Breaking the Black Box of Deep Learning

Long time no see, time for a new meetup! We have a very interesting theme about transparency and making sense of the otherwise black box of deep learning. Happy to have speakers from Intel, IBM, and I'll probably do an interactive demo from some work at Peltarion as well. Here are some more details about the speakers:

Justin Shenk develops and presents innovative AI projects internationally as an Intel Software Innovator. He previously studied biology and conducted neuroscience research in the USA. He is writing his master's thesis on machine learning at the Institute of Cognitive Science at the University of Osnabrueck in Germany and is creating interactive visualizations for the Institute of Mathematics.

Mikael Huss currently works as a senior data scientist at IBM in Stockholm. Prior to that, he has worked in academia (PhD 2007 from KTH) as a researcher and consulting bioinformatician, and for a couple of early-stage biotech startups as employee or data science consultant

Anders Arpteg, old guy that worked with AI for 15+ years in academia and industry, previously heading up a research group at Spotify and now works at Peltarion with democratization and industrialization of AI.

Tentative Agenda

18:00 Mingle time with food and drinks.

19:00 "How to visualize weights and activations in neural networks" by Justin Shenk, Intel

19:40 "Explaining predictions in a text classification scenario" by Mikael Huss, IBM

20:20 "Gradient-based feature importance for art classification" by Anders Arpteg, Peltarion

21:00 The end