Disentanglement through Nonlinear ICA with General Incompressible-flow Networks (GIN)

A central question of representation learning asks under which conditions it is possible to reconstruct the true latent variables of an arbitrarily complex generative process. Recent breakthrough work by Khemakhem et al. (2019) on nonlinear ICA has answered this question for a broad class of conditional generative processes. We extend this important result in a direction relevant for application to real-world data. First, we generalize the theory to the case of unknown intrinsic problem dimension and prove that in some special (but not very restrictive) cases, informative latent variables will be automatically separated from noise by an estimating model. Furthermore, the recovered informative latent variables will be in one-to-one correspondence with the true latent variables of the generating process, up to a trivial component-wise transformation. Second, we introduce a modification of the RealNVP invertible neural network architecture (Dinh et al. (2016)) which is particularly suitable for this type of problem: the General Incompressible-flow Network (GIN). Experiments on artificial data and EMNIST demonstrate that theoretical predictions are indeed verified in practice. In particular, we provide a detailed set of exactly 22 informative latent variables extracted from EMNIST.

[1]  Geoffrey E. Hinton,et al.  Distilling the Knowledge in a Neural Network , 2015, ArXiv.

[2]  Yoshua Bengio,et al.  Generative Adversarial Nets , 2014, NIPS.

[3]  Ryan P. Adams,et al.  Composing graphical models with neural networks for structured representations and fast inference , 2016, NIPS.

[4]  Aapo Hyvärinen,et al.  Nonlinear independent component analysis: Existence and uniqueness results , 1999, Neural Networks.

[5]  Yoshua Bengio,et al.  NICE: Non-linear Independent Components Estimation , 2014, ICLR.

[6]  Bernhard Schölkopf,et al.  Challenging Common Assumptions in the Unsupervised Learning of Disentangled Representations , 2018, ICML.

[7]  Emilien Dupont,et al.  Joint-VAE: Learning Disentangled Joint Continuous and Discrete Representations , 2018, NeurIPS.

[8]  Kilian M. Pohl,et al.  Variational Autoencoder with Truncated Mixture of Gaussians for Functional Connectivity Analysis , 2019, IPMI.

[9]  Aapo Hyvärinen,et al.  Unsupervised Feature Extraction by Time-Contrastive Learning and Nonlinear ICA , 2016, NIPS.

[10]  Pierre Comon,et al.  Independent component analysis, A new concept? , 1994, Signal Process..

[11]  Max Welling,et al.  Auto-Encoding Variational Bayes , 2013, ICLR.

[12]  Jimmy Ba,et al.  Adam: A Method for Stochastic Optimization , 2014, ICLR.

[13]  Pieter Abbeel,et al.  Flow++: Improving Flow-Based Generative Models with Variational Dequantization and Architecture Design , 2019, ICML.

[14]  Murray Shanahan,et al.  Deep Unsupervised Clustering with Gaussian Mixture Variational Autoencoders , 2016, ArXiv.

[15]  Aapo Hyvärinen,et al.  Nonlinear ICA of Temporally Dependent Stationary Sources , 2017, AISTATS.

[16]  Gregory Cohen,et al.  EMNIST: an extension of MNIST to handwritten letters , 2017, CVPR 2017.

[17]  Ullrich Köthe,et al.  Guided Image Generation with Conditional Invertible Neural Networks , 2019, ArXiv.

[18]  Aapo Hyvärinen,et al.  Nonlinear ICA Using Auxiliary Variables and Generalized Contrastive Learning , 2018, AISTATS.

[19]  Christopher Burgess,et al.  beta-VAE: Learning Basic Visual Concepts with a Constrained Variational Framework , 2016, ICLR 2016.

[20]  Prafulla Dhariwal,et al.  Glow: Generative Flow with Invertible 1x1 Convolutions , 2018, NeurIPS.

[21]  Samy Bengio,et al.  Density estimation using Real NVP , 2016, ICLR.

[22]  Pieter Abbeel,et al.  InfoGAN: Interpretable Representation Learning by Information Maximizing Generative Adversarial Nets , 2016, NIPS.

[23]  Aapo Hyvärinen,et al.  Variational Autoencoders and Nonlinear ICA: A Unifying Framework , 2019, AISTATS.

[24]  Roger B. Grosse,et al.  Isolating Sources of Disentanglement in Variational Autoencoders , 2018, NeurIPS.