After some intense conference days we want to share our key takeaways from deep learning world and predictive analytics world 2018:
1. The deep learning community in Germany is growing
but it is still a small and welcoming family. The deep learning world conference reminded us a lot of the PAW conference five years ago: a discipline in it's early days, but plenty of good reasons to believe it will become a huge thing.
2. While deep learning is hyped for many areas, it's really important to look behind all the buzz. For time series prediction and tabular data more classical ML methods often perform on par
or even better (as highlighted by Prof. Sven Crone
, Dr. Michael Allgöwer
and Dr. Szilard Pafka
) and this perception seems to manifest in the community. On the other hand, for image, audio and text data deep learning opened up a whole new spectrum of use cases
with unseen performance (e.g. presented cases by Bilfinger & Microsoft, Dida).
3. With the increasing maturity of ML applications we need to think more about security of deployed models
. While avoiding SQL injections is a standard, we lack awareness for similar attacks on ML models as highlighted by Calvin Seward
4. The AI hype results in more and more corporates fearing of missing out. However, rushing into dozens of POC projects is not the solution and can lead to the POC trap
with not one of them making it into production. As pointed out by Norbert Wirth
, you need to solve a clear objective and an interdisciplinary team (including product owners from the business side).
We enjoyed all the brain food and conversations a lot and are excited to see the community and maturity of ML / NLP in Germany growing rapidly.
Looking forward to next year!