Here is a list of proposed thesis topics for 2016. My work focuses on finding better ways to compute using FPGA and cluster technology. For more details about my work, take a look at the Computer Engineering Laboratory website. Feel free to email me and make an appointment to discuss any ideas for projects within these areas.

PHWL01 Foreign Exchange Hedging Strategies (1 student)

Most major Australian companies require foreign exchange (FX) services for buying and selling services and goods in a foreign currency. This is typically done through a bank which in turn faces the problem of hedging the FX risk, i.e. managing the accumulated positions via transactions with liquidity providers. This project involves exploring new methods of applying cloud computing and machine learning techniques to predict customer flows and exchange rate changes, and will enable them to better manage foreign exchange risk. It will be in collaboration with Dr Barry Flower at Westpac Bank within the research project described at https://cel.eng.sydney.edu.au/financial/ . Possible projects include:

  • Develop an Amazon EC2 Cloud-based environment for the parallel back-testing of FX hedging strategies.
  • Develop improved customer order flow prediction based on machine learning techniques.
  • Develop techniques to better understand and model the dynamics of exchange rate variations, particularly the effect of the FX market on the spot price and spread.

PHWL02 FPGA-based Machine Learning for Algorithmic Trading (1 student)

A field programmable gate array (FPGA) is an array of logic gates in which the functionality and interconnection can be configured by downloading a bitstream into its memory. They combine the programmability of microprocessors with the speed and flexibility of application specific integrated circuits (ASICs). FPGAs can be used to accelerate problems in areas as diverse as signal processing, networking, scientific computing and financial engineering, this field of research being known as reconfigurable computing.

Low latency is a focus of many trading systems on Wall Street (e.g. see http://www.informationweek.com/news/infrastructure/showArticle.jhtml?articleID=199200297) and their systems use traditional PC technology and have latencies measured in milliseconds. In this project, improved approaches to high speed online machine learning will first be developed. An FPGA-based implementation will then be designed and implemented. Using such an approach, it is expected that machine learning algorithms which can track the market in real-time and discover trading arbitrage opportunities can be developed. This project will involve FPGA design using AutoESL (a commercial C to HDL compiler). A strong interest in computer architecture and digital systems design is required.

PHWL03 Mixed Precision Machine Learning Algorithms (1 student)

We are working with Harbin Institute of Technology to develop high-speed FPGA-based implementations of machine learning algorithms. In software implementations, designers are restricted to either single or double precision floating-point operations. Mixed precision algorithms in which most of the computations are performance in single precision, with some critical ones in double have been proposed (e.g. see http://citeseerx.ist.psu.edu/viewdoc/summary?doi= Mixed precision allows improved speed without sacrificing accuracy.

In this project, new mixed-precision algorithms for machine learning will be developed. FPGAs allow arbitrary precision to be used, allowing more aggressive tradeoffs between precision and speed exploited. This project will involve simulation of mixed-precision algorithms and a theoretical study of their properties but no hardware implementation. This project is an application of computer arithmetic.