Submitted by Carl Henrik Ek on Wed, 26/04/2023 - 15:21
For their paper on Deep Gaussian Processes, our colleague Professor Neil Lawrence and his former PhD student Dr Andreas Damianou have today won the AISTATS 2023 'Test of Time' Award.
AISTATS (the International Conference on Artificial Intelligence and Statistics) is one of the most prominent conferences in machine learning. Each year it features a Test of Time Award recognising a paper from 10 years ago that has had a prominent impact in the field.
This year the award has gone to Dr Andreas Damianou (now Senior Scientist at Spotify) and Neil Lawrence (pictured right), now DeepMind Professor of Machine Learning here. At the time they authored the paper 'Deep Gaussian Processes', they were both based at the University of Sheffield.
The development and adaptation of Gaussian processes in machine learning provided the field with a solid foundation to develop regression models. Importantly they provided a statistical framework and gave a principled narrative to the heuristic methodologies at the time.
Ten years ago when the paper was presented, the Deep Learning revolution had already begun and the benefit of compositing simple models into complex structured was starting to show empirical benefits.
With a statistical foundation the authors set out to bring principle to this modelling paradigm by presenting a framework that allows a composition of Gaussian processes.
Providing the mathematical foundation to do inference in these highly intractable structures, the authors showed that composite structures are not just beneficial in the big data setting, but given a sound statistical formulation can be used when data is scarce.
The paper has had significant impact across many different domains and given the recent prominence of composite probabilistic models, it was work that was clearly ahead of its time and inspired lots of innovative research.
Congratulations to Damian and Neil!
