search

CompPhysVienna / n2p2

n2p2 - A Neural Network Potential Package
https://compphysvienna.github.io/n2p2/
GNU General Public License v3.0
217 stars 82 forks source link

Kalman Filter Optimizer #151

Open moabe84 opened 2 years ago

moabe84 commented 2 years ago

Hi. I am wondering if the (extended) Kalman filter optimizer can be applied to classification problems. I am interested in using a training code which is designed for molecular property prediction (both regression and classification problems can be considered). It uses PyTorch deep learning framework and the torch optimizers. From the N2P2 paper, it is clear that the Kalman filter performs much better than Adam and SGD. I am very curious to see if the Kalman filter optimizer can be used in that code and if it can improve the training performance. I also need to know if it's straightforward to write a python version of this optimizer. Many thanks.

Kyvala commented 2 years ago

Hi, it can. Check the article DOI: 10.3103/S1060992X14020088.