Bayesian representation learning with oracle constraints

Abstract

Representation learning systems typically rely on massive amounts of labeled data in order to be trained to high accuracy. Recently, high-dimensional parametric models like neural networks have succeeded in building rich representations using either compressive, reconstructive or supervised criteria. However, the semantic structure inherent in observations is oftentimes lost in the process. Human perception excels at understanding semantics but cannot always be expressed in terms of labels. Thus, \emph{oracles} or \emph{human-in-the-loop systems}, for example crowdsourcing, are often employed to generate similarity constraints using an implicit similarity function encoded in human perception. In this work we propose to combine \emph{generative unsupervised feature learning} with a \emph{probabilistic treatment of oracle information like triplets} in order to transfer implicit privileged oracle knowledge into explicit nonlinear Bayesian latent factor models of the observations. We use a fast variational algorithm to learn the joint model and demonstrate applicability to a well-known image dataset. We show how implicit triplet information can provide rich information to learn representations that outperform previous metric learning approaches as well as generative models without this side-information in a variety of predictive tasks. In addition, we illustrate that the proposed approach compartmentalizes the latent spaces semantically which allows interpretation of the latent variables.

16 pages, publishes in ICLR 16

Cited by in corpus (23)

• Learning Factorized Multimodal Representations
• Harnessing Deep Neural Networks with Logic Rules
• Isolating Sources of Disentanglement in Variational Autoencoders
• Early Visual Concept Learning with Unsupervised Deep Learning
• Are Disentangled Representations Helpful for Abstract Visual Reasoning?
• Disentangling Hate in Online Memes
• Modeling Uncertainty with Hedged Instance Embedding
• Variational Inference of Disentangled Latent Concepts from Unlabeled Observations
• Disentangling Factors of Variation Using Few Labels
• Stochastic Prototype Embeddings
• Evaluating the Disentanglement of Deep Generative Models through Manifold Topology
• Context-aware learning for generative models
• On the Fairness of Disentangled Representations
• Factorized Discriminant Analysis for Genetic Signatures of Neuronal Phenotypes
• Uncertainty Estimates for Ordinal Embeddings
• Semantic Adversarial Network for Zero-Shot Sketch-Based Image Retrieval
• Learning Flat Latent Manifolds with VAEs
• Multi-modal data generation with a deep metric variational autoencoder
• Scalable and Effective Deep CCA via Soft Decorrelation
• Odd-One-Out Representation Learning
• Open-Ended Content-Style Recombination Via Leakage Filtering
• Variational learning across domains with triplet information
• Semi-Supervised Few-Shot Classification with Deep Invertible Hybrid Models