Neural Petri flows for chemical reactions
The paper keeps Petri-net chemistry rules fixed and lets the neural part learn only the firing rate.
The authors argue that learned reaction models often lose the semantics that make Petri nets chemically meaningful. Their Neural Petri Flow hard-wires conservation, locality, and enabling rules, then uses learned rate laws or readouts on top. They report strong results across atom mapping, product prediction, enzyme classification, and elementary-step prediction, including valid top-1 molecule outputs without a filter. Source: ArXiv · AI/CL/LG's note.
The authors argue that learned reaction models often lose the semantics that make Petri nets chemically meaningful. Their Neural Petri Flow hard-wires conservation, locality, and enabling rules, then uses learned rate laws or readouts on top. They report strong results across atom mapping, product prediction, enzyme classification, and elementary-step prediction, including valid top-1 molecule outputs without a filter. Source: ArXiv · AI/CL/LG's note.
score 4