Learning Deep Disentangled Embeddings with the F-Statistic Loss

Abstract

Deep-embedding methods aim to discover representations of a domain that make explicit the domain's class structure and thereby support few-shot learning. Disentangling methods aim to make explicit compositional or factorial structure. We combine these two active but independent lines of research and propose a new paradigm suitable for both goals. We propose and evaluate a novel loss function based on the $F$ statistic, which describes the separation of two or more distributions. By ensuring that distinct classes are well separated on a subset of embedding dimensions, we obtain embeddings that are useful for few-shot learning. By not requiring separation on all dimensions, we encourage the discovery of disentangled representations. Our embedding method matches or beats state-of-the-art, as evaluated by performance on recall@$k$ and few-shot learning tasks. Our method also obtains performance superior to a variety of alternatives on disentangling, as evaluated by two key properties of a disentangled representation: modularity and explicitness. The goal of our work is to obtain more interpretable, manipulable, and generalizable deep representations of concepts and categories.

Cited by in corpus (37)

• Multi-Object Representation Learning with Iterative Variational Inference
• Unsupervised deep learning identifies semantic disentanglement in single inferotemporal neurons
• Are Disentangled Representations Helpful for Abstract Visual Reasoning?
• Learning Disentangled Representations in the Imaging Domain
• Weakly-Supervised Disentanglement Without Compromises
• Weakly Supervised Disentanglement with Guarantees
• Spatial Broadcast Decoder: A Simple Architecture for Learning Disentangled Representations in VAEs
• Disentangling Factors of Variation Using Few Labels
• Unsupervised Model Selection for Variational Disentangled Representation Learning
• Learning Disentangled Behaviour Patterns for Wearable-based Human Activity Recognition
• Robustly Disentangled Causal Mechanisms: Validating Deep Representations for Interventional Robustness
• Towards Nonlinear Disentanglement in Natural Data with Temporal Sparse Coding
• 3D-IDS: Doubly Disentangled Dynamic Intrusion Detection
• Generative causal explanations of black-box classifiers
• Semi-Supervised StyleGAN for Disentanglement Learning
• Theory and Evaluation Metrics for Learning Disentangled Representations
• An Identifiable Double VAE For Disentangled Representations
• Self-Adversarial Disentangling for Specific Domain Adaptation
• On the Transfer of Disentangled Representations in Realistic Settings
• Modularity in Deep Learning: A Survey
• Demystifying Inter-Class Disentanglement
• On the Fairness of Disentangled Representations
• Factorized Discriminant Analysis for Genetic Signatures of Neuronal Phenotypes
• A Plug-in Method for Representation Factorization in Connectionist Models
• Disentangling by Subspace Diffusion
• Measuring the Biases and Effectiveness of Content-Style Disentanglement
• Generalization and Robustness Implications in Object-Centric Learning
• Video Reenactment as Inductive Bias for Content-Motion Disentanglement
• Discond-VAE: Disentangling Continuous Factors from the Discrete
• DCI-ES: An Extended Disentanglement Framework with Connections to Identifiability
• Algorithm Fairness in AI for Medicine and Healthcare
• Linear Disentangled Representations and Unsupervised Action Estimation
• Revisiting Factorizing Aggregated Posterior in Learning Disentangled Representations
• Disentanglement Learning via Topology
• DEFT: Distilling Entangled Factors by Preventing Information Diffusion
• Quantifying and Learning Linear Symmetry-Based Disentanglement
• Towards Better Understanding of Disentangled Representations via Mutual Information