Data Science Ireland July 2019: What We Learned

Dsijuly

It was another packed house at the Data Science Ireland meetup this past Thursday evening, as we enjoyed more fascinating presentations with three top-class speakers on our return to Huckletree D2.

We kicked things off with Chief Data Scientist at Corvil Fergal Toomey who talked about Making Markets Faster, Safer and more Efficient. He discussed how Corvil uses data and machine learning to optimize the financial trading infrastructure. Recently acquired by Pico, an Infrastructure-As-A-Service offering for financial trading, it is a very exciting time at Corvil. 

Fergal explained how data tapped directly from networks can help operators overcome these difficulties. He described the characteristics of network data, the different ways in which it can be analyzed and some of the customer problems that Corvil solves using machine learning powered by network data.

One of the most interesting takeaways from Fergal's talk on the importance of forecasting and anomaly detection to allow users to gain a helping hand to foresee trouble spots. He also gave an insight into their work with root-cause analysis, which allows them to track orders and trade events occurring in multiple parts of a complex trading system.

Amelia Kelly, Director of Speech Technology at Soapbox Labs talked about the future of children's voice technology. Voice is set to replace typing, mouse clicks, touch and gesture as the dominant interface with technology, but these voice interfaces don't work well for children's speech.

Soapbox Labs have received plenty of plaudits for their work in speech technology over the past few years but it is not just developing their children's voice technology platform that is the challenging part, but also the need to be transparent with their data to ensure that they are building a platform that is accurate, appropriate and leading the field in data privacy.

Their platform is also being rolled out in an educational setting and help improve childhood literacy. Backed up by a Tech for Good partnership by Microsoft, SoapBox Labs are leading the way in children's voice technology and a perfect example of AI for good.

Finally, Machine Learning Engineering Leader at Hubspot Hector Urdiales Llorens introduced us to his fascinating work sharing how HubSpot mixes multiarmed bandits and deep learning to deal with exploration/exploitation trade-offs when there is no data yet.

We learned about simple trick "ε-greedy." Choose a value and when deciding with an ε probability get a random option amongst all the available one. With a 1-ε probability, choose the one that has performed the best so far. However, you may be using cMAB unknowingly!

If you’re using an ML model in production today where you take the best option most of the time but let through “bad options” a small percentage of the time, congratulations! You’re using a combination of ε-greedy and Machine Learning.

It was great to see a big turnout at Huckletree for the 16th Data Science Ireland meetup and we want to thank all our guests who joined us for a fascinating evening. A big thank you also to Amelia, Hector and Fergal for their amazing talks with plenty of insightful thoughts to takeaway.

We are back at Huckltree in our usual slot on the first Wednesday of the month, which is the 7th of August from 6pm. You can now register and RSVP to join us for August's meetup with more top class speakers, networking opportunities and pizza! Register here.

Liam McNamara

Leave a Comment

* Indicates a required field