Deep symbolic regression

Using a suitable reward mechanism with the reinforcement learning to identify the correct mathematical expression from an exponentially growing space of expressions that may describe a given dataset.

Objective

Symbolic regression is a challenging task in both physics and artificial intelligence, requiring the identification of a symbolic expression that accurately fits unknown function data. Although deep neural networks have shown remarkable performance in solving complex tasks, developing deep learning methods for symbolic regression remains an open issue. The key challenge in symbolic regression is the exponential growth of the space of mathematical expressions, which is both discrete (in model structure) and continuous (in model parameters). To find the optimal solution within this space, an appropriate reward mechanism is needed to guide the model in identifying the correct mathematical expression. It means reinforcement learning is a potential solution for this problem.