Intel expands AI research cooperation with Mila, the worlds largest deep learning research institute

Intel announced three-year strategic research and joint innovation cooperation with Mila, AI research institute in Montreal, Canada. As part of the announcement, which expanded and renewed the scope of cooperation announced in April 2021, more than 20 researchers from Intel and Mila will focus on developing advanced AI technologies to solve global tasks such as climate change, new materials, and digital biology.


Joshua Begin Founder and Science General Manager said, We are facing various global tasks. We have great expectations for collaboration with Intel, which improves carbon capture, accelerates new drug discoveries, and quickly studies new materials needed to realize a more sustainable future.

In order to accelerate the research and development of advanced AI to solve the difficulties faced by the world, it requires a responsible approach to AI and the expansion of computing technology. Intel and Mila, who have a common goal to be a positive and powerful driving force to change the world as a leader of computing and AI, accelerate the project that started in 2021, add a third track, and lead to visible results. We plan to expand support.

Kavita Prasad, Vice President and Data Center, AI, Cloud Execution and Strategy Manager said, AI research is needed, he said. Today’s presentation will play an important role in providing core insights and promoting technological innovation to researchers. Intel is expected to work with Mila to solve the challenges today and to contribute to creating a better world for future generations.

AI-based new material discovery automation: Development of chemical simulation technologies such as Density-Functional Theory (DFT) enabled ways to simulate important characteristics of complex material systems. However, these technologies have limited the complexity of the material system that can be modeled, considering the increase in computing costs as the number of atoms increases. The AI technology, especially the graph neural network (GNN), is greatly reduced as the system size increases, helping to approximate chemical simulations. This uses AI-based simulation technology to open a huge possibility of replicating more complex materials systems. The potential discovery of new materials can contribute to cost and carbon footprints.

Intel and Mila cooperate in science and technological innovation development to improve the performance of atomic simulation GNN, such as the Open Catalyst Data Set. This effort can enhance the relevant technology pipeline to democratize the competence of researchers related to atomic material data. The team will develop a learning-based framework that can effectively search in the vast search space of material design applications. These frameworks can draw ideas from reinforced learning, search algorithms, and creation models, as well as other machine learning algorithms, including the generated flow network pioneered by Mila.

Application of causal and machine learning for climate science: Standard physical-based climate model can help to predict the impact of climate change, but involves high complexity and computing costs. It takes several months in special super computing hardware environments, so the frequency of providing simulation and subdivided and localized predictions is inevitably reduced. In addition, such models cannot generally explain the reasoning or causal relationship that is the basis of prediction. Intel and Mill aims to build a new type of climate model emulator based on the phosphorus and machine learning. This method allows you to identify which variables are predicted in the high-level input of traditional climatic models. The project will accelerate the development of climate science through thorough and reliable predictions on the impact of climate change and directly provide information necessary for related policies.

Research on the molecular drivers of diseases and drug discovery: New drug discovery is a long process of $2.6 billion in approved drugs. Finding a small molecule that combines a specific object is a dangerous and very uncertain process that can take more than 10 years. In addition, even if a molecule is found, it may fail later. Intel and Mila researchers work together to faster and simply identify better drug candidates.

For example, predicting a complex phenotype, including a genotype based on a single-nucleotide polymorphism (SNP), is affected by many SNPs throughout the genome. For this reason, the complex phenotype prediction was a continuous challenge of digital biology. The main computing task is to use large population data to jointly learn the causal effects of all SNPs included in the genome on the phenotype. The exact solution has a search space that is proportional to the number of SNPs. As millions of SNPs are detected, the exact solution is difficult to deal with. However, with the availability of high resolution data, the breakthrough development of AI, and the improvement of computing density led by Moore’s law, Intel and Mila will develop AI technology and achieve the following goals.

Leave a Reply

Your email address will not be published.