Machine Learning demystified

You can’t escape the discussion about artificial intelligence these days. AI was also a big topic at this year’s Digital Tech Summit. A lot of keynotes, panels, and teams in the hackathon talked and used AI, machined learning, or deep neural networks to solve the big problems in climate change, health, or mobility.

All data generated by humankind until 2003 is now created every 10 minutes. That’s impressive. Even though I hate the expression ?data ist the new oil“. Data is an expression of the past. Without context, data can be misleading and often is just used to build a better version of the past than build for the future. Machine learning still struggles with context.

You recognize that this photo is upside down. An ML algorithm struggles. Upside down is contextual. Content is king, and context is kingdom.

For machine learning algorithms, it’s crucial to have a specific use case in mind. We did that at the Tech Summit in two amazing workshops about ML and how to get to know it in the IT-best-practice-way to learn anything: trial & error.

Christian Winkler showed us how to derive data-driven insights from user-generated content. It’s incredible how easy it is to derive business value with unbiased insights from open data like the one from Tripadvisor, from technical documentation, company wikis or scientific publications. You can thus back your decisions for product design, marketing, or category management by relevant data.

Philipp Baron Freytag von Lorginhoven got practical with us with the Microsoft’s Deep Learning Studio. You can get started if you have an O365 account right away. We had a fantastic time demystifying machine learning with a comprehensible use case: predicting the survival of Titanic passengers. Explore Kaggle if you also want to give it a try and get your hands dirty with machine learning.

Philipp’s tip searching for machine learning algorithm cheat sheets was great. Here is my favorite from Stefan Kojouharov.

We saw in both workshops how important it is to perform quality assurance with the whole data set. Approximately 60-70% of the time creating a machine learning solution deals with the quality of the data.

You can find Chrisitan’s and Philipp’s slide deck for the workshops in SlideShare.

Machine Learning? It’s not scary, it’s not magic… most time, it’s just math.

This article is part of a short series about my highlights of this year's Digital Tech Summit in Nuremberg. You find the links to all other articles here.