By Michael Hughes
The key takeaway for the speakers, exhibitors and delegates who attended BDA Rotterdam was the cross-sector relevance of
By anchoring your core objectives to the mastering of , and supplementing it with such complementary technologies as AI, machine learning, IoT and automation, to mention but a few, you will limit the potential for systemic failure and increase your potential for success. However, before mastering such forces at your disposal, it is vital that you put in place the people, processes and structure necessary to make sense of your production processes.
Best practice examples of what is required to achieve success in the production process were expertly detailed by Teren Teh, Vice-President, Market & Customer Insights, Barclays, Redouane Boumghar, Former mentor at NASA Frontier Development Lab; Member, Libre Space Foundation and Kasper Knol, Data Scientist, Ford Motor Company.
The focus of Teren’s presentation was on how to successfully deploy your models to production through the maximisation of AI and ML. Addressing a Dutch audience, Teren highlighted the fact that 61% of Dutch companies are still only in the planning or piloting stages of AI. Teren stated that this was due primarily to workplace culture which in turn impacts the route to deployment. First and foremost, it is vital to articulate a vision of what it is you wish to achieve in order to change yourself and your organisation. Next, you must set goals and embrace all outcomes. By starting simple, you can then scale up in an ordered and professional manner. So, before all else, get your culture right, understand that progress is never a clear straight line, but fraught with many deviations and corrections which require working backwards in order to move forward.
Following on from Teren, Redouane spoke of the ingredients required for successful ML projects. Red highlighted the positive role played by communities in establishing a space for learning and development, with particular reference to hackathons, open source platforms and open data forums. Through such communities and forums comes the necessary data context and the binding of information and relations, from which reasoning and understanding accumulate. The collectivisation of these individual components results in a playground for progress allows for learning to come from failure and establishes recognised modes for success.
Kasper, in a departure from Teren and Red, addressed how best to improve your training data for ML in production. As all data scientists are fully aware, being able to trust your data results is of tantamount importance when you are looking to deploy ML models in production. Using the unique example of the connected vehicle, Kasper highlighted the role that IoT and big data are beginning to play in creating a fleet of vehicles with a reputation as the most reactive, pre-emptive and technologically advanced in the history of the automotive industry. Through the establishment of a reliable training data model, it is possible to rely on three key metrics in order to measure performance; these being data visualisation and the plotting of signal data to see behaviour and patterns, finding out what is special by providing data to the ML model that helps it learn what is normal and what is not, and choosing your prediction period, which in turn helps you to find a balance between the predictive power of the data and time to action.