Location: Boxever, Tara Street

Date: 25 Apr


[18:00 – 18:30] Networking

[18:30 – 19:45] Welcome message by Boxever and two presentations followed by Q&A.

• Dominik Dahlem (Lead Data Scientist at Boxever): “Reinforcement Learning with Deep Learning Q-Function Approximation”.
Online booking sites for example in the travel industry often use personalised recommendations to drive conversions. The ability to link behavioural clickstream data with transactional order data affords us the opportunity to utilise feedback online with respect to the recommendations made. For example, it is common to deploy a recommendation model, test it as part of an A/B testing framework, and evaluate retrospectively for which customers the model worked and for which ones it didn’t. This is often a manual process and as a consequence may fail to exploit the richness in the behavioural and transactional features. This talk will introduce a machine learning approach using reinforcement learning coupled with deep neural networks that can be applied to this setting. In our example above recommender models (of any kind) can be deployed into a managed container that automatically adjusts a weight vector for the models according to the real-time features of a web session. The feedback of whether a customer purchases a recommended product can now be incorporated to optimise the mixture of models. This talk will demonstrate the machine learning pipeline using a simple maze demo.

• Peter Cahill (Founder at Voysis): “Introduction to Speech Recognition with Deep Learning”

[19:45 – 20:30] Networking