Human-centric machine learning.

March 9, 2015
Drew Purves | Microsoft Research

How can we enable a human being to carry out machine learning that is of value to themselves and to others? Not, I would argue, by reducing that human to an operative that simply selects some data and fires off a giant automated algorithm! Rather, we need to enable that human to interact creatively with every step of the machine learning process, and with other humans, in order to extract whatever it is that they want and need: understanding, new protocols for gathering data, predictive models, whatever. Illustrated with examples based on many years of experience of using machine learning within a particular domain – ecological and environmental science – I’ll outline 6 key barriers that I believe need to be broken in order to realize the true societal potential of machine learning.

Speaker Details

“Drew Purves is head of the Computational Ecology and Environmental Science group (CEES) at Microsoft Research Cambridge. His overarching research interest is in combining ecological theory, with large and varied data sets, via computational statistics to produce quantitative, predictive models of ecological phenomena. Following Purves’ lead, the CEES group is using this approach to build new models to address global environmental challenges—such as carbon-climate, food security, wood production, biodiversity and ecosystem function, and pandemics—while developing new software tools to enable others to conduct this kind of ecological modeling. In 2012, Purves was one of 40 young scientists worldwide invited to attend the World Economic Forum’s “Summer Davos.” “