Achieving Financial Fairness in Deep Learning
Fairness in machine learning - including deep learning - is a very complex concept that does not have a single solution. At the heart of the issue is Ethics and challenges in the training datasets – which humans has been trying to figure out since the beginning of civilization. There are several challenges, as Fairness considerations apply to human decision-making too. Most uncertain situations can be solved on a case-by-case basis. This will not work ‘at scale’ for algorithmic decision-making. Discrimination happens even when nobody is evil!, as ML is very good at picking up unintended proxies in the data. This session will examine the latest research and approaches with some concrete know how as a result of our work in the area to support all our ML products at Experian. As an example Insight of this session, one thing is to detect a biased algorithm which with the right methodology and approach can be done in many cases but a very different one is to fix them – which in certain cases is extremely complex.
As Head of the Datalabs, Javier oversees innovation with emphasis on development of new products and services across all business units in UK&I and EMEA. Pioneering application of artificial intelligence in mobile, voice, fraud, credit, marketing, social media, digital advertising and healthcare. He came from WPP, where he spent 7 years: two as Kantar ( WPP) Global CTO, leading the strategy of the global technology function and focusing on the new generation of market research platforms.He also spent 5 years in GroupM (WPP), as the EMEA Chief Information Officer (CIO). He was part of the Xaxis Global Technology Board. Javier has more than 26 years’ experience globally within the Finance, AI, Market Research, Media and Technology fields including serving as Global Chief Technology Officer at Havas Media. Prior to joining HavasMedia, Javier was an Executive consultant for the largest Media companies in the world within Accenture’s London Media & Entertainment practice for nearly 10 years.