Innovation to deployment: Machine learning systems design
People: Andrei Paleyes, Eric Meissner, Neil D. Lawrence
It used to be true that computers only did what they were programmed to do, but today AI systems are learning from our data. This introduces new problems in how these systems respond to their environment. We need to better monitor how data is influencing decision making and take corrective action as required. This project addresses that challenge.
Explaining the science
The world of software engineering in the 1980s and 1990s was dominated by the software crisis, the wide availability of computers led to more software being written, but it was difficult to maintain, deploy and re-use code. The modern era is being dominated by the data-crisis. We have oceans of data, but it is unconsumable due to poor data quality. As we incorporate data into our complex software systems new challenges are emerging.
These systems are the “AI” that we have created today. The focus of this project is developing new approaches in systems, software, organisational culture and machine learning to manage the complex interacting systems that are being built that are reliant on this data.
This project aims to develop a set of tools, cultural practices and technical solutions for maintainable and interpretable AI systems. The overarching goal is to reduce the number of technical experts required to create and maintain these systems, making safe and reliable AI available to a wide range of companies.
Applications range from across all areas of data-driven system deployment, from health to logistics to ride sharing systems. The project’s main initial focus is on deploying machine learning systems in Africa in collaboration with Data Science Africa.