Hamiltonian Graph Networks with ODE Integrators
This advances the state-of-the-art in learned simulation, with potential applicability beyond physical domains, though it appears incremental as it builds on existing graph network and ODE integrator methods.
The paper tackled the problem of improving learned simulation models by imposing physically informed inductive biases, combining graph networks with a differentiable ODE integrator and Hamiltonian representation. The result was outperforming baselines in predictive accuracy, energy accuracy, and zero-shot generalization to unseen time-step sizes and integrator orders.
We introduce an approach for imposing physically informed inductive biases in learned simulation models. We combine graph networks with a differentiable ordinary differential equation integrator as a mechanism for predicting future states, and a Hamiltonian as an internal representation. We find that our approach outperforms baselines without these biases in terms of predictive accuracy, energy accuracy, and zero-shot generalization to time-step sizes and integrator orders not experienced during training. This advances the state-of-the-art of learned simulation, and in principle is applicable beyond physical domains.