With more businesses approaching Multitel for help solving incredibly complex challenges, how could the organization scale to meet growing demand for world-leading AI research?
To enhance its machine learning and deep learning capabilities, Multitel implemented IBM Watson Machine Learning Accelerator—turbocharging the pace of innovation.
10x fastertraining of machine learning models accelerates research
Boosts productivityby empowering multiple researchers to use powerful GPU resources at the same time
Supports innovationin avionics, automotive manufacturing, medical science and industry 4.0
Business challenge story
Fulfilling revolutionary potential
Artificial Intelligence (AI) is allowing researchers to address some of the biggest and most urgent questions facing the world. For example, how to tackle climate change or find effective treatments for degenerative diseases.
While many organizations are interested in AI, few have the data science expertise and tools to use deep learning solutions to solve their business challenges. Enter Multitel—an ICT innovation center and a leader in machine learning research committed to helping companies across sectors reap the rewards of AI.
Jean-Yves Parfait, AI Team Leader at Multitel comments: “There is a lot of hype surrounding AI, which is making it more difficult for business leaders to have realistic expectations about how it could drive their business forward. We’ve made it our mission to help organizations realize the true potential of this emerging technology. We’ve been so successful that we’ve seen a steady increase in the number of companies commissioning AI research projects with us in recent years.”
Multitel wanted to ensure that it had the powerful, reliable and flexible IT systems in place required to support increasing demand for cutting-edge research.
Parfait continues: “Until a few years ago, we were able to implement our machine learning research projects with our existing cluster of machines, which was equipped with a single GPU. However, we had hit a performance wall for modern deep learning activities—limiting the pace of innovation. We knew that by going to multiple-GPU servers, we could dramatically increase the pooled compute resources available to our researchers and take on more client projects.”
To scale its AI capabilities, Multitel deployed IBM Watson Machine Learning Accelerator—a comprehensive suite of open source machine learning frameworks, a rich variety of AI development tools, and ultra-reliable IBM Power System AC922 servers equipped with four Tesla V100 GPUs.
Parfait says: “We thought of IBM straight away, due to their excellent reputation in the tech industry for AI innovation. With four integrated GPUs, IBM Watson Machine Learning Accelerator could resolve many of the bottlenecks that our machine learning experts encounter, by reducing latency and accelerating neural network training.
“What’s more, we were impressed that the IBM solution supports open source frameworks that the research community is familiar with. We saw an opportunity to use Docker containers to package each research project into self-contained units, enabling our researchers to share resources more effectively.”
Multitel engaged technical consultants from IBM Belgium to help with the implementation. The IBM team architected the solution to optimize performance, and then benchmarked it to ensure that it would deliver.
“We wanted to upgrade our machine learning systems as quickly as possible so that we could get back to focusing on research,” explains Parfait. “We set ourselves the target of deploying IBM Watson Machine Learning Accelerator in under eight weeks. The experts from IBM Belgium were instrumental in helping us to achieve this goal.”
New avenues for innovation
With IBM Watson Machine Learning Accelerator underpinning its AI capabilities, Multitel can accelerate research and dedicate more time to making breakthroughs that could change the world.
“Thanks to IBM Watson Machine Learning Accelerator we can now train machine learning and deep learning models up to 10 times faster—reducing the total training time from weeks to just days,” says Parfait. “Training our machine learning algorithms faster frees up more time for us to test and refine these models, which in turn enables us to speed up downstream development.”
By containerizing the AI workloads with Docker, Multitel can share resources between its team of researchers more evenly too. Parfait explains: “By accelerating machine learning training and sharing our resources across our team, we are much better prepared to handle increasing demand for our services.”
Established by the Engineering Faculty of the University of Mons, Multitel Innovation Center has led the charge in developing innovative engineering solutions using the latest innovations in science and technology for more than two decades. The non-profit organization works across industries to develop specialist solutions in signal and embedded systems, network engineering, applied photonics, computer vision and railway certification.