The Leverhulme Centre for the Future of Intelligence (CFI; http://lcfi.ac.uk/) and the Machine Learning Group (http://mlg.eng.cam.ac.uk ) at the University of Cambridge invite applications for a Postdoctoral Research Associate in the study of trust and transparency in Artificial Intelligence (AI). The appointment will be for 3 years.

CFI is a new, highly interdisciplinary research centre addressing the challenges and opportunities posed by artificial intelligence (AI). Funded by the Leverhulme Trust, CFI is based at the University of Cambridge, with partners in the University of Oxford, Imperial College London, and UC Berkeley, and close links with industry.

This is a new Research Associate post within CFI’s Trust and Transparency project, based in central Cambridge. The post-holder will be a member of both CFI, and the Machine Learning Group, run by Prof Zoubin Ghahramani, in the Department of Engineering. This project, led by Dr Adrian Weller and involving partners at Imperial College, aims to develop processes to ensure that AI systems are transparent, reliable and trustworthy.

Deadline for applications is 19 December 2016. Please see http://lcfi.ac.uk/careers/postdoc-study-trust-transparency/ for further details.

The University values diversity and is committed to equality of opportunity.

The University has a responsibility to ensure that all employees are eligible to live and work in the UK.

Zoubin Ghahramani and Adrian Weller are involved with the Leverhulme Centre for the Future of Intelligence which was launched this week.

Professor Stephen Hawking gave a widely reported speech at the launch.

Zoubin, who introduced the talk, is the Deputy Director of the Centre.

Adrian is an Executive Fellow at the centre and is leading the project on Trust and Transparency, which is part of his general interest in the social implications of AI.

Professor Magaret Boden, who also spoke, said:

AI is hugely exciting. Its practical applications can help us to tackle important social problems, as well as easing many tasks in everyday life. And it has advanced the sciences of mind and life in fundamental ways. But it has limitations, which present grave dangers given uncritical use. CFI aims to pre-empt these dangers, by guiding AI development in human-friendly ways.

Prof. Zoubin Ghahramani has spoken to the BBC world service Forum radio show on the question: “Do we need artificial intelligence?”. The full show can be found here.

Shane Gu was involved with a Google shared experience robotics project this summer. The project was written up on Google’s research blog and in the MIT technology review. Shane is the first author on one of the corresponding papers which can be found here.

MLG members are involved with the organisation of one symposium and three workshops at NIPS. These are:

Symposium

Machine Learning and the Law
Adrian Weller, Thomas D. Grant, Conrad McDonnell and Jat Singh

Workshops

Bayesian Deep Learning
Yarin Gal, Christos Louizos, Zoubin Ghahramani, Kevin P Murphy and Max Welling

Towards an Artificial Intelligence for Data Science
Charles Sutton, James Geddes, Zoubin Ghahramani, Padhraic Smyth and Chris Williams

Reliable machine learning in the wild
Jacob Steinhardt, Dylan Hadfield-Menell, Adrian Weller, David Duvenaud, Percy Liang

MLG members are shown in bold. A full schedule for the NIPS workshops and symposia can be found here.

Five new papers from the group are to appear at the 2016 conference on Advances in Neural Information Processing Systems (NIPS 2016), to be held in December in Barcelona, Spain.

The list of papers are:

Renyi Divergence Variational Inference
Yingzhen Li and Rich Turner

A Theoretically Grounded Application of Dropout in Recurrent Neural Networks
Yarin Gal and Zoubin Ghahramani

Understanding Probabilistic Sparse Gaussian Process Approximations
Matthias Stephan Bauer, Mark van der Wilk, and Carl Edward Rasmussen.

Consistent Kernel Mean Estimation for Functions of Random Variables
Adam Ścibior*, Carl-Johann Simon-Gabriel*, Iliya Tolstikhin, Bernhard Schölkopf. * equal contribution.

Distributed Flexible Nonlinear Tensor Factorization
Shandian Zhe, Kai Zhang, Pengyuan Wang, Kuang-chih Lee, Zenling Xu, Yuan Qi, and Zoubin Ghahramani.

For the most up to date versions of the papers, visit the authors’ webpages, which may be found through our group members page.

Alex Matthews has been given a software award from the Google Open Source Programs Office for his work on TensorFlow. The award is for “…people outside of Google that they thought were doing great things in the world of open source”, in particular for his work implementing Cholesky backpropagation. The work removes one of the main barriers to implementing Gaussian processes in TensorFlow and is part of the ongoing GPflow project.

Ten papers involving authors from MLG will appear at the International Conference on Machine Learning 2016. They are:

Unitary Evolution Recurrent Neural Networks
Martin Arjovsky, Amar Shah and Yoshua Bengio

Predictive Entropy Search for Multi-objective Bayesian Optimization
Daniel Hernández-Lobato, José Miguel Hernández-Lobato, Amar Shah and Ryan P. Adams.

Pareto Frontier Learning with Expensive Correlated Objectives
Amar Shah and Zoubin Ghahramani

Continuous Deep Q-Learning with Model-based Acceleration
Shixiang Gu, Timothy Lillicrap, Ilya Sutskever and Sergey Levine.

Black-box alpha-divergence Minimization.
José Miguel Hernández-Lobato*, Yingzhen Li*, Mark Rowland, Daniel Hernández-Lobato, Thang Bui and Richard E. Turner.
(* joint first author)

Deep Gaussian Processes for Regression using Approximate Expectation Propagation.
Thang Bui, Daniel Hernández-Lobato, Yingzhen Li, José Miguel Hernández-Lobato and Richard E. Turner.

Scalable Discrete Sampling as a Multi-Armed Bandit Problem.
Yutian Chen and Zoubin Ghahramani

Dropout as a Bayesian Approximation: Representing Model Uncertainty in Deep Learning
Yarin Gal and Zoubin Ghahramani

Uprooting and Rerooting Graphical Models
Adrian Weller

Train and Test Tightness of LP Relaxations in Structured Prediction
Ofer Meshi, Mehrdad Mahdavi, Adrian Weller and David Sontag

For more details see the author’s personal webpages.

We are delighted to announce that Dr. José Miguel Hernández-Lobato will join Cambridge MLG as a University Lecturer in Machine Learning later this year. Miguel joins us from Harvard and has a broad range of interests in probabilistic machine learning. For more details see his website.

Cambridge Machine Learning Group has a strong involvement with the new Alan Turing Institute. The ATI is the national institute for research in Data Science and is headquartered in London. It brings together researchers from five top British universities: the Universities of Cambridge, Edinburgh, Oxford, and Warwick, and University College London.

Our own Prof. Zoubin Ghahramani has been appointed University of Cambridge Liason Director.

Group members Prof. Carl Rasmussen and Dr. Adrian Weller have been appointed as ATI Faculty Fellows.