August 29, 2016

Facebook to Accelerate Global AI Research with New GPU Program Recipients

By: Ari Entin

Today, Facebook is announcing new recipients to the GPU Partnership Program announced earlier this year.

This program, fashioned with the goal of helping others make faster progress in the field of AI, is intended to overcome a large challenge, ensuring the world’s top researchers have the infrastructure, tools and techniques to solve some of the largest technical problems.

To help accomplish this, we are distributing 22 high-powered GPU servers to 15 world-class research groups across 9 European countries. The Facebook AI Research (FAIR) team will also work with recipients to ensure they have the software required to run the servers as well as directly collaborate with them in their ongoing research efforts.

FAIR is well known for its open research efforts, and the insights from the GPU program are intended to be shared with the global scientific community. “There is a tremendous diversity of work happening across AI research and we believe that providing these talented teams of researchers with the technology they need to aid their efforts will deliver faster progress and drive innovation across the industry,” said Yann LeCun, Director of Facebook AI Research.

Recipients research focus areas range from computer vision to learning systems, deep neural networks and a host of technical challenges in between. Making progress in these areas will propel the entire AI field forward, enabling new services and solutions across a range of issues ranging from science, health care, language and the ability to intelligently automate tasks.

The recipients are:

IST Austria

KU Leuven, EE Department
“The EE department at KU Leuven is grateful and proud to receive Facebook’s support. No doubt, it will greatly boost the research of our staff and students, working on diverse topics from machine learning theory, over novel vision and speech applications, to hardware implementations.”

Czech Republic
Czech Technical University in Prague, Cybernetics, Faculty of Electrical Engineering
“The Visual Recognition and Machine Learning groups of the Center for Machine Perception at the Czech Technical University in Prague are excited to participate in the Facebook AI Research Partnership program. The new GPU servers provided within the program will speed-up our research in topics ranging from visual recognition and search to theoretical analysis of relations between graphical models and deep networks. We are looking forward to the collaboration with FAIR, a research establishment that attracted many top-notch scientists.”

Brno University of Technology, Department of Computer Graphics and Multimedia
“We are performing top research and education in human-computer and human-human interaction. The donated GPU servers will enhance DCGM’s computing power helping the research groups at DCGM to perform data-intensive computations in the areas of speech data-mining, image and video processing, computer graphics, knowledge extraction from big data, and computational photography.”

École normale supérieure (ENS), Département d’Informatique

The GPU servers donated by Facebook will allow the THOTH research team at INRIA Grenoble to speed-up their research on large-scale machine learning and computer vision. “We are already actively collaborating with FAIR on the topic of semantic image segmentation, and this gift strengthens our partnership with FAIR. It will in particular extend our capability to do research on memory and compute intensive generalized convolutional network architectures, and applications of visual recognition in video,” said Jakob Verbeek.

University Pierre and Marie Curie, LIP6 Machine Learning and Information Access Lab
“This partnership is decisive today for testing at large scale our models and algorithms” said Dr. Denoyer. The donation will help to advance our theoretical understanding of deep architectures, for our research on Reinforcement/Unsupervised/Generative learning. Dr. Cord, whose interests include computational cuisine, is fully convinced by this collaboration: “To fight against junk-food, we need to reconnect people with the act of cooking; smart mobile apps to suggest recipes from ingredient or plate pictures can inspire people to give it a try!”

University of Freiburg, Computer Science, Computer Vision Group

University of Tübingen, Center for Integrative Neuroscience
“We are highly delighted about the partnership with FAIR and the interaction with their researchers. The additional computational resources will help us to reach our goal of understanding the brain and constructing machine intelligence. We very much appreciate the support of Facebook and their openness to academia in general.”

Technical University of Munich (TUM), Informatics and Visual Computing
“Our AI research is centered around computer graphics and vision, dealing with the visualization and analysis of complex three- and higher-dimensional data. The partnership program will allow us to develop faster and better algorithms, for example, in physically-based animation and image-based biomedical diagnostics by using new machine learning techniques.”

University of Modena and Reggio Emilia
“We are proud to have the possibility to collaborate with Facebook AI research labs in these projects. The GPU servers received by Facebook AI research together with the HPC computer facilities of CINECA supercomputer labs will allow us to reduce the computational time, dramatically. Neural networks have proven to be a best way to add the sight sense to computers but they need, as the humans do, days and days of training. With the specialized hardware our research will compete with the research of the best labs over the world.”

University of Edinburgh

University of Cambridge, ML and Computer Vision
“The partnership between the Machine Learning Group and Facebook will benefit our research enormously. The GPU servers will help us apply new complex algorithms to challenging datasets, including those from speech and computer vision. They will also allow us to extend our existing software packages that automate aspects of probabilistic inference to leverage these resources. This will benefit the machine learning community at large.”

Moscow Institute of Physics and Technology (MIPT), Neural Networks and Deep Learning
“Our lab is focusing on the areas where recent breakthroughs were achieved mainly because of the progress in hardware: neural dialogue systems, reinforcement learning, and their combination. This partnership will allow us to significantly reduce the turnaround time in our experiments and thus accelerate the progress in our research much faster closing the loop from research to market. Overall, it will lead to a sooner availability of human level AI agents for general public.”

ETH Zurich, Data Analytics Lab
“The partnership with Facebook AI research will help us to pursue our ideas with much higher efficiency, as learning deep models requires computationally intensive data processing. This is in particular true for recurrent netwoks and attention-driven models that have shown great promise in this area. The new servers will help us speed up our research cycles and make progress faster” said Dr. Hofmann.