Symbolic Mathematics Finally Yields to Neural Networks | Quanta Magazine

Symbolic Mathematics Finally Yields to Neural Networks | Quanta Magazine

Symbolic maths yields to neural networks

Deep learning lets computers work out statistical patterns in massive amounts of data. It is a more brute-force approach than symbolic AI, in which programmers encode explicit rules in their algorithms. Now, two computer scientists have shown that deep learning can handle mathematical symbols, too. Borrowing techniques from automated translation, they taught their neural networks to solve mathematical problems, such as integration. “Mathematicians will in general be very impressed if these techniques allow them to solve problems that people could not solve before,” says mathematician Anders Hansen. Some hope that a similar strategy could enable computers to find their own mathematical proofs.

Quanta | 8 min readSource: arXiv preprint