Automating deep learning workflows, lessons from 100 models

Deep learning is a very exciting topic and will continue to be in the future. Having set new benchmarks on text, speech, image, and video analysis a lot of data scientists are looking to using deep learning technologies. One of the main issues with deep learning models is limiting the amount of custom design and development. Ben will discuss some of the best practices for building models quickly that he has learned over the years. He will also discuss what is on the horizon for improving deep learning and where the future improvements will come from.

Ben Taylor has over 13 years of machine learning experience. He has worked for 5 years in the semiconductor industry for Intel and Micron in photolithography, process control, and yield prediction. He has also worked as a Wall Street “quant” building sentiment stock models for a hedge fund trading the S&P 1500 on the news content. During that time Ben helped build a 600 GPU computing cluster from the ground up that he used to backtest up to 10M trading scenarios per day. Ben left finance and semiconductor to work for a new HR start-up called HireVue in 2013 and lead their machine learning efforts around digital interviewing. His greatest accomplishment has been developing the features and methods which have allowed short unstructured video recorded interviews to see r-values in the 0.3-0.4 range in the HireVue insights product. He has a M.S. in chemical engineering from the University of Utah where he is currently finishing his Ph.D., also in chemical engineering.