In this great post: What sort of thing a brain is, Nate Soares makes a case for the need for human rationality, but I think that his view about the brain as a mutual-information machine is even more profound:

A brain is a specialty device that, when slammed against its surroundings in a particular way, changes so that its insides reflect its outsides. A brain is a precise, complex machine that continually hits nearby things just so, so that some of its inner bits start to correlate with the outside world.

A brain is a machine that builds up mutual information between its internals and its externals.

Is it turns out, this principle of mutual information will help us to automatically learn disentangled representations

Lo and behold, InfoGAN does exactly that, by building upon the framework of Generative Adversarial Nets: InfoGAN: Interpretable Representation Learning by Information Maximizing Generative Adversarial Nets

This paper describes InfoGAN, an information-theoretic extension to the Generative Adversarial Network that is able to learn disentangled representations in a completely unsupervised manner. InfoGAN is a generative adversarial network that also maximizes the mutual information between a small subset of the latent variables and the observation. We derive a lower bound of the mutual information objective that can be optimized efficiently. Specifically, InfoGAN successfully disentangles writing styles from digit shapes on the MNIST dataset, pose from lighting of 3D rendered images, and background digits from the central digit on the SVHN dataset. It also discovers visual concepts that include hair styles, presence/absence of eyeglasses, and emotions on the CelebA face dataset. Experiments show that InfoGAN learns interpretable representations that are competitive with representations learned by existing supervised methods.


In this paper, we present a simple modification to the generative adversarial network objective that encourages it to learn interpretable and meaningful representations. We do so by maximizing the mutual information between a fixed small subset of the GAN’s noise variables and the observations, which turns out to be relatively straightforward.


These results suggest that generative modelling augmented with a mutual information cost could be a fruitful approach for learning disentangled representations.

It has many desirable properties:

  • It does Representation Learning
  • In an unsupervised way
  • Its representations are interpretable
  • And disentangled
  • And the principle to push it to do that is conceptually elegant (based on maximising mutual information)

Note that a previous good attempt to build interpretable and disentangled representations is this one: Deep Convolutional Inverse Graphics Network, although the learning of the representations is done in a supervised fashion.