Open AccessProceedings Article
Evaluating the Disentanglement of Deep Generative Models through Manifold Topology
Sharon Zhou,Eric Zelikman,Fred Lu,Andrew Y. Ng,Gunnar E. Carlsson,Stefano Ermon +5 more
- 03 May 2021
TL;DR: This article proposed a method for quantifying disentanglement that only uses the generative model, by measuring the topological similarity of conditional submanifolds in the learned representation.
read more
Abstract: Learning disentangled representations is regarded as a fundamental task for improving the generalization, robustness, and interpretability of generative models However, measuring disentanglement has been challenging and inconsistent, often dependent on an ad-hoc external model or specific to a certain dataset To address this, we present a method for quantifying disentanglement that only uses the generative model, by measuring the topological similarity of conditional submanifolds in the learned representation This method showcases both unsupervised and supervised variants To illustrate the effectiveness and applicability of our method, we empirically evaluate several state-of-the-art models across multiple datasets We find that our method ranks models similarly to existing methods We make our code publicly available at https://githubcom/stanfordmlgroup/disentanglement
read more
Chat with Paper
AI Agents for this Paper
Find similar papers on Google Scholar, PubMed and Arxiv
Write a critical review of this paper
Analyze citations of this paper to find unaddressed research gaps
Citations
•Posted Content
Metrics for Exposing the Biases of Content-Style Disentanglement
Xiao Liu,Spyridon Thermos,Gabriele Valvano,Agisilaos Chartsias,Alison O'Neil,Sotirios A. Tsaftaris +5 more
- 27 Aug 2020
TL;DR: This paper identifies key design choices and learning constraints on three popular models that employ content-style disentanglement and derive ablated versions and proposes metrics to measure how (un)correlated, biased, and informative the content and style representations are.
15
•Posted Content
On Disentangled Representations Learned From Correlated Data
Frederik Träuble,Elliot Creager,Niki Kilbertus,Francesco Locatello,Andrea Dittadi,Anirudh Goyal,Bernhard Schölkopf,Stefan Bauer +7 more
TL;DR: In this article, the authors analyzed the behavior of disentanglement approaches on correlated data in a large-scale empirical study (including 4260 models) and showed that systematically induced correlations in the dataset are being learned and reflected in the latent representations, which has implications for downstream applications such as fairness, and demonstrated how to resolve these latent correlations, either using weak supervision during training or by post-hoc correcting a pre-trained model with a small number of labels.
8
•Posted Content
Measuring the Biases and Effectiveness of Content-Style Disentanglement
Xiao Liu,Spyridon Thermos,Gabriele Valvano,Agisilaos Chartsias,Alison O'Neil,Sotirios A. Tsaftaris +5 more
TL;DR: In this paper, the authors investigate the role of different biases in content-style disentanglement settings and unveil the relationship between the degree of disentangling and task performance.
4
•Posted Content
Interventional Assays for the Latent Space of Autoencoders.
TL;DR: In this paper, the authors propose a framework called latent responses for probing the learned data manifold using interventions in the latent space, and evaluate how their analyses improve the quality of the generated samples using the VAE on a variety of benchmark datasets.
2
•Posted Content
Structure by Architecture: Disentangled Representations without Regularization
Felix Leeb,Guilia Lanzillotta,Yashas Annadani,Michel Besserve,Stefan Bauer,Bernhard Schölkopf +5 more
TL;DR: In this article, a self-supervised structured representation learning method using autoencoders for generative modeling is proposed, which relies solely on the independence of latent variables and avoids the tradeoff between reconstruction quality and generative performance inherent to VAEs.
1
References
•Proceedings Article
Domain Adaptation for Large-Scale Sentiment Classification: A Deep Learning Approach
Xavier Glorot,Antoine Bordes,Antoine Bordes,Yoshua Bengio +3 more
- 28 Jun 2011
TL;DR: A deep learning approach is proposed which learns to extract a meaningful representation for each review in an unsupervised fashion and clearly outperform state-of-the-art methods on a benchmark composed of reviews of 4 types of Amazon products.
Barcodes: The persistent topology of data
TL;DR: In this paper, a survey of the use of algebraic topology for feature detection and shape recognition in high-dimensional data is presented. But the main focus of the survey is on the application of topology to the classification of natural images.
•Proceedings Article
Challenging Common Assumptions in the Unsupervised Learning of Disentangled Representations
Francesco Locatello,Stefan Bauer,Mario Lucic,Gunnar Rätsch,Sylvain Gelly,Bernhard Schölkopf,Olivier Bachem +6 more
- 24 May 2019
TL;DR: The authors show that the unsupervised learning of disentangled representations is fundamentally impossible without inductive biases on both the models and the data, and suggest that future work on disentanglement learning should be explicit about the role of inductive bias and (implicit) supervision.
•Posted Content
BEGAN: Boundary Equilibrium Generative Adversarial Networks
TL;DR: This work proposes a new equilibrium enforcing method paired with a loss derived from the Wasserstein distance for training auto-encoder based Generative Adversarial Networks, which provides a new approximate convergence measure, fast and stable training and high visual quality.
1.2K
Barycenters in the Wasserstein Space
Martial Agueh,Guillaume Carlier +1 more
TL;DR: This paper provides existence, uniqueness, characterizations, and regularity of the barycenter and relates it to the multimarginal optimal transport problem considered by Gangbo and Świech in [Comm. Pure Appl. Math., 51 (1998), pp. 23–45].