AI upgrade: New algorithm could lead to cheaper, more accurate and light-weight AI models

A group of researchers has solved a foundational problem in machine learning, creating the first method for handling symmetric data that is guaranteed to be efficient in both computation and data needs. The main challenge is that AI can be easily confused by symmetry; for example, it might see a rotated molecule as a completely new object instead of recognizing it as the same structure.

These symmetries are important because they are some sort of information that nature is telling us about the data, and we should take it into account in our machine-learning models. We’ve now shown that it is possible to do machine-learning with symmetric data in an efficient way. — Behrooz Tahmasebi, an MIT graduate student and co-lead author.

While some current models like Graph Neural Networks can handle symmetry, researchers have not fully understood why they work so well. This MIT team says they took a different approach — they designed a new algorithm by combining mathematical concepts from algebra and geometry to create a system that can efficiently learn and respect symmetry.

This provably efficient method requires fewer data samples for training, which can improve a model's accuracy and adaptability. The researchers say their work could lead to the development of more powerful and less resource-intensive AI models for a wide range of applications, “from discovering new materials, to identifying astronomical anomalies and unraveling complex climate patterns.” The research was recently presented at the International Conference on Machine Learning.