a High-Performance Framework for Matrix Factorization Methods
SMURFF is an easy to use, high performance and flexible matrix factorization package. In this talk we will give a brief introduction to matrix factorization as a means of solving machine learning problems that are multi-task, with large training sets and scarce data. We will then outline the benefits of using SMURFF to solve them.
Virtual Screening on FPGA
Some ML problems require massive prediction throughput. Accelerators are useful in such cases to improve performance and reduce costs, but can be difficult to program. In this talk, we briefly describe such an ML problem from drug discovery, and then describe the effort required and the results achieved by implementing an inferencing method on an FPGA.
Thomas Ashby received his PhD from the University of Edinburgh in 2005, focusing on computational solvers for scientific computing simulations. After doing a postdoc on computer architecture, he joined Imec in Belgium in 2007. Since then he has done research on parallel programming tools, machine learning in vision, HPC, and large scale machine learning for the life sciences.