The Artificial Intelligence Group conducts research in many areas of artificial intelligence. The group has active interests in models of intelligent interaction, multi-agent systems, natural language understanding, constraint programming, computational vision, decision-theoretic planning and learning, machine learning, affective computing, human-in-the-loop systems, neural networks and computational social science; together with cross-appointed faculty we also investigate multi-disciplinary applications.
Intelligent user interfaces
Integrating natural language processing models and user models for the purpose of producing more effective human-computer interaction. This includes designing interfaces that allow for mixed-initiative interaction. Application areas include interface agents, electronic commerce, and recommender systems.
Studying how computational limitations influence strategic behaviour in multi-agent systems, as well as developing approaches to overcome computational issues that arise in practical applications of mechanism design and game theory. Designing systems of collaborative problem solving agents, with an emphasis on issues of communication and co-ordination for applications of multi-agent systems to the design of effective electronic marketplaces and adjustable autonomy systems. Modeling trust, reputation and incentives in multi-agent systems, including the use of social networks.
Natural language processing
The design of chatbots and conversational robots, by deep learning. The exploration of statistical and linguistic techniques to automate the analysis of natural text, the synthesis of clusters of documents, the retrieval of information from unstructured documents and the developments of methods and software tools for computational rhetoric. Application domains include personalized mobile health and web analysis.
Investigating methodologies for solving difficult combinatorial problems by emphasizing modeling and the application of general purpose search algorithms that use constraint propagation. Current projects are focused on solving optimization problems that arise in machine learning, such as learning the structure of a Bayesian network from data.
Developing general graphical and geometric models, optimization methods, and fast approximation algorithms useful in a wide range of computer vision applications, including image segmentation, recognition, stereo correspondence, multi-view reconstruction, optical flow, image synthesis and biomedical image analysis. Significant interest in supervised, weakly-supervised, and unsupervised methods for computer vision based on machine learning.
Developing computational theories of perception, based on Bayesian inference, preference rules, and qualitative probabilities, and applying such methods to problems in object recognition, motion estimation, and learning. Computational perception of scene dynamics, with applications in event recognition, human computer interaction, and robotics; the analysis and categorization of image motion, particularly in densely cluttered scenes, and the recognition of human behaviours in natural environments with application to assistive technology.
Decision-theoretic planning and learning
Design of algorithms to optimize a sequence of actions in an uncertain environment. The emphasis is on probabilistic and decision-theoretic techniques such as (fully and partially observable) Markov decision processes as well as reinforcement learning. Applications include assistive technology for persons with physical and cognitive disabilities and spoken-dialogue systems.
Developing solutions for machine learning from an academic perspective and for commercial applications. Examining the issue of how computers can “learn”, that is, how processes drawing useful conclusions from massive data sets can be automated. Considering the central role played by machine learning in a wide range of important applications, emerging from a need to process data sets whose sizes and complexities are beyond the ability of humans to handle. Developing efficient and scalable algorithms for machine learning with formal guarantees and analysis. Bayesian learning and reinforcement learning approaches. Theoretical analysis of the limitations and of the unreasonable power of machine learning models. Understanding the universal approximation properties of sum-product networks, graphical models, tensors, deep neural networks and generative models.
Studying how intelligent systems can be improved by reasoning about emotions. Investigating theories of culturally shared affective sentiments during human-machine interaction. Application areas include tutoring, sentiment analysis, assistive technologies and computational social science.
Human-in-the-loop intelligent systems
Across many AI-related research areas, there are various models for combining human and machine intelligence to solve computational problems, including human computation (aka crowdsourcing), learning by demonstration, mixed initiative systems, active learning from human teachers, interactive machine learning, etc. In these systems, humans are a critical part of the computational process -- they serve as teachers and collaborators to the AI system, providing feedback and corrections, or performing computational tasks that are difficult for existing algorithms. This area of research is at the intersection of AI, Human-Computer Interaction (HCI) and EconCS, involving the design of interfaces, algorithms and incentive mechanisms to harness human processing power to tackle challenging computational problems.
Using deep learning to understand how our brains work, where one of the secrets of natural intelligence is the use of deep, hierarchical, generative networks. Discovering biologically-plausible learning algorithms to train these networks. Using deep learning to develop solutions for natural language understanding and for problems of bioinformatics.
Computational Social Science
The exploration and understanding of human behaviour in groups and teams through computational simulation of socio-psychological agent models. Current projects include research on online collaborative networks, social networks for older adults, and communities of practice. Other work involves answering open questions in the social sciences through computational modeling, including work on stratification, ethics, social structures, implicit biases, cultural differentiation, and on the sociology of knowledge.
Together with cross-appointed faculty, exploration of intelligent autonomous vehicles, computational neuroscience, computational statistics and its application to data analytics, and human-like understanding of natural language.