High Performance AI Systems

HIGH PERFORMANCE ARTIFICIAL INTELLIGENCE SYSTEMS RESEARCH TEAM

LEAD PI: SATOSHI MATSUOKA 

PROJECT MENTOR(S): Aleksandr Drozd and Emil Vatai

The HPAIS team is pursuing the convergence of high performance computing (HPC) and artificial intelligence (AI) in multiple aspects. One of the projects we would like to invite the candidates to participate in is an ongoing cross-institutional effort to develop and improve large scale natural models and methods for natural language processing (NLP). This research interfaces into adjacent areas like extending NLP methods to scientific data (e.g. chemical reactions in textual representation), understanding computer programs and mathematical theorems. Some of the results of our work are listed at the project website: https://vecto.space/

Students accepted to work on this project will participate in: (i) the development of novel types of language models; (ii) improving performance and scalability of existing language models (up to being able to use more than a hundred thousand nodes of Fugaku supercomputer); (iii) applications of NLP and other deep learning methods to real applications of different life and science domains.

The relevant technologies are still evolving, and the convergence of HPC and AI still has a long way to go, therefore the potential for the project to become a long-term collaboration is very high.

We expect the materials, methods and outcomes of this research to be aimed to the goals of peace, fairness and, in general, to the benefit of humanity as a whole.

REQUIRED SKILLS

Requirements for the internship can be categorized under: HPC and AI. The candidate should have strong experience in at least one of these fields and enough knowledge in the other one to be able to communicate efficiently. HPC requirements include a basic understanding of concepts related to parallel and distributed computing to experience with supercomputers and HPC systems, such as schedulers, I/O, MPI etc. AI requirements cover familiarity with deep learning frameworks, experience in NLP and/or theorem proving.

COMPUTATIONAL RESOURCES

Participants will be able to work with Fugaku supercomputer - world’s fastest machine as of November 2020, as well as smaller test-bed machines and clusters used by HPAIS team if needed.