Hamiltonian Neural Networks
This work addresses the challenge of improving inductive biases in neural networks for physics-based learning, though it is incremental as it builds on Hamiltonian mechanics.
The paper tackled the problem of neural networks struggling to learn basic physics laws by introducing Hamiltonian Neural Networks, which learn and respect exact conservation laws in an unsupervised manner, resulting in faster training and better generalization compared to regular neural networks on tasks like the two-body problem and pendulum observations.
Even though neural networks enjoy widespread use, they still struggle to learn the basic laws of physics. How might we endow them with better inductive biases? In this paper, we draw inspiration from Hamiltonian mechanics to train models that learn and respect exact conservation laws in an unsupervised manner. We evaluate our models on problems where conservation of energy is important, including the two-body problem and pixel observations of a pendulum. Our model trains faster and generalizes better than a regular neural network. An interesting side effect is that our model is perfectly reversible in time.