News Archive
Latest news is here
29.11.2017
I gave a TEDx talk in Exeter in September, with the ambitious title "Artificial intelligence, video games, and the mysteries of the mind." I definitely covered the first two points; I'll leave the mysteries of the mind to the neuroscientists and philosophers! Enjoy the talk.
I gave a TEDx talk in Exeter in September, with the ambitious title "Artificial intelligence, video games, and the mysteries of the mind." I definitely covered the first two points; I'll leave the mysteries of the mind to the neuroscientists and philosophers! Enjoy the talk.
22.4.2018
One month til ICRA in Brisbane! I'm looking forward to talking about recent research at DeepMind during my plenary talk, also looking forward to plenaries from Rodney Brooks and Mandyam Srinivasan, visits to the university labs (QUT and QCAT), and ROBOWARS.
One month til ICRA in Brisbane! I'm looking forward to talking about recent research at DeepMind during my plenary talk, also looking forward to plenaries from Rodney Brooks and Mandyam Srinivasan, visits to the university labs (QUT and QCAT), and ROBOWARS.
25.9.2017
I am organising a NIPS 2017 workshop: "Acting and Interacting in the Real World: Challenges in Robot Learning" with Ingmar Posner, Martin Riedmiller, Markus Wulfmeier, and Rohan Paul.
CfP is up - submission deadline is one week after ICLR - November 3!
I am organising a NIPS 2017 workshop: "Acting and Interacting in the Real World: Challenges in Robot Learning" with Ingmar Posner, Martin Riedmiller, Markus Wulfmeier, and Rohan Paul.
CfP is up - submission deadline is one week after ICLR - November 3!
22.8.2017
Giving a keynote at ICML 2017 was an amazing experience. I really enjoyed presenting research from DeepMind - continual learning, navigation, parkour, feudal nets, and streetlearn! This fall I'll mainly be keeping my head down working on new research, but I'll take a break in October to try my hand at TEDxExeter.
Giving a keynote at ICML 2017 was an amazing experience. I really enjoyed presenting research from DeepMind - continual learning, navigation, parkour, feudal nets, and streetlearn! This fall I'll mainly be keeping my head down working on new research, but I'll take a break in October to try my hand at TEDxExeter.
1.7.2017
Great research has come from colleagues at DeepMind in the last few months. A few papers I'm excited about:
Great research has come from colleagues at DeepMind in the last few months. A few papers I'm excited about:
- Sasha Vezhnevets et al. FeUdal Networks for Hierarchical Reinforcement Learning
Inspired by an algorithm called Feudal RL that was first proposed by Peter Dayan and Geoff Hinton back in 1993, this paper describes a manager+worker architecture that uses latent policy representations as sub-goals, succeeding in making something almost magical happen: high level behaviours are learned for 'impossible' tasks like Montezuma's Revenge without any top-down or privileged information, enabling better memory, credit assignment, generalization, exploration, and transfer!
- Nicolas Hees et al. Learning and Transfer of Modulated Locomotor Controllers
Animals and humans do not learn task-specific control of their bodies immediately. Rather, first they learn the structure, kinematics, and dynamics of their bodies through simple exploration. This paper mimics that learning process by first learning a low-level neural controller that only sees proprioceptive information, then training a high-level, goal-directed controller for increasingly difficult control tasks. Video.
- Nicolas Watters et al. Visual Interaction Networks
and Santoro et al. A simple neural network module for relational reasoning
These two papers explore the ability for deep neural networks to perform complicated relational reasoning with unstructured data. A novel network architecture is used to model the physical relationship between partially observed bodies and to infer events and outcomes of physical worlds, purely from visual observations. This model is the first to achieve superhuman performance on the CLEVR dataset.
30.6.2017
RAAIS - The Research and Applied AI Summit - took place today at Google's offices in King's Cross. Lots of great speakers and a very knowledgeable audience that seemed to be evenly split between the amazing universities in and around London and the thriving tech startup scene. Thanks to Nathan Benaich for organizing the event.
Update: here's the video of my talk
RAAIS - The Research and Applied AI Summit - took place today at Google's offices in King's Cross. Lots of great speakers and a very knowledgeable audience that seemed to be evenly split between the amazing universities in and around London and the thriving tech startup scene. Thanks to Nathan Benaich for organizing the event.
Update: here's the video of my talk
25.5.2017
I spent a summery afternoon at Oxford a few days ago, giving a talk about catastrophic forgetting and Deep RL at the school of Computer Science. Thank you to OxWoCS (Oxford Women in Computer Science) for inviting me! I gave a second talk at a Diversity in ML event in the department of Stats, aimed at undergraduates considering machine learning degrees. Slides for the first talk here.
I spent a summery afternoon at Oxford a few days ago, giving a talk about catastrophic forgetting and Deep RL at the school of Computer Science. Thank you to OxWoCS (Oxford Women in Computer Science) for inviting me! I gave a second talk at a Diversity in ML event in the department of Stats, aimed at undergraduates considering machine learning degrees. Slides for the first talk here.
