Papers of the day   All papers

Hamiltonian Neural Networks


Sam Greydanus: My most recent @GoogleAI residency project is out: "Hamiltonian Neural Networks" Blog: Paper: Starting from noisy (pixel) data, we can learn _exact_ conservation of energy-like quantities.

7 replies, 836 likes

Alfredo Canziani: «Hamiltonian NN» uses a physical invariance (conservation of total energy) to reduce the variance of NN and get better generalisation for physical systems by training the net's *input gradients*! This, coupled with an (why not convolutional?) AE can be used directly on pixels!!!

3 replies, 77 likes

Kyunghyun Cho: a very nicely written blog accompanying the paper (h/t @alfcnz )

2 replies, 48 likes

John Platt: You can create a recurrent neural network out of a symplectic integration of a Hamiltonian, which learns physics from noisy examples. HT Leon Bottou at #NeurIPS . Based on by @samgreydanus

0 replies, 35 likes

Daisuke Okanohara: Hamiltonian Neural Network uses Hamiltonian mechanics as inductive bias and learns a Hamiltonian-like scalar function in an unsupervised manner, s.t the dynamics follow the gradients of the learned scalar function. Trains faster and generalizes better

0 replies, 22 likes

Sam Greydanus: [1/3] I’ll be in Vancouver at #NeurIPS2019 this week! I’m presenting my paper “Hamiltonian Neural Networks” (coauthors @jasonyo and Misko Dzamba) 👨‍🔬 Poster session (Thurs Dec 12, 5-7pm, Exhibition Hall B + C #67): Thread:

1 replies, 11 likes

Vincent Boucher 🕊: "Hamiltonian Neural Networks" Greydanus et al.: Blog: #Hamiltonian #NeuralNetworks #UnsupervisedLearning

0 replies, 10 likes

Jason Yosinski: @oh_that_hat @lanjanice @savvyRL 3/ Thurs 5:00 pm, poster #67. Hamiltonian Neural Networks, by @samgreydanus with Misko Dzamba. Bonus: shirt giveaway (designer: Misko)!

1 replies, 10 likes

Charles Fisher: @vijaypande Nice paper Vijay. Here's another physics + ML paper from a few weeks back that I really liked: Hamiltonian Neural Networks

0 replies, 2 likes

Miles Cranmer: 2/10 First, papers referenced in main tweet: LNNs - HNNs - Graph Nets - (+ refs therein...) Group-CNNs -

1 replies, 1 likes

Dimitris Kalatzis: @sam_power_825 Maybe close to what you're looking for (?), although probably not 100% A neural net learning a Hamiltonian vector field:

1 replies, 1 likes


Found on Jun 10 2019 at

PDF content of a computer science paper: Hamiltonian Neural Networks