Environmental Systems:

1. Algal Blooms (agent-based models)

Social / Human Systems:

3. Large-scale models for human interactions

4. Simulation of Social Networks (complex network models). We are now looking for interns to work on this topic!

5. Exploiting Parallel Treebanks to Improve DataDriven Machine

6. Traffic modelling

6. Hedge fund data (wavelet & spectral models)

7. Correlation Matrix Analysis (random matrix & portfolio theory)

8. Trader Motivation/Behaviour (agent-based models)

Environmental Systems

1. Algal Blooms: agent-based models (Ray Walshe)

 Description will be available soon.

Social/Human Systems (R. Walshe, Prof. H. Ruskin, D. Perrin)

3. Large-scale models for human interactions  

The concept of compartmentalised influence on the overall dynamics of complex systems is very promising, and appears very suited to the context of human interactions.

Similarly to cell-level interactions controlling whole-body infection progression in the context of immune models, (see our biosystem projects), specific one-to-one interactions in human groups are key to the emergence of system-wide features. Problems of particular interest in this context include the spread of diseases (and rumors) in a large population.

Each individual is represented by a software agent with autonomous behaviour. Using a large-scale parallel implementation and our centre's computing facilities, the objective is to model realistic urban environments and known patterns of individual mobility.

A key outcome of this project will be a detailed and quantitative evaluation of intervention policies during disease outbreaks.

4. Simulation of Social Networks (complex network models)

The social networks project involves the application of sophisticated probabilistic and statistical methods (graph theory, complex networks) to the simulation of social networking sites like Bebo, Facebook, LinkedIn etc. This is a new project and we are currently looking for intern students interested to work with us on it.

The initial phase of this work involves the creation of a parallel treebank for the English/French language pair from a preexisting parallel corpus of approximately 1 million sentence pairs. Given that we have already parsed the English side of the corpus, the main task outstanding is to parse the French side. This will be done using Bikel's (2002) history based, lexicalised, generative parser trained on a monolingual French treebank. The remaining task will be to induce the word and phrase alignment information in the parallel treebank using a statistical tree alignment tool which we developed in our MT group (Tinsley et al., 2007b). Having completed this portion of work, we will have a large fully annotated parallel treebank with which we can continue our translation experiments. The monolingual parser we will use to parse French takes approximately 40 seconds per sentence on a single CPU Pentium 4, 3.8GHz and with <4GB of RAM. These estimates are based on initial experiments carried out on a set of 5,000 sentences. Parsing the entire French corpus of ~1M sentence pairs will require 11,111 CPU hours. As the sentences are parsed independently of each other, a taskfarming solution is appropriate for this task. Thus, access to a multiprocessor architecture would allow us to greatly expedite the task.

6. Traffic Modelling

Traffic congestion is a major problem in most major cities around the world with few signs that this is diminishing, despite management efforts. In planning traffic management and control strategies at urban and inter urban level, understanding the factors involved in vehicular progression is vital. One important project on the topic, carried out by Dr. Puspita Deo, was Modelling Heterogeneous Motorised Traffic Flow Modelling using Cellular Automata. Most work to date has been restricted to single vehicle-type traffic. Study of heterogeneous traffic movements for urban single and multi-lane roads has been limited, even for developed countries and motorised traffic mix, (with a broader spectrum of vehicle type applicable for cities in the developing world). The aim of the research, was thus to propose and develop a model for heterogeneous motorised traffic, applicable to situations, involving common urban and interurban road features in the western or developed world. A further aim of the work was to provide a basis for comparison with current models for homogeneous vehicle type.

A two-component cellular automata (2-CA) methodology was used to examine traffic patterns for single-lane, multi-lane controlled and uncontrolled intersections and roundabouts. In this heterogeneous model (binary mix), space mapping rules are used for each vehicle type, namely long (double-unit length) and short (single-unit length) vehicles. Vehicle type was randomly categorised as long or short, with different fractions considered. Update rules were defined based on given and neighbouring cell states at each time step, on manoeuvre complexity and on acceptable space criteria for different vehicle types. Inclusion of heterogeneous traffic units increased the algorithm complexity as different criteria apply to different cellular elements, but mixed traffic is clearly more reflective of the real-world situation. The impact of vehicle mix on the overall performance of an intersection and roundabout (one-lane one-way, one-lane two-way and two-lane two-way) has been examined. Investigation of performance metrics for heterogeneous traffic (short and long vehicles), can be shown to reproduce main aspects of real-world configuration performance. This has been validated, using local Dublin traffic data.

The developed model has potential to extend its use to linked transport network elements and can also incorporate further motorised and non-motorised vehicle diversity for various road configurations.

New work on traffic modelling is in progress in the group. You can read more about it here.

Economic Systems (M. Crane, Prof H. Ruskin)

7. Hedge fund data (wavelet & spectral models)

Description will be available soon.

8. Correlation Matrix Analysis (random matrix & portfolio theory)

The research of the Financial Modelling Group is primarily focused in the  relatively new field of Econophysics. Otherwise termed Statistical Finance, Econophysics is an interdisciplinary subject where techniques from Statistical Physics are applied to problems in Finance.

To date, the group has applied filtering techniques based upon Random Matrix Theory to correlation matrices formed from Equity, Currency, Commodity and Hedge Fund returns. Other work has focused on the characterisation of financial markets through the use of signal processing techniques.  Recent projects have focused on the dynamics of correlations between Financial Assets, to provide insight into the strategies and time horizons underlying the interactions between the numerous constituents involved.

Read about on-going work on this topic here.

9.Trader Motivation/Behaviour (agent-based models)

 Description will be available soon.