New method disentangles latent subspaces under correlation shifts.
problem Correlations between factors of variation make disentanglement models less robust.
method Enforces independence between subspaces conditioned on available attributes using adversarial CMI minimization.
result Models are disentangled and robust under correlation shifts, including in weakly supervised settings.
New method disentangles hidden data structures using HSIC and supervision.
problem Tackles the challenge of interpreting high-dimensional data.
method Supervised Independent Subspace Principal Component Analysis (sisPCA) using HSIC.
result Identifies and separates hidden data structures effectively.
New method learns fair representations by separating out protected attributes.
problem Learning fair representations invariant to protected attributes.
method FD-VAE: disentangles latent space into target, protected, and mutual attributes.
result FD-VAE outperforms previous methods in fairness metrics.
Unified framework for disentangled representations using mechanistic independence.
problem Identifiability of disentangled latent factors under statistical dependencies.
method Introduces mechanistic independence to characterize latent factors by their actions on observed variables, proposing various independence criteria.
result Establishes conditions for identifiability of latent subspaces without statistical assumptions.
GEOMANCER learns manifold factors without supervision.
problem Learning to factorize Lie group orbits from data.
method Subspace diffusion estimation of invariant subspaces.
result Unsupervised disentanglement of complex manifolds possible.
Learning the disentangled representation of interpretable generative factors of data is one of the foundations to allow artificial intelligence to think like people. In this paper, we propose the analogical training strategy for the unsupervised disentangled representation learning in generative models. The analogy is …
Image generating neural networks are mostly viewed as black boxes, where any change in the input can have a number of globally effective changes on the output. In this work, we propose a method for learning disentangled representations to allow for localized image manipulations. We use face images as our example of cho…
In this work we present an adversarial training algorithm that exploits correlations in video to learn --without supervision-- an image generator model with a disentangled latent space. The proposed methodology requires only a few modifications to the standard algorithm of Generative Adversarial Networks (GAN) and invo…
Linear disentangled representations improve unsupervised action estimation.
problem Learning linear disentangled representations for unsupervised action estimation.
method Developed a method to induce irreducible representations in VAE models without labeled action sequences.
result Linear disentangled representations are a desirable property for unsupervised action estimation.
Recently there has been an increased interest in unsupervised learning of disentangled representations using the Variational Autoencoder (VAE) framework. Most of the existing work has focused largely on modifying the variational cost function to achieve this goal. We first show that these modifications, e.g. beta-VAE, …
We develop a generalisation of disentanglement in VAEs---decomposition of the latent representation---characterising it as the fulfilment of two factors: a) the latent encodings of the data having an appropriate level of overlap, and b) the aggregate encoding of the data conforming to a desired structure, represented t…
VDA improves disentanglement of latent representations in complex signals.
problem Learning disentangled and interpretable representations in nonstationary, high-dimensional time-evolving signals.
method Variational decomposition autoencoding (VDA) framework, incorporating signal decomposition, contrastive self-supervised task, and variational prior approximation.
result DecVAEs surpass state-of-the-art VAE-based methods in disentanglement quality and generalization.
The paper formalizes how concepts are encoded in text-guided generative models and provides a method to manipulate them.
problem Encoding and manipulating concepts in text-guided generative models.
method Formalizing concepts as subspaces of a representation space, developing algebraic manipulation methods.
result The ability to manipulate concepts in generative models through algebraic operations on the representation.
This paper analyzes AJIVE for estimating shared subspace across multiple datasets, revealing its strengths and limitations.
problem Estimating shared subspace across multiple datasets with varying degrees of misalignment.
method Angle-based Joint and Individual Variation Explained (AJIVE) method, a two-stage spectral approach.
result AJIVE's performance in high signal-to-noise ratio (SNR) regimes and its non-diminishing error in low-SNR settings.
We tackle the problem disentangling the latent space of an autoencoder in order to separate labelled attribute information from other characteristic information. This then allows us to change selected attributes while preserving other information. Our method, matrix subspace projection, is much simpler than previous ap…
SIG model identifies invariant variables for MSDA with fewer domain constraints.
problem Challenges in enforcing minimal changes across domains for MSDA.
method Subspace identification theory and variational inference.
result SIG model outperforms existing techniques on various benchmark datasets.
