Hamiltonian Neural Networks

Ayush Garg, Sammed Shantinath Kagi

Abstract: In today’s world, neural networks are being in almost every discipline resulting in significant improvement in all the tools and applications. But in the field of Physics, they struggle to attain the basic laws like conservation of momentum. The paper Hamiltonian Neural Networks addresses this issue by using Hamiltonianmechanics to train the neural network in an unsupervised method. The attached report is an explanation of the paper and the code to reproduce the claimed results.

Track: Replicability

NeurIPS Paper Id: https://openreview.net/forum?id=B1f2TrBxUr