The two use cases outlined in this paper demonstrate two very different approaches to using ML to solve a problem: one that relies on AI-as-a-Service, and one that creates an in-house framework for developing ML models.
Automated object recognition for vehicle licence plates
This use case draws on AIaaS products to integrate licence plate recognition with business logic in Alfa’s own asset finance software platform, Alfa Systems. It demonstrates how these services enable experimentation with ML techniques, at little cost and without requiring great expertise.
Picking up from our first paper on AI, this use case analyses the results of the automated testing of Alfa's internal source code - this time using our own reusable framework. A neural network trained on historic failures learns patterns between the areas of code that are causing test failures, and the failures themselves.
The experiments detailed in this paper explore different ways to develop ML solutions, and each requires different levels of time, effort and expertise.
However, both solutions rely on domain knowledge of the data used; the first relies heavily on research in object recognition and a resulting dataset that was created by domain experts in that field, while the second is based on data with which most people at Alfa are familiar.
This paper features a foreword from Blaise Thomson, whose speech technology start-up VocalIQ was acquired by Apple and formed an important part of the Siri development team.