14.5.2017
I'm going to the Advances in Data Science workshop in Manchester tomorrow. I'll give a talk about scaling up deep reinforcement learning for real world complexity.
I'm going to the Advances in Data Science workshop in Manchester tomorrow. I'll give a talk about scaling up deep reinforcement learning for real world complexity.
5.5.2017
I'm honored to have been asked to give a keynote talk at ICML this year, along with Peter Donnelly, Latanya Sweeney, and Bernhard Schölkopf. It's a long flight to Sydney, but I'll be there with bells on! I'll aim to give a compelling picture of current research at DeepMind, focusing on Deep RL in the Real World - how we can adapt and extend deep reinforcement learning so that it can be used for complex tasks such as robotics and navigation.
I'm honored to have been asked to give a keynote talk at ICML this year, along with Peter Donnelly, Latanya Sweeney, and Bernhard Schölkopf. It's a long flight to Sydney, but I'll be there with bells on! I'll aim to give a compelling picture of current research at DeepMind, focusing on Deep RL in the Real World - how we can adapt and extend deep reinforcement learning so that it can be used for complex tasks such as robotics and navigation.
4.5.2017
2017 is the meta-year. Meta-learning is on the rise, and I am a meta-reviewer (area chair) for 5 conferences (ICLR, ICML, NIPS, CoRL, and AAAI). It's a nice change from the drudgery of reviewing, I must say - AC-ing comes with a little more breadth, a little more perspective, and a lot more pow-- I mean, decision-making :).
2017 is the meta-year. Meta-learning is on the rise, and I am a meta-reviewer (area chair) for 5 conferences (ICLR, ICML, NIPS, CoRL, and AAAI). It's a nice change from the drudgery of reviewing, I must say - AC-ing comes with a little more breadth, a little more perspective, and a lot more pow-- I mean, decision-making :).
2.3.2017
I spoke at the LONDON.AI meetup last night, at the beautiful Alphabeta building in Finsbury Park. It was a very welcoming audience, with lots of students from Imperial, UCL, and Oxford, as well as researchers from various cool startups. I spoke about my group's recent work on catastrophic forgetting and on navigation with deep reinforcement learning.
I spoke at the LONDON.AI meetup last night, at the beautiful Alphabeta building in Finsbury Park. It was a very welcoming audience, with lots of students from Imperial, UCL, and Oxford, as well as researchers from various cool startups. I spoke about my group's recent work on catastrophic forgetting and on navigation with deep reinforcement learning.
26.2.2017
The European Robotics Forum is being held in Edinburgh, Scotland on March 22-24, and I'll be giving a keynote on challenges for deep learning and end-to-end robotics. Right now, there are more challenges than solutions, but I truly believe that robotics as a field is on the verge of being transformed by deep reinforcement learning.
Update: I've been asked for my keynote slides by a few people. Here is the pdf.
The European Robotics Forum is being held in Edinburgh, Scotland on March 22-24, and I'll be giving a keynote on challenges for deep learning and end-to-end robotics. Right now, there are more challenges than solutions, but I truly believe that robotics as a field is on the verge of being transformed by deep reinforcement learning.
Update: I've been asked for my keynote slides by a few people. Here is the pdf.
24.2.2017
I spoke about Progressive Nets and robotics at the RE•WORK Deep Learning Summit in London, on 22-23 September 2016. I don't think that the talk was recorded, but this is the video of an interview about my research.
I spoke about Progressive Nets and robotics at the RE•WORK Deep Learning Summit in London, on 22-23 September 2016. I don't think that the talk was recorded, but this is the video of an interview about my research.
12.12.2016
I've joined the executive board for the Women in Machine Learning organisation (WiML). The annual WiML workshop was held on Monday at NIPS Barcelona, and it was impressive - 600 attendees, posters, speakers, and mentoring round tables. I'm looking forward to being more involved in future WiML events at ICML and NIPS!
I've joined the executive board for the Women in Machine Learning organisation (WiML). The annual WiML workshop was held on Monday at NIPS Barcelona, and it was impressive - 600 attendees, posters, speakers, and mentoring round tables. I'm looking forward to being more involved in future WiML events at ICML and NIPS!
1.12.2016
The Conference on Robot Learning (CoRL) is a new event aiming to bring together approximately 250 of the best researchers at the intersection of robotics and machine learning. It will be held in Mountain View, California November 13 - 15, 2017.
The fields of robotics, autonomous perception and control are undergoing a machine learning revolution. Now is the time to dedicate a venue that will combine fundamental advances in machine learning with an empirical exploration of robotics applications and theory. We hope to see you in November.
The Conference on Robot Learning (CoRL) is a new event aiming to bring together approximately 250 of the best researchers at the intersection of robotics and machine learning. It will be held in Mountain View, California November 13 - 15, 2017.
The fields of robotics, autonomous perception and control are undergoing a machine learning revolution. Now is the time to dedicate a venue that will combine fundamental advances in machine learning with an empirical exploration of robotics applications and theory. We hope to see you in November.