Inverse Learning of Symmetries
This addresses the challenge of modeling symmetries in domains like chemistry, where traditional methods fail, though it is incremental as it builds on deep information bottleneck techniques.
The paper tackles the problem of learning symmetry transformations in complex domains like chemical space where invariances are observable but not analytically describable, achieving state-of-the-art performance on artificial and molecular datasets.
Symmetry transformations induce invariances which are frequently described with deep latent variable models. In many complex domains, such as the chemical space, invariances can be observed, yet the corresponding symmetry transformation cannot be formulated analytically. We propose to learn the symmetry transformation with a model consisting of two latent subspaces, where the first subspace captures the target and the second subspace the remaining invariant information. Our approach is based on the deep information bottleneck in combination with a continuous mutual information regulariser. Unlike previous methods, we focus on the challenging task of minimising mutual information in continuous domains. To this end, we base the calculation of mutual information on correlation matrices in combination with a bijective variable transformation. Extensive experiments demonstrate that our model outperforms state-of-the-art methods on artificial and molecular datasets.