Researchers at the Massachusetts Institute of Technology (MIT) have introduced an algorithm that allows artificial intelligence to effectively learn from symmetric data. The solution overcomes one of the key problems in machine learning — the misinterpretation of symmetry, where, for example, an inverted molecule is perceived as a new object rather than the same structure.
The new method combines mathematical concepts from algebra and geometry, allowing the model to not only recognize symmetry but also take it into account during training. According to MIT graduate student Behrouz Tahmasebi, the algorithm is demonstrably efficient in terms of both computational resources and the amount of data required. This means that models can achieve high accuracy using fewer training samples.
Unlike graph neural networks, which already partially work with symmetry, the new approach allows for a better understanding of the mechanisms underlying the processing of such data. The development can be applied in various fields, from materials science to astronomy and climatology.
The results of the study were presented at the International Conference on Machine Learning. The authors believe that their work will contribute to the creation of more accurate and less resource-intensive AI models for scientific and applied tasks.