Machine Learning Models Walk the Runway at 2018 DataSci Awards Event

Corvil’s “You vs. the Machine” game proved popular with DatSci attendees asking how machine learning models go from the lab into real world products

Corvil continued an annual tradition this year with a stand in the “GeekOut Zone” at the 2018 Data Science (DatSci) awards in Dublin. We were on hand to introduce Corvil to a host of delegates from across industry and academia, who were eager to learn about Corvil’s approach to model building, anomaly detection, and big data stacks.

As you would expect from an awards ceremony honoring the best and brightest amongst Ireland’s considerable stock of data scientists, the crowds attending the awards were highly attuned to the variety of algorithms and techniques Corvil are using for both big data analysis and user recognition modelling.

How easy is it to build machine learning model?

A common question that is asked at various types of events is “how easy is it to build a machine learning model to recognize someone’s behaviour?” The answer, of course, depends on how accurate the model must be. For example, it is very difficult to build a highly accurate model. We know this because machine learning for anomaly detection and behavioral analysis are areas of intense research and development effort to fuel our Security Analytics and Intelligence solutions.

However, it is surprisingly easy to build a model that will perform better than random chance. To demonstrate this, Corvil’s Data Science team developed a simple game using a single neuron from a Neural Network (known as a Perceptron). The game learns patterns from a player’s left or right mouse clicks, and to predict whether the next click will be left or right. The more the user plays, the more the game learns about their patterns, the more accurate the models become, and the more difficult the game gets.

The “You Vs. The Machine” game proved highly popular at this year’s awards ceremony, with queues of people lining up to test their mettle against the Perceptron. Players ranged from students and junior analysts, through managers, and up to director and CTO level. The winner of the day was Pawel Lee of First Data, whose innovative and winning technique was to use the rhythms of jazz standards to create complex patterns, which were perfect for confusing the machine -- at least for a while.

Queues of enthusiastic data scientists vying for a spot on the Corvil leaderboard

The winner of the game, Pawel Lee of First Data

In the end, the machine does what machines do - tirelessly learns more and executes based on the knowledge - while Pawel’s fingers and attention got tired. It was great fun to watch, and everyone keen to find out the link between the game and Corvil’s user recognition research and asked us a ton of questions, such as:

  • Why learn user behaviors on the network? If the traffic patterns for these employees deviate significantly from their learned models, it could be an indication that their account has been breached (or other concerning behavioral change).
  • Why push the model to +99% accuracy? Reducing alert noise is a priority as security teams already get too many alerts from their log and other monitoring tools. For machine learning techniques to effectively detect anomalies in the real world, we have to be continually improving the model.
  • How do models go from the lab into a product? Through the magic of engineering (aka lots of hard work by very smart people) and customer testing programs, machine learning models are delivered through solutions like Intelligence Hub.

Taking Machine Learning Models From the Lab into The Real World

The Intelligence Hub demo was a steady attraction across the day, with John Carey kept busy introducing people to Corvil’s data analytics and operational business intelligence solutions. There was also some discussion around the technical details like:

People were very impressed with the dashboards/visualizations and how easy and intuitive they were to use. It was good validation, since one of our goals is enabling individuals with varying levels of data science expertise to leverage machine learning applied to network data. As our Co-Founder & Chief Scientist, Fergal Toomey, says in his “Bringing AI to Network Data” blog

For algorithms to produce meaningful results they need people to apply them to the areas that matter.

With the a wide range of people who were interested in our research, Intelligence Hub, and the business challenges we are solving, it is clear that Corvil is to the fore in the world of Data Science in Ireland!

Learn more about Corvil Intelligence Hub
Explore the Solution   Watch Now
Download the Solution Brochure
Learn about Self-Service Intelligence for Electronic Trading

Michael Fenton

Michael Fenton, Senior Data Scientist in Development
Pico is a leading provider of technology services for the financial markets community. Pico provides a best-in-class portfolio of innovative, transparent, low-latency markets solutions coupled with an agile and expert service delivery model. Instant access to financial markets is provided via PicoNet™, a globally comprehensive network platform instrumented natively with Corvil analytics and telemetry. Clients choose Pico when they want the freedom to move fast and create an operational edge in the fast-paced world of financial markets.