Failure Modes of Variational Autoencoders and Their Effects on Downstream Tasks
arXiv:2007.07124
Abstract
Variational Auto-encoders (VAEs) are deep generative latent variable models that are widely used for a number of downstream tasks. While it has been demonstrated that VAE training can suffer from a number of pathologies, existing literature lacks characterizations of exactly when these pathologies occur and how they impact downstream task performance. In this paper, we concretely characterize conditions under which VAE training exhibits pathologies and connect these failure modes to undesirable effects on specific downstream tasks, such as learning compressed and disentangled representations, adversarial robustness, and semi-supervised learning.
arXiv commentAccepted at the International Conference on Machine Learning (ICML) Workshop on Uncertainty and Robustness in Deep Learning (UDL) 2020
References in corpus (26)
- Auto-Encoding Variational Bayes
- Estimating Mutual Information
- Categorical Reparameterization with Gumbel-Softmax
- Neural Discrete Representation Learning
- Challenging Common Assumptions in the Unsupervised Learning of Disentangled Representations
- An ensemble-based system for automatic screening of diabetic retinopathy
- Glow: Generative Flow with Invertible 1x1 Convolutions
- Improving Variational Inference with Inverse Autoregressive Flow
- Lagging Inference Networks and Posterior Collapse in Variational Autoencoders
- Towards Deeper Understanding of Variational Autoencoding Models
- The Robust Manifold Defense: Adversarial Training using Generative Models
- Towards Realistic Individual Recourse and Actionable Explanations in Black-Box Decision Making Systems
- Inference Suboptimality in Variational Autoencoders
- Learning Latent Subspaces in Variational Autoencoders
- Sticking the Landing: Simple, Lower-Variance Gradient Estimators for Variational Inference
- Reinterpreting Importance-Weighted Autoencoders
- Unsupervised Model Selection for Variational Disentangled Representation Learning
- A Survey of Inductive Biases for Factorial Representation-Learning
- On the Quantitative Analysis of Decoder-Based Generative Models
- Tighter Variational Bounds are Not Necessarily Better
- Doubly Reparameterized Gradient Estimators for Monte Carlo Objectives
- Counterfactual Reasoning for Fair Clinical Risk Prediction
- Adversarial Defense of Image Classification Using a Variational Auto-Encoder
- On the Need for Topology-Aware Generative Models for Manifold-Based Defenses
- Learning Implicit Generative Models Using Differentiable Graph Tests
- Characterizing and Avoiding Problematic Global Optima of Variational Autoencoders