It’s a fact that computing hardware is now moving fast towards a new dimension with GPUs and these recent progresses in high performance computing (HPC) change the nature of machine learning research. Full exploitation of the available computing power implies finding new efficient ways to parallelize learning algorithms over many threads. Because of its recent breakthroughs, this includes novel efficient approaches to designing and training deep networks that leverages inexpensive GPU computing power. But the need for HPC in machine learning goes beyond deep architectures. For instance dictionary and kernel machine training could benefit from high performance sparse orthogonal and Cholesky factorization implementations. More generally, specific machine learning optimization algorithms have to be adapted for HPC, allowing new applications. Our vision is to widespread the use of GPU based machine learning to yet underexplored domain applications such that embedded real time urban environments perception, audio scene understanding and medicine/healthcare.
Deep in Normandy team brings together researchers from machine learning, image processing and intelligent vehicles from the NormaSTIC research federation. NormaSTIC concentrates the whole potential of public research in information and computer science in Normandy. It constitutes the only research structure in information sciences of the COMUE Normandie Université.
Researchers involved in Deep in Normandy belong to inter-disciplinary research groups, that include computer scientists, mathematicians, statisticians, physicists and engineers, working in close cooperation with application domain such as intelligent vehicles, image processing, medical imaging and signal processing. They are expert in machine learning with an international recognized experience in kernel machine, optimization issues (convex and non convex), signal processing. They have a strong experience in machine learning software development including public available Matlab toolboxes.