New method neutralizes gender bias in word embeddings without losing semantic information.
problem Gender biases in word embeddings trained on human-generated corpora.
method Latent Disentanglement and Counterfactual Generation with siamese auto-encoder and gradient reversal layer.
result Our method outperforms existing debiasing methods in preserving semantic information and neutralizing gender biases.
We tackle the challenge of disentangled representation learning in generative adversarial networks (GANs) from the perspective of regularized optimal transport (OT). Specifically, a smoothed OT loss gives rise to an implicit transportation plan between the latent space and the data space. Based on this theoretical obse…
Weakly-supervised RL identifies meaningful tasks, improving performance in complex environments.
problem Learning to efficiently explore and distinguish between meaningful and irrelevant tasks.
method Weak supervision to automatically disentangle meaningful tasks from a large space of nonsensical tasks.
result The learned subspace of meaningful tasks leads to substantial performance gains, especially in complex environments.
Proposes HeteroJIVE for joint subspace estimation in multi-view data with statistical and structural heterogeneity.
problem Joint subspace estimation in multi-view data with varying statistical and structural heterogeneity.
method HeteroJIVE: A weighted two-stage spectral algorithm addressing statistical and structural heterogeneity.
result HeteroJIVE achieves the O(K−1/2) rate without iterative refinement, validating the oracle-optimal weighting scheme. Generative networks are analyzed using spline operators to understand their properties and limitations.
problem Understanding and optimizing the properties of deep generative networks.
method Characterizing latent space partition, manifold dimension, and disentanglement using spline operators.
result Characterized the latent space partition, manifold dimension, and disentanglement of GDNs.
We propose the factorized action variational autoencoder (FAVAE), a state-of-the-art generative model for learning disentangled and interpretable representations from sequential data via the information bottleneck without supervision. The purpose of disentangled representation learning is to obtain interpretable and tr…
Improved disentanglement in VAEs using aggregated feature maps.
problem Improving disentanglement in Variational Autoencoders (VAEs).
method Regionally aggregated feature maps extracted from pre-trained CNNs on ImageNet.
result 2nd place in NeurIPS 2019 disentanglement challenge.
To evaluate disentangled representations several metrics have been proposed. However, theoretical guarantees for conventional metrics of disentanglement are missing. Moreover, conventional metrics do not have a consistent correlation with the outcomes of qualitative studies. In this paper we analyze metrics of disentan…
New technique learns causally disentangled representations for better generation.
problem Learning disentangled representations for accurate generation.
method Causally Disentangled Generation (CDG) approach with supervised regularization.
result CDG is necessary and sufficient for accurate disentangled generation.
Disentangled representations, where the higher level data generative factors are reflected in disjoint latent dimensions, offer several benefits such as ease of deriving invariant representations, transferability to other tasks, interpretability, etc. We consider the problem of unsupervised learning of disentangled rep…
Neural recordings are nonstationary time series, i.e. their properties typically change over time. Identifying specific changes, e.g. those induced by a learning task, can shed light on the underlying neural processes. However, such changes of interest are often masked by strong unrelated changes, which can be of physi…
Improved disentanglement through learned feature aggregation.
problem Disentangling latent factors in images.
method Variational autoencoder trained on regionally aggregated feature maps from ImageNet.
result 2nd place in NeurIPS 2019 disentanglement challenge.
We analyze disentangled representations under a causal generative process, proposing new metrics and datasets.
problem Addressing fairness and interpretability through disentangled representations with a causal perspective.
method Work under a causal generative process, proposing new metrics and datasets to study disentanglement.
result Proposed metrics capture the desiderata of disentangled causal process.
New framework for disentangling graph node and edge features.
problem Learning disentangled representations for attributed graphs with node and edge features.
method Proposes a novel variational objective and architecture for node and edge deconvolutions to disentangle latent factors.
result Demonstrates effectiveness of the proposed model and its extensions on synthetic and real-world datasets.
New framework learns disentangled causal representations from observed labels.
problem Learning meaningful disentangled causal representations from observed data.
method ICM-VAE framework using flow-based diffeomorphic functions and causal disentanglement prior.
result Induces highly disentangled causal factors and improves robustness.
Proposes a new method for disentangling data representations using topological analysis.
problem Learning disentangled representations for better model explainability and robustness.
method Integrates a multi-scale topological loss term into the training of deep learning models.
result Improves disentanglement scores compared to state-of-the-art methods.
