Group-invariant tensor train networks for supervised learning
This work addresses the need for efficient group-invariant models in machine learning, particularly for problems like protein binding classification, but it is incremental as it builds on existing tensor network frameworks.
The authors tackled the problem of constructing group-invariant tensors for supervised learning by introducing a new numerical algorithm that is up to several orders of magnitude faster than previous approaches, and applied it to a protein binding classification problem, achieving prediction accuracy in line with state-of-the-art deep learning methods.
Invariance has recently proven to be a powerful inductive bias in machine learning models. One such class of predictive or generative models are tensor networks. We introduce a new numerical algorithm to construct a basis of tensors that are invariant under the action of normal matrix representations of an arbitrary discrete group. This method can be up to several orders of magnitude faster than previous approaches. The group-invariant tensors are then combined into a group-invariant tensor train network, which can be used as a supervised machine learning model. We applied this model to a protein binding classification problem, taking into account problem-specific invariances, and obtained prediction accuracy in line with state-of-the-art deep learning approaches.