Machine Learning Conference 2018

By 18/12/2018Machine learning

2018 is almost over. But will still have some time to share yet another conference overview. This time it was a Machine Learning oriented conference. To be more precise the ‘ML Conference – The Conference for Machine Learning Innovation‘ held in Berlin from 5 to 7 of December, just a week ago. Obviously JArchitects could not let it go without being part of this event. Our Machine Learning expert was there, and now he shares his impressions about the conference.

It was a very decent event, and to be honest exceeded my expectations. The organizers divided the conference on keynotes, short and long sessions and workshops. Session speakers presented live demos and case studies of real-world problems. Mostly, they focused on the struggles they had to tackle to deliver the promises of Machine Learning. And to close the show a day of hands-on workshops given by experts.

There were many topics presented: frameworks & libraries, like Tensorflow, scikit-learn and Deeplearning4J; use case scenarios like job prediction, conversational AI with NLP, recommender systems and even ‘How to win a Kaggle competition’. Let’s walk through the highlights.

 

The keynotes

I’m not a big fan of keynotes. Usually they are quite marketing oriented (like the ones from Google and Amazon). So, you don’t get to learn anything from them. This time however, one has caught my attention. It was entitled ‘Making Enterprises Intelligent with Machine Learning’ and presented by SAP. Interesting enough, the presenter didn’t focus on SAP’s products, but rather on how to make Machine Learning a reality on enterprises customers. For that, he discussed the real big challenges companies have to handle.

On the second interesting keynote the presenter discussed AI as a commodity, and how consumers will benefit from the fact that many providers will probably battle to create better products and services. Providers need to keep looking for a way to stand out and differentiate themselves from the pack.

 

The sessions

Let’s start like this: I apologize for not been able to attend all session. Unfortunately, we cannot be in more than one place at time, yet. Having said that, let’s put out attention on the ones I’ve attended and liked the most.

 

‘Building smart apps with Machine Learning, from magic to reality’ (by Laurent Picard from Google)

This session was presented by a Google employee, and as usual, very marketing oriented. They are really good at making the presentation flawless even in live demos, like this one.

Mr. Laurent Picard showed us how to use some of the pre-trained models from Google, and also how to apply AutoML on specific scenarios without any AI or Machine Learning previous knowledge.

What a liked about it was the opportunity to see a few different Google API working on real time with almost no effort. Mr. Laurent Picard gave us some pointer on how to get started with Google APIs for machine learning. But once again it is a kind of ‘marketing of Google products’.

Takeaway: it is super easy to get started with machine learning nowadays.

 

‘Making a smart chat bot made smarter’ (by Aleksandra Vercauteren from Faktion)

In this short session Mrs. Aleksandra Vercauteren showed how difficult it can be to train smart chatbots that really understand the users’ intents, especially if your target language is not English.

She works mostly with Dutch language, for which the pre-trained word embeddings are not good enough. Therefore, she had to come up with innovative techniques to generate text examples to improve their NLP models.

 

‘The challenge of putting ML into practice’ (by Vladimir Rybakov from WaveAccess)

The words from Mr. Vladimir Rybakov: “… it is a common situation when specialists encounter problems they did not have while doing research … At the same time, companies that want to introduce ML in their business process, often do not realize what is needed for that to happen, what difficulties might occur and how to estimate the outcome of a project. We will look at two projects implemented in production. Based on those examples we will cover some of the problems you might encounter in your projects including the fields of data processing, algorithm selection, communication with the customer and other.”

The snippet above depicts exactly the scope of the session and what Mr. Vladimir has shared with the attendees.

 

‘Deep Recommendation Systems for real Personalization’ (by SK Reddy from Haxagon)

By far the most detailed slides and a very engaging presentation. Mr. Reddy charisma and energy were spot on. Well, he used some very clever tricks to keep everyone’s attention 🙂

 

 

This presentation was a mix of ‘clear explanation’ and ‘ technical/detailed discussion’ about machine learning models. Mr. Reddy showed us some beautiful and scary math equations. It was another session from which the attendees could learn some machine learning skills.

 

‘Winning a Kaggle competition in 2015’ (by Pieter Buteneers from Robovision)

Winning a Kaggle competition, who doesn’t want that? Do you think it is easy or even feasible? Well, Dr. Pieter Buteneers has shown us the way to do it.

On his own words he and his 6 colleagues have spent ‘about 80% of their time trying to solve the challenge and win it’. And they did it. So, yes, it is feasible. But it is a humongous task. Don’t be afraid though. The session was super interesting. He showed us how they have divided the work and how they have tackled each aspect of the challenge in a detailed way. The algorithms they have used and why. The variations, experiments and some nice discussion. There were too much to explore in 1 hour. You should definitely check it out. Get in contact with him (he is on twitter also @pieterbuteneers) and ask for the directions/links. He was so excited with the presentation that (I believe) he will gladly share the information.

 

 

‘How to build a job recommender SaaS with Deep Learning to disrupt the job market!’ (by Laurent Sorber from Radix.ai)

The best session by far. Why? Because not only Mr. Laurent Sorber presented the complete context of the application, but also the real nuts-and-bolts of their solution. How they evolved from the data collection task to the current solution step-by-step. Whenever Mr. Laurent will present another topic, I’ll be in the first line to watch and don’t miss a single detail. Thanks Mr. Laurent, this is the kind of presentation we expect in a technical conference. Something we can learn from.

‘Text classification and NLP on Starwars characters’ (by Natalie Beyer from LAVRIO.solutions)

 

In the last session I would like to bring to your attention the speaker brought to the stage some very familiar characters. It was also interesting to see her approach to explain how we can use toy datasets to explore nice ideas and machine learning techniques .

 

 

 

Last day, the workshop day

There were 3 workshops available. I’ve chosen ‘Deep Learning with TensorFlow’ by ML6 (https://ml6.eu/). A full day playing around with quite interesting dataset and TensorFlow. I’ve actually enjoyed the workshop and the experience, fast paced and hands-on approach. My only criticism is that Keras was used instead of TensorFlow directly. Why am I unhappy about his since Keras simplifies our lives in so many ways? because there was no mention of Keras on the title or the description of the workshop. So, the expectations were TensorFlow.

In the end, it was very well presented by the team from ML6.

 

Takeaway

It is super easy to get started with machine learning nowadays. But sometimes (most of the time, actually) you need much more than ‘getting started’. You need to dig deeper and come up with creative solutions to remove the hurdles from the path to succeed. At times what you need is creativity and experience. The super-ultra-mega trending and complex algorithm is not always the best solution, but rather some simple heuristic and strategy can deliver better results, or good enough results. Experiment, try out ideas and keep learning.