Implementing ML Solutions While Avoiding Traps

It is all-to-easy for someone who is just interested in the results to look for and use existing code that can help with the solution. The higher-level the better. This raises the problem of tying one’s future to a specific set of techniques. In this talk we will look at implementing Machine Learning  solutions which take advantage of the great innovations in hardware without binding ourselves to a specific technology which might go out of fashion or gets abandoned.

Required audience experience

General audience but developers will appreciate more details

Objective of the talk

To make clear that bridging the gap between the great pieces of hardware, bought or produced (FPGA), and an ML solution does not and should not require software that cannot be controlled fully. Proprietary stacks or abandonware can and will cause problems. Judging risk vs cost is needed.

You can Ulrich’s slides and presentation below:

Ulrich Drepper M3 presentation


Track 3
Location: Stephenson Date: October 15, 2018 Time: 12:35 pm - 1:20 pm Ulrich Drepper, Red Hat