier

Rethinking Controllable Variational Autoencoders

Multi-Channel Multi-Scale Convolution Attention Variational Autoencoder (MCA-VAE): An Interpretable Anomaly Detection Algorithm Based on Variational Autoencoder

MMD-encouraging convolutional autoencoder: a novel classification algorithm for imbalanced data

Class-Specific Variational Auto-Encoder for Content-Based Image Retrieval

How to train your VAE

SepVAE: a contrastive VAE to separate pathological patterns from healthy ones

NVAE: A Deep Hierarchical Variational Autoencoder

Cross-Modal Information Recovery and Enhancement Using Multiple-Input–Multiple-Output Variational Autoencoder

mcVAE: disentangling by mean constraint

VAE-based Deep SVDD for anomaly detection

Improving Variational Autoencoders with Density Gap-based Regularization