Alex Honchar is Partner and Head of AI at Neurons Lab where he helps HealthTech and EnergyTech startups to accelerate R&D and product launch. Before he worked as an independent researcher with different startups and investment management firms and taught at the University of Verona, University of Copenhagen, Ukrainian Catholic University. Alex writes a blog on Medium with 1M+ views and is co-author of scientific articles with hundreds of quotations.
Today, when we say "AI" or "machine learning", most people immediately think of classic deep learning applications and results in computer vision, text understanding and generation, speech processing. One of the data types which don't look that sexy but are widely present in business is time series and digital signals. They are coming from the human bodies, machines, companies, and even the universe. Blindly applying state-of-the-art deep learning models to time series rarely produces great results. In this speech, I want to decompose different time series sources and related problems and explain the playbook, how to choose the right deep learning approach for time series to produce superior results that actually beat the standard benchmarks