Turing Lecture: Learning how to learn efficiently

Event Information

Share this event

Date and Time

Location

Location

Kings Place

90 York Way

London

N1 9AG

United Kingdom

View Map

Event description

Description

NOTE: Please note that booking for this event does not guarantee entry. As this is a free event, we may overbook the venue to ensure a full audience. We therefor ask that you arrive on time to avoid disapointment. Doors open at 18:00; talk begins at 18:30.

Training a large neural network with lots of data and subsequently deploying this model to carry out specific tasks, such as speech recognition, machine translation, game playing, image recognition, image and text generation, text-to-speech, and lipreading has been incredibly fruitful. Instead of focusing on few tasks with massive amounts of data, this talk will however focus training neural networks to solve many tasks with few data each. The objective is not to learn a fixed-parameter classifier, but rather to learn a “prior” neural network that can be adapted rapidly to solve new tasks with few data. The output of training is not longer a fixed model, but rather a fast learner. That is, the goal is to build tools that learn.

Some technical knowledge required.

About the speaker (in his own words)

Nando was born in Zimbabwe, with malaria. He was a refugee from the war in Mozambique. Thanks to his parents getting in debt to buy him a passport from a corrupt official, he ended up living in a small volcanic rock hut in Madeira, Portugal, without water and electricity, before the EU got there, and without his parents who were busy making money to pay their debt. At the age of eight, he joined his parents in Venezuela and began school in Catia; a neighbourhood of Caracas. He moved to South Africa after high-school and sold beer illegally in black-townships for a living until 1991; learning to solve ODEs in his free time. Apartheid was the worst thing he ever experienced. He obtained his BSc in electrical engineering and MSc in control at the University of the Witwatersrand, under the guidance of amazing teachers and friends, and where he strived to be the best student to prove to racists that anyone can do it. He was privileged to obtain a PhD on Bayesian methods for neural networks at Trinity College, Cambridge University, thanks to scholarships by benevolent people who donate and invest in education. He did a postdoc in Artificial Intelligence at UC Berkeley, and became a Professor at the University of British Columbia, Canada, in 2001 and subsequently at the University of Oxford, UK, in 2013. In 2017, he joined DeepMind full-time as a principal scientist to help with the vision of solving intelligence so that future generations can have a better life. Nando is also a Senior Fellow of the Canadian Institute for Advanced Research and has been the recipient of several academic awards.

Agenda

18:00-18:30 - Registration

18:30-18:35 - Welcome and introduction - Mark Briers (The Alan Turing Institute, UK)

18:35-19:25 - Learning how to learn efficiently - Nando de Freitas (Google DeepMind, UK)

19:25-19:40 - Q&A - Nando de Freitas and Mark Briers

19:40-20:30 - Drinks reception

For help, information or to arrange special provisions for accessibility, please contact events@turing.ac.uk.

More information about the event and speaker can be found on the event website.

Dr Jem Rashbass studied medicine at University College London, was a graduate student of Professor Sir John Gurdon in Cambridge, then trained in diagnostic pathology. He has worked on large-scale healthcare data systems for the last 25 years in a variety of different settings - first as an academic, then through national policy and within the health service. He is now the National Director for Disease Registration and Cancer Analysis in Public Health England and the PHE Cancer Lead. In this role he is responsible for 350 staff in the National Disease Registration Services in England.

He was the founder of Clinical and Biomedical Computing Ltd., which developed the online service “Medicines Complete” for the Pharmaceutical Press; this included the British National Formulary, Martindale Drug Reference and Merck Index. In 2011 he launched a social enterprise, Health Data Insight C.I.C. to act as an ethical information intermediary for health data; in 2018 HDI released the first version of the Simulacrum – a synthetic dataset that mirrors some of the National Cancer Registry data in PHE.

Share with friends

Date and Time

Location

Kings Place

90 York Way

London

N1 9AG

United Kingdom

View Map

Save This Event

Event Saved