DeepMind has been getting a lot of press lately after the AI company created a language model to compete with the likes of OpenAI, Google, Facebook, and Microsoft. Now, they’re helping the field of chemistry. A team led by scientists from DeepMind has built a machine learning (ML) model that can guess how a molecule is made up by predicting how many electrons it has. While others have tried, DeepMind’s results are said to be more accurate than current techniques used for this problem.
Improving DFT with deep learning
In the past 30 years, density functional theory (DFT) has emerged as the most widely used electronic structure method to predict the properties of various systems in chemistry, biology, and materials science. Despite a long history of successes, state-of-the-art DFT functionals have crucial limitations. In particular, significant systematic errors are observed for charge densities involving mobile charges and spins. Kirkpatrick et al. developed a framework to train a deep neural network on accurate chemical data and fractional electron constraints (see the Perspective by Perdew). The resulting functional outperforms traditional functionals on thorough benchmarks for main-group atoms and molecules. The present work offers a solution to a long-standing critical problem in DFT and demonstrates the success of combining DFT with the modern machine-learning methodology. —YS
via ‘Pushing the frontiers of density functionals by solving the fractional electron problem‘ on Science.org
The successful results were achieved by training the model on 1,161 accurate solutions and included fundamental laws of physics into the network. Two of the scientists, James Kirkpatrick and Aron Cohen, said DeepMind is making the system open for anyone to try.
This is what I want to see ML used for and, on the surface, this looks like a remarkable achievement so far.
(via Nature.com)
Filed under: DeepMind machine learning neural networks