Discovering Hidden Factors of Variation in Deep Networks

Abstract

Deep learning has enjoyed a great deal of success because of its ability to learn useful features for tasks such as classification. But there has been less exploration in learning the factors of variation apart from the classification signal. By augmenting autoencoders with simple regularization terms during training, we demonstrate that standard deep architectures can discover and explicitly represent factors of variation beyond those relevant for categorization. We introduce a cross-covariance penalty (XCov) as a method to disentangle factors like handwriting style for digits and subject identity in faces. We demonstrate this on the MNIST handwritten digit database, the Toronto Faces Database (TFD) and the Multi-PIE dataset by generating manipulated instances of the data. Furthermore, we demonstrate these deep networks can extrapolate `hidden' variation in the supervised signal.

Presented at International Conference on Learning Representations 2015 Workshop

References in corpus (3)

• ADADELTA: An Adaptive Learning Rate Method
• Going Deeper with Convolutions
• Better Mixing via Deep Representations

Cited by in corpus (19)

• Disentangling factors of variation in deep representations using adversarial training
• Promises and pitfalls of deep neural networks in neuroimaging-based psychiatric research
• GeneGAN: Learning Object Transfiguration and Attribute Subspace from Unpaired Data
• Disentangled Representations for Short-Term and Long-Term Person Re-Identification
• A Survey of Inductive Biases for Factorial Representation-Learning
• Challenges in Disentangling Independent Factors of Variation
• Controllable Data Generation by Deep Learning: A Review
• Effect of The Latent Structure on Clustering with GANs
• Attribute Guided Unpaired Image-to-Image Translation with Semi-supervised Learning
• All-In-One: Facial Expression Transfer, Editing and Recognition Using A Single Network
• Conditional Adversarial Generative Flow for Controllable Image Synthesis
• An Image is Worth More Than a Thousand Words: Towards Disentanglement in the Wild
• Improving Content-Invariance in Gated Autoencoders for 2D and 3D Object Rotation
• Decomposing Normal and Abnormal Features of Medical Images into Discrete Latent Codes for Content-Based Image Retrieval
• Disentangling images with Lie group transformations and sparse coding
• Efficient decorrelation of features using Gramian in Reinforcement Learning
• Quantifying the Effects of Enforcing Disentanglement on Variational Autoencoders
• Attribute-controlled face photo synthesis from simple line drawing
• Novel View Synthesis via Depth-guided Skip Connections