ML in Industry - Beyond Installing Packages
Building cutting edge production machine learning (ML), systems is hard. We shall discuss variety of rules of thumb that can improve the chances of success. We shall range from what is "good" code; researching ML; architectural design; devops burden and big company inertia.
Sadik Kapadia is a veteran of the machine learning world. In the late 1980s, early 1990s he made major contributions to the venerable HTK speech recognition toolkit which has won many DARPA trials. His recent work includes writing the production recommendations systems at Netflix which are publically acknowledged to be worth in excess of $500 million/year. In a testament to it's quality that even after 4 years of trying Netflix has not been able to improve them. Currently he is involved with new technology recommenders that require less than 1% of the data and have over double the lift of Netflix.