Particle Swarm Optimization for Multi-Agent Simulation

Particle Swarm Optimization offers a range of highly effective optimization approaches for finding incrementally better solutions to otherwise complex problem areas. These optimization problems are critical to a wide range of application areas. This project aims to explore how new PSO approaches can be developed which will lead to further advances in determining more consistently optimal solutions to these particular computer science problems. A classic application of these approaches is the Travelling Salesman problem. This project will incrementally explore alternative approaches to various sample problems before examining a number of real world applications. The project team already has experience of applying these approaches to areas such as supply chain optimization. Aspects of population health and optimal inoculation strategies may also be explored.

Exploring Learning Approaches on the Cloud

The increased movement of applications to the cloud offers a range of new opportunities and challenges to organizations. Hybrid implementations have emerged as one of the preferred solutions for many organizations. Cloud applications now straddle between both private and public clouds. Moreover, multiple public cloud providers maybe used and thereby making the brokering between each of these providers increasingly complex. Furthermore, it is clear that the Cloud market is moving towards a spot price model where costs will be continuously changing to reflect demands at peak and off peak times. Furthermore, cloud providers will openly bid for business so they can maximize utilization of their cloud capacity. This project examines a number of challenges in this area, and in particular explores how learning approaches will allow these interactions to be managed and optimized. This project will incrementally explore approaches using a simulated environment before moving to number of real world applications. The project team already has extensive experience in this area and would hope that this project would involve a number of industry partners.

Public Health Analytics

The long-standing challenge of devising national treatment programs for public health initiatives is a complex one. Accurate data analytics is an essential aspect of this challenge along with scenario exploration and simulation. Over the lifetime of this project, the candidate will work with a range of practitioners in this area and develop models, which can be used to address a number of complex challenges in this space. Typical examples of these include annual inoculation strategies, which are developed on a national basis. The project team already has extensive experience in this area and would hope that this project would involve a number of industry partners.

Evolution of Cooperation through in P2P File Sharing Environments

In the area of social dilemmas, the decision to cooperate with others is always a challenging one. The decision is quite fundamental to the emergence of file sharing websites such as BitTorrent were individuals choose to become seeders and offer their material freely for others to download. Others chose to be predominantly leechers, and simply download without ever offering their material to others. Various mathematical models of these interactions have been studied throughout the area of Game Theory. This project proposes to study the various ways such levels of cooperation emerge in these environments and how this can me modeled in a simulation environment where various alternative mechanisms can be explored.

The role of social structure and norms on agent behaviors

Social networks are a fundamental factor in the communication and spread of information and social norms. The inherent structures of society and our individual social interactions determine many of our beliefs and traditions. For example, there are natural hierarchies based on social standing, age and wealth, which often serve to bias the influence of individuals in a social network. The most fundamental of these would be how connected a particular individual is. These factors have many implications for society such as online marketing campaigns, or to social scientists. The project will examine network structures through both real world data analysis, and also through simulation of an agent society. Both simulated and real world data maybe involved in this project. Data sources such as Twitter and Wikipedia offer rich sources of data involving these complex interactions.

Multi-Agent Simulation for building planning

Crowd simulation has become a growth area with respect to planning of public spaces and buildings. Increased awareness of building safety and evacuation routes are now a major consideration with respect to planning and design. This project aims to build on existing approaches to develop more scientific and advanced methods of crowd simulation for building design.

Business Intelligence and Big Data analysis for Finance

Business Intelligence and Big Data offer the single biggest challenges to corporations in the current climate. They area increasingly aware of the great potential for their massive amounts of business data, but figuring out the meaning of this data is a major challenge. This project proposes the use of some commonly used approaches along with the Systems Dynamics methodology to aid business decision making in a modern finance environment. This can link to aspects corporate governance and reporting which are all essential to the current economic climate.

Learning for Robocup Soccer Tournament

The Robocup Soccer Tournament was founded in 1997 as a means of promoting and advancing robotics and AI research. Since 1997, the competition has been held each year in a different part of the world. AI researchers convene to pitch their latest designs against each others in various categories of the completion. This project proposes to develop new techniques in this area and to work in a collaborative project with some of the most successful researchers in this area.

Trading Commercial Agents

The Trading Agent Competition (TAC) is an international competition, which was designed to develop research into complex, intelligent trading agents. Since 2002 a vibrant community has developed in this space where new agent strategies have been developed. Trading in electronic markets is an increasingly commonplace economic activity, as well as a topic of special interest within the AI, Electronic Commerce, and Multi-agent Systems (MAS) research communities. This project will build on the existing research in this area and submit strategies to the annual TAC competition.

Multi-Agent Learning for Smart Energy Management

Energy is becoming increasingly flexible to channel from one device to another, and particularly where the majority of devices are rechargeable. Smart energy solutions offer us the potential to take energy from many alternative sources and also the ability to sell it back onto the national grid. Each device whether it is a mobile phone, a pc, a electric car or a solar panel has a unique set of traits and characteristics. These devices can be modelled as an agent with a set of aims and priorities. With this in mind, the aim of this project is to develop a learning framework to allow smart energy devices to plan their energy needs, so they can guarantee to have sufficient energy when they need it, while also helping each other when there is an energy shortage. Furthermore, these smart energy devices should be capable of maximizing energy efficiency and thereby resulting is lower costs to their owner.