GFT’s insights from MLPrague 2018 conference

MLPrague is the biggest Central-European conference on applications of AI and machine learning. GFT had its representation there, so here’s what, in our view, were the most important trends and the most interesting talks in the 2018 edition.

Theme and presenters

MLPrague is about applications and trends in the industry. It manages to touch the sweet spot of focusing on tangible industry achievements without blurring the message with sales pitches or tech details. The notable presenters were: Google, Microsoft, DHL, CEAi, eBay, Wolfram, Adobe, BigML and PLOS – a good balance of platform vendors and practitioners. The most interesting speakers were Sepp Hochreiter, the inventor of LSTM networks and Jacob Biamonte, theoretical physicist working on quantum computers.


Microsoft’s talk was a strikingly honest account of applying AI in a non-obvious environments. CEAi’s talk on money laundering showed how it is more about process augmentation rather than replacing humans. O2’s presentation on Machine Learning in telco provided the most insights into how ML workflow can be embedded into established corporate infrastructure.

No longer a novelty

From the diversity of industries and organizations represented by speakers, one thing becomes apparent. AI is virtually everywhere today. From cell phones to smart buildings and from small, low profile non-profits to multinational corporations, it is ubiquitous. There were no discussions on what AI is and where it will take us. The community seems to be very much informed in this regard. Only methods, opportunities and return on investment were discussed this year. It was fascinating to see how in a year it all moved from praising the virtues of TensorFlow and cloud into assuming it is there. TensorFlow seems like Java in 2018.


Another “it is there” theme was the Cloud. Except for the Google talk, no-one really mentioned it explicitly, but those delving into tech details of their solutions seemed to usually use the Cloud, especially for image processing. Cloud vendors were represented by Google and Microsoft, but not AWS. The funny fact is Microsoft’s talk (the most likeable talk during the conference) seemed to explicitly favour TensorFlow over CNTK. The most often mentioned software platform used in the Cloud was, still, Spark.

Augmenting rather than replacing

The household fear of being replaced by robots was not really mentioned. Today’s applications of AI focus on helping people out. The common theme for multiple talks was enabling people to make correct, informed decisions quicker by using AI to provide better search results, extract relevant data from unstructured sources and enable insights into vast datasets impossible to reason from with purely human-based approach. This was most striking in the Microsoft talk which discussed applying AI in the somewhat tech-backward industry of non-profits.


This was not discussed directly much, as the Machine Learning community is more focused on applications, but what one could read between the lines was: Python and a little Scala, Spark, TensorFlow, CNTK, MxNet, Teradata Aster, Hortonworks and AWS, Google and Microsoft as major cloud providers for computing power. Platforms and languages seem to be pretty well established. The case is different for more specific libraries tackling problems of Natural Language Processing, time series, etc. There is a huge diversity there and no outstanding leaders. Several technologies seemed quite obscure.

Is it worth it?

Yes! This is a mature conference featuring a diverse set of industries and speakers. Most importantly, it is an important promoter of great potential of highly skilled Central-European data scientists. Also, the organization with single track, 30-minute-sharp talks and questions from the audience is something other conferences should perhaps learn from. See you next year!

Check out the videos of keynotes and panel discussions on MLprague’s website