AI Learns To Work Backward and Reveal Hidden Forces in Nature

University of Pennsylvania researchers developed Mollifier Layers, an AI method for solving inverse partial differential equations by inferring hidden causes from observed patterns.

The approach uses neural networks and automatic differentiation to work backward from data, helping reveal unseen processes in complex systems.

It could reduce the cost and difficulty of studying hard-to-model problems in biology, materials science, climate, aging, and related fields.

The method was published in Transactions on Machine Learning Research and is slated for presentation at NeurIPS 2026.