Insights into Machine Learning at Jazoon

Jazoon is a tech conference in Zürich which manages to bring brilliant speakers to talk about a particular subject - on Friday, 27th October 2017 it was Artificial Intelligence.

Kudos must go to the organisers of the conference, Nithin Mathews, Dmitrij Burlak and Julia Imhof. The venue was excellent, the speakers were world-class, and they had spent a day with the speakers beforehand, showing our beautiful city of Zürich. It was really rewarding to hear the speakers wax lyrical about Switzerland.

Paige Bailey really impressed me. Very much at ease presenting, witty (I'm loud, I'm Texan, we all are loud), and a bad-ass coder.

One of the surprises for me was to learn that essentially, most of the artificial intelligence/Machine learning software is open source. Every big IT firm packages this open source software in their own environment, and most of them try to lock you in. Except for one. Microsoft. This made my head buzz. Microsoft? The arch-proprietors, the mean guys, they are cool now? I had the opportunity to chat with Paige and she waxed lyrical about Satya Nadella, who has managed to completely change the company culture in a surprisingly short amount of time.

One tool she used for presenting her ideas were Jupyter Notebooks, an impressive way to share code, interactive working code and demos within a browser, and easily transportable. Mind-blowing. Have a look at Microsoft's offering on https://notebooks.azure.com/

Another noteworthy insight was Microsoft's Project Pendelton, which applies machine learning to help data scientists clean up data (which is apparently what takes up most of their time and which they call 'data wrangling')

Christoph K?rner showed us deep neural nets working on the browser, and a particularly fun demonstration was Teachable machine, where you use your camera to teach a simple neural network to respond to video information - smiles, for example. Loads of fun, try it out!

Romeo Kinzler, IBM's Chief Data Scientist, took us through some really simple examples of how artificial intelligence works and then went on to explain how to apply that using Apache Spark, which is a huge distributed computing framework, which comes in handy when one needs big processing power, but sometimes is overkill. My jaw dropped as he off-handedly dropped the comment 'Oh yeah, this is node.red, which my team developed'. Wow.

Romeo showed this slide, from Techleer, which I found particularly informative to classify the different flavours of machine learning.

Apache Spark was also demonstrated in depth by Juliet Houghland. The one surprise was that the preferred standard for storing models is in PMML, which is an XML format. I thought that XML was by now completely uncool, but this variant is apparently the best. Another thought-provoking statement was her demonstration of using Apache Spark to train many slightly different models in parallel, instead of the usual use case of using a distributed network to deal with huge amounts of data.

?ke Forslund gave an impressive first-time-ever presentation and went full monty, by going against the common wisdom of not showing hardware, and not live-coding. But he did! He was presenting Mycroft.ai, which is an open-source voice assistant. I was delighted to hear that Mycroft gets its name from the sentient computer from my favourite book of all times, Robert Heinlein's The Moon is a Harsh Mistress. And the concept of an open-source voice assistant is reassuring to me. I spent my teenage years gobbling George Orwell's books, and, frankly, I find it's downright creepy to have one of Googles' or Amazon's little gadgets listening in to me.

Daniel Svonava showcased us Googles' open source data processing library. They have made a fantastic playground website which shows you how neural networks work. I can't recommend it enough, go and have a try: https://playground.tensorflow.org/

You can select your original dataset, add some neurons, add some layers, and then you press the play button and see how the network tweaks itself until it gets good results - the weighings are beautifully shown. A really effective learning tool to understand what these networks are.

This should give you an appetite to learn more. Nithin made a survey of the developers in the room, and almost none of us was already making money with this. I think that these technologies will really have an impact, now and in the future, and it's wise to learn up on them. It's not a hype! I invite you to watch the talks on video, as they were all posted on YouTube!