Carl Henrik Ek talks about the Free Lunch concept and what actually happens machines learn from data
Carl Henrik Ek is a Senior Lecturer in Computer Science at the University of Bristol, UK. In his entertaining and informative manne, he peels back the hype from deep learning and machine learning and looks at what is really going on under the hood.
The slides of the talk available for download here.
Also, check out the Gaussian Process Summer Schools to learn more about the research that Carl Henrik mentions in his talk..
Machine learning has evolved from a branch of statistics into a technology that drives some of the biggest corporations in the world.
We are now seeing politicians, practitioners, and business leaders creating a narrative about the possibilities of this technology and how it will change society.
Looking into the future is a challenging endeavor, even more so if one does not fully understand the present. Associating fundamentally human concepts like "learning" and "intelligence" with algorithms makes this process evermore daring as we anthropomorphize them based on capabilities instead of the underlying principles that achieve them.
So what is it that allows learning from data?
In this talk, Carl Henrik Ek peels off the hype to get down to the fundamentals behind learning from data. What is required of a method to create knowledge from data? The first part of the talk will focus on the philosophical aspects, while the second part will introduce Bayesian non-parametric processes - and specifically Gaussian processes - in the context of the fundamental principles of machine learning.
The aim of the talk is to create a discussion and present some ideas orthogonal to the current mainstream view of machine learning.
About Carl Henrik
Before joining Bristol, he was an Assistant Professor in Machine Learning at KTH. Other stations of his academic career include the University of California at Berkeley, Oxford Brookes University, as well as the University of Manchester.
His research focuses on the theoretical concepts underlying the field of machine learning. He is especially interested in how to specify efficient and interpretable assumptions that allow learning from small amounts of data. Most of his work is centered on Bayesian non-parametric methods and in specific Gaussian processes.
To see our library of all past and upcoming webinars, click here.