A team of researchers has solved a fundamental problem in machine learning, creating the first way to process replica data with guaranteed efficiency in terms of computations and the amount of data required.

The difficulty is that AI is often confused when confronted with homology; for example, it might treat a role molecule at a certain angle as a completely different object rather than recognize it as a single structure.Co-researcher Behraz Tahmasebi, a graduate student at MIT, emphasizes: "These symmetries carry important information that nature tells us about data, and we should take it into account in machine learning models. We have now established that it is possible to handle replica data with high efficiency.