Pete & Martin recently took part in a 3-day Artificial Intelligence Hack at the Microsoft Reactor in London. Where developers, designers and architects can advance their AI skills on Azure by designing, building and testing their AI ideas and code, in a practical, hands-on environment. Here Martin tells us about the event:
"At Felinesoft we are just starting out and exploring AI and this event was fantastic for accelerating our learning. Pete & I went to the hack with an idea of what we wanted to achieve and some data in a database, but neither of us had much experience in the field.
After speaking to some of the experts, we were directed to some great documentation and GitHub repos with quick-start projects that closely matched part of what we wanted to achieve. This allowed us to be up and running with a skeleton App that could be used to perform model training on our data set in no time. We quickly learned that good data is 90% of the battle!
During the hack we explored the use of use of ML.NET for creating our recommendation/ personalisation system for recommending events to users in our userbase, based on either user interests/ metadata or collaborative filtering (specifically binary classifiers/field aware factorization) – and more importantly what approach would work in which context. We also explored the Azure Text Analytics API as a quick and easy way of condensing our event descriptions down into the main keywords to allow us to match user interests to topic keywords more easily.
We also attended ‘lightning talks’ and soaked up all the different tools, services and buzzwords that we had not been exposed to before – so learnt a great deal!"
"We’ve learned what would work in which scenarios and what would not – and have a far better understanding of the kinds of data we need to be capturing in order to be able to get the best possible models. We were also given some good ideas of how to extract user-specific data based on time spent on particular pages etc, which seems like a very interesting way of improving user models “behind the scenes”. We were able to train a decent model using “simulated data” – we just need to get real data that approaches the quality of our simulated data now!
The hack allowed us to pull together a proof of concept system based on simulated data and gave us a deep understanding of many of the issues surrounding ML – which will allow us to have a head start when we come to implementing something in earnest.
We are now in a position where we know what would work (and more importantly what wouldn’t, and why!). We also have a good knowledge of the tools and services out there that would help get us where we want to go. Overall, a fantastic event – thanks Microsoft!"
Martin Rhodes, Developer @ Felinesoft