Yesterday was my third week in Munich and to join the community of the city’s machine learning peeps, I went to Kaggle Munich’s meetup, held at the Google office.
The agenda was:
- Listen to two talks
- Chat with people in the breaks
- Do some hands-on Kaggle hacking either by yourself, but preferably by creating a small team
Talks
The first talk was about the Vectorflow framework, which is a lightweight neural network framework implemented in the D language; and optimized for processing sparse data. Maybe someday I’ll get my hands on this while dabbling in a Kaggle competition - I’d like to see what advantages it would bring me.
The second talk was about Hierarchical Temporal Memory, which is an experimental approach to machine learning where the neurons are different than the usual neural network neurons most of us are used to. This approach is inspired by our neocortex and it could have some advantages for certain types of problems. I don’t think I will get my hands on this in the short term, but still, it was interesting to learn about this new way of thinking. The slides of this talk are here on github, if you’d like to check them out.
Hands on Kaggling
Although two of the talks were engaging, the most fun part of the meetup was the hands on Kaggle hacking session. I was curious about this Avito Demand Prediction Challenge competition so I started a team, by acting as the noob seed and then we became a small team of three, when two other Kagglers joined me. We started by taking a look at the existing kernels and checked out some exploratory data analysis posts and some model implementations. We then submitted a super basic prediction output - just to have a baseline result to improve later on :]
Verdict😛
All in all, this was a super fun and friendly meetup. I am looking forward to the next one and I absolutely recommend it to you too, if you happen to be around Munich at that time and are keen to spend a good 3-4 hours thinking of and with data.