Machine learning models are programs that challenge the assumptions of traditional software engineering and computer science. They differ in their construction: instead of being composed by humans, they are algorithmically inferred from data. They also differ in their domain: instead of dealing in other formalized software constructs, they deal in raw phenomena as represented in images, audio, and text. For practitioners, these differences lead to a profound rethink across engineering toolkit: from version control, testing, IDEs, modularity, and the purpose of computation itself. This talk will summarize the origin and nature of these challenges, the current state of the art, and the potential shape of a future synthesis.
Kovas is a senior engineer at Twitter Cortex, building infrastructure for Deep Learning. Previously he was the head of analytics at Weebly, and a senior research associate at Wolfram Research, working on Wolfram Alpha and Mathematica. His favorite language is Clojure.