Artificial intelligence could benefit people in many ways but it also has the potential to be dangerous, biased and ultimately catastrophic to humanity. We are focused on understanding, creating and championing ethical and sustainable innovation in artificial intelligence.
Intelligent systems often use machine learning to recognise our voice commands, correct our spelling, and suggest music and films we like, from just a few keywords. They help us drive our cars and fly our planes, they learn to predict our desires and behaviours, and they can even help us do science or art. None of the achievements of artificial intelligence come from a single algorithm, equation or data set. Developing machines with intelligent behaviour takes constructing a system of interacting parts: learning, sensing, memory, action, interaction, planning, reasoning, users, and professionals.
As robots and other intelligent systems increase in intelligence and autonomy, we need answers to fundamental questions around issues including behaviour, autonomy, ethics and even the very nature of life itself.
Key topics
- data mining
- reinforcement learning
- learning on data manifolds: semi-supervised learning, spectral clustering, non-linear data embedding, link prediction
- Bayesian inference: large-scale approximate Bayesian inference, latent variable models
- learning for computer graphics: Bayesian inference for shape modelling, tracking, sampling, and transfer, machine learning for computational photography, videography, and 3D data analysis
- sparse Bayesian models (the “relevance vector machine”) and related novel learning techniques
- probabilistic approaches to tree-based pattern recognition
- adaptive analysis of multivariate time series
- methods for intelligent statistical automation
- new perspectives on deep neural networks
- model-driven data mapping and visualisation techniques