NVIDIA AI Technology Center makes researchers better equipped to develop computationally demanding AI applications
NVAITC Finland has been now operational for one and half years. It has enabled academics to do their research more efficiently.
Artificial intelligence researchers have benefited from NVIDIA’s expertise in utilizing graphics processing units (GPUs) and AI software in AI applications. NVIDIA has also contributed to research carried out at FCAI.
“We have been nothing but impressed by the world-class research that is conducted at FCAI”, says Dr. Niki Loppi, AI & HPC Solution Architect at NVIDIA.
NVIDIA AI Technology Center (NVAITC) is a joint research center of the Finnish Center for Artificial Intelligence FCAI, NVIDIA, and the Finnish IT Centre for Science CSC. NVAITC Finland accelerates research, education and adoption of artificial intelligence in Finland.
NVAITC is supplying a considerable amount of computing power for AI researchers.
“GPU computing is a crucial enabling technology for AI, as the performance and energy efficiency of modern GPU architectures has made them invaluable tools for AI research. Collaborating with NVIDIA as part of NVAITC has given our researchers access to world class expertise in GPU computing, enabling new and optimized GPU based AI algorithms to be developed in FCAI research projects”, says Professor Keijo Heljanko from Department of Computer Science at University of Helsinki.
So far NVAITC have undertaken nine projects from all participating organisations, University of Helsinki, VTT and Aalto University. The topics have ranged across different AI fields, e.g. Gaussian processes, computer vision, generative modelling and natural language processing.
“I’ve thoroughly enjoyed all projects that we’ve worked on. For example with Professor Arno Solin and Doctor Will Wilkinson I worked on Spatio-temporal Variational Gaussian Processes, and the research was published in NeurIPS 2021. In this project I helped the researchers to improve various implementation aspects e.g. parallelization of certain state-updates and profiling and benchmarking the temporally parallel filtering/smoothing approach”, says Loppi.
Loppi also undertook a project with Professors Antti Honkela and Sami Kaski, and PhD student Lukas Prediger to work on their new Python Package for Differentially-private Probabilistic programming called D3P.
“I helped them to fix earlier performance issues on GPUs and to implement a GPU-optimized subsampler to draw minibatches in i.i.d fashion. This research was just accepted for publication in PoPETs,” says Loppi.
Right now Loppi is working with Professor Jörg Tiedemann’s group. He is helping the researchers to scale their codes to large supercomputer architectures e.g. Puhti-AI and Mahti-AI at CSC.
“Harnessing the compute capability of modern GPU architectures at scale is one of the core missions of the center, so we are very excited about this project”, says Loppi.
COVID and the remote working model has been a challenge for AI & HPC Solution Architect Loppi.
“We’ve had to adjust to the circumstances. COVID has prevented us from working collaboratively on campuses and meeting new researchers spontaneously on corridors and various in-person events. When the situation gets better and campuses truly re-open, we hope that we can reach even more researcher as nothing beats in-person engagements”, says Loppi.
Training and education are important tasks
NVAITC is not only about the collaborative projects but also about training and higher education.
“Last year we created a webinar series “Introduction to efficient deep leaning” and delivered it live to the FCAI community. This year we have organized a webinar series on “AI applications in Computational Sciences” as a part of the FCAI AIX forum, where invited domain experts to showcase how they use AI as part of their numerical workflows.
Academics at all levels can contact NVAITC and get help with challenges related to GPU computing and AI. Please visit: fcai.fi/nvaitc or contact Niki Loppi directly via nloppi@nvidia.com.
“We very much welcome early ideas from PhD students and postdocs, and we are always happy to help with the project submission”, says Loppi.