Institute Homepage
Institute Homepage EN Sign In

Back

Conference Paper 2020

Counterfactuals uncover the modular structure of deep generative models

{Deep generative models can emulate the perceptual properties of complex image datasets, providing a latent representation of the data. However, manipulating such representation to perform meaningful and controllable transformations in the data space remains challenging without some form of supervision. While previous work has focused on exploiting statistical independence to \textit\textbraceleftdisentangle\textbraceright latent factors, we argue that such requirement can be advantageously relaxed and propose instead a non-statistical framework that relies on identifying a modular organization of the network, based on counterfactual manipulations. Our experiments support that modularity between groups of channels is achieved to a certain degree on a variety of generative models. This allowed the design of targeted interventions on complex image datasets, opening the way to applications such as computationally efficient style transfer and the automated assessment of robustness to contextual changes in pattern recognition systems.}

Author(s): Besserve, M and Mehrjou, A. and Sun, R and Schölkopf, B
Book Title: Eighth International Conference on Learning Representations (ICLR 2020)
Year: 2020
Bibtex Type: Conference Paper (inproceedings)
Address: Addis Ababa, Ethiopia
Electronic Archiving: grant_archive

BibTex 

@inproceedings{item_3270699,
  title = {{Counterfactuals uncover the modular structure of deep generative models}},
  booktitle = {{Eighth International Conference on Learning Representations (ICLR 2020)}},
  abstract = {{Deep generative models can emulate the perceptual properties of complex image datasets, providing a latent representation of the data. However, manipulating such representation to perform meaningful and controllable transformations in the data space remains challenging without some form of supervision. While previous work has focused on exploiting statistical independence to \textit\textbraceleftdisentangle\textbraceright latent factors, we argue that such requirement can be advantageously relaxed and propose instead a non-statistical framework that relies on identifying a modular organization of the network, based on counterfactual manipulations. Our experiments support that modularity between groups of channels is achieved to a certain degree on a variety of generative models. This allowed the design of targeted interventions on complex image datasets, opening the way to applications such as computationally efficient style transfer and the automated assessment of robustness to contextual changes in pattern recognition systems.}},
  address = {Addis Ababa, Ethiopia},
  year = {2020},
  slug = {item_3270699},
  author = {Besserve, M and Mehrjou, A. and Sun, R and Sch\"olkopf, B}
}