A new method for disentangling action sequences improves model stability.
problem Challenges in unsupervised disentanglement learning due to incomplete theories and abstract notions.
method Introducing disentangling action sequences and a novel fractional variational autoencoder (FVAE) framework.
result FVAE improves the stability of disentanglement for action sequences.
GCAE uses density estimation to achieve reliable disentanglement in latent space.
problem Disentangled learning representations suffer from reliability issues.
method GCAE uses Gaussian Channel Autoencoder with Dual Total Correlation (DTC) to avoid the curse of dimensionality.
result GCAE achieves highly competitive and reliable disentanglement scores.
The paper connects disentanglement to manifold charts and commutativity.
problem Discovering local charts of the data manifold for disentanglement.
method Interpreting disentanglement as local charts of the data manifold and studying commutativity.
result Commutativity is a central property in disentanglement, as shown in manifold, group theoretic, and probabilistic frameworks.
Improved VAE learns disentangled representations with less supervision.
problem Learning disentangled representations is challenging.
method Semi-supervised disentanglement learning with label replacement.
result Significant improvement in disentanglement with minimal supervision.
New metric for disentangling multivariate representations, accounting for more complex entanglements.
problem Current disentanglement metrics fail to detect entanglements involving more than two variables.
method Partial Information Decomposition framework to analyze information sharing and propose a new disentanglement metric.
result The proposed metric correctly identifies entanglements in high-dimensional spaces.
Improved disentanglement of data factors using recursive training.
problem Current unsupervised disentanglement methods are inconsistent and fail to achieve levels of disentanglement seen in supervised approaches.
method Introduced PBT for VAEs, used UDR for heuristic scoring, and developed recursive rPU-VAE approach.
result Recursive training leads to robust disentanglement of data factors across multiple datasets.
We define and address the problem of unsupervised learning of disentangled representations on data generated from independent factors of variation. We propose FactorVAE, a method that disentangles by encouraging the distribution of representations to be factorial and hence independent across the dimensions. We show tha…
New method learns disentangled discrete representations using categorical variational autoencoders.
problem Learning disentangled representations from discrete latent spaces.
method Replaced standard Gaussian VAE with a categorical VAE to mitigate rotational invariance.
result Categorical distributions improve learning of disentangled representations.
Disentangled representations have recently been shown to improve fairness, data efficiency and generalisation in simple supervised and reinforcement learning tasks. To extend the benefits of disentangled representations to more complex domains and practical applications, it is important to enable hyperparameter tuning …
Method quantifies disentanglement of generative models using manifold topology.
problem Challenging and inconsistent measurement of disentanglement in generative models.
method Measures topological similarity of conditional submanifolds in learned representation.
result Method ranks models similarly to existing methods across multiple datasets.
A generative model with a disentangled representation allows for independent control over different aspects of the output. Learning disentangled representations has been a recent topic of great interest, but it remains poorly understood. We show that even for GANs that do not possess disentangled representations, one c…
C2VAE learns disentangled and coupled representations without prior knowledge.
problem Learning disentangled and coupled representations in latent space.
method Introduces C2VAE, a self-supervised VAE that factorizes posterior and uses Gaussian copula for dependencies. result Demonstrates strong effect in enhancing disentangled representation learning.
This paper critically examines unsupervised disentangled representation learning, revealing challenges and limitations.
problem The difficulty of unsupervised learning of disentangled representations and the challenges in evaluation metrics.
method Theoretical analysis and a large-scale experimental study covering 8 datasets and 14000 models.
result Well-disentangled models cannot be identified without supervision, and different evaluation metrics disagree on what constitutes disentanglement.
Simple method disentangles content and style from pre-trained vision models.
problem Learning interpretable features in visual representations.
method Probabilistic linear entanglement model and simple disentanglement algorithm.
result Method provably disentangles content and style features.
Unsupervised mesh disentanglement separates identity and pose.
problem Geometric disentanglement for 3D deformable models.
method CFAN-VAE architecture using conformal factor and normal features.
result CFAN-VAE achieves state-of-the-art performance on unsupervised geometric disentanglement.
Study evaluates scalability and real-world impact of disentangled representations.
problem Scalability and real-world impact of disentangled representations.
method New high-resolution dataset and architectures for disentangled representation learning.
result Disentanglement predicts out-of-distribution task performance.