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.

Radio Galaxy Morphology Simulation via Residual Conditional Variational Autoencoder

Radio Galaxy Zoo: Leveraging latent space representations from variational autoencoder

Radio Galaxy Morphology Generation Using DNN Autoencoder and Gaussian Mixture Models

Simulation-based inference of the 2D ex-situ stellar mass fraction distribution of galaxies using variational autoencoders

CosmoVAE: Variational Autoencoder for CMB Image Inpainting

Structured Variational Inference for Simulating Populations of Radio Galaxies

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

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

From Images to Features: Unbiased Morphology Classification via Variational Auto-Encoders and Domain Adaptation

Classification of Radio Galaxy Images with Semi-supervised Learning.

Radio Galaxy Zoo: Knowledge Transfer Using Rotationally Invariant Self-Organising Maps

A Physics-Informed Variational Autoencoder for Rapid Galaxy Inference and Anomaly Detection

Similar image retrieval using Autoencoder. I. Automatic morphology classification of galaxies

E(2)-Equivariant Features in Machine Learning for Morphological Classification of Radio Galaxies

Conditional Variational Autoencoder for Learned Image Reconstruction

Deep Learning for Morphological Identification of Extended Radio Galaxies using Weak Labels

A model local interpretation routine for deep learning based radio galaxy classification

Morphological Classification of Radio Galaxies with wGAN-supported Augmentation

Morphological classification of compact and extended radio galaxies using convolutional neural networks and data augmentation techniques

Morphological classification of Radio Galaxies using Semi-Supervised Group Equivariant CNNs

Automatic Morphological Classification of Galaxies: Convolutional Autoencoder and Bagging-based Multiclustering Model