Evolvable hardware (EHW) is a method of designing hardware through evolutionary methods. Through rapid evaluation of thousands of candidate solutions on an FPGA, good solutions can be found in a reasonable amount of time.
The Xilinx Zynq reconfigurable device.
The ROBIN group has long experience in research on evolvable hardware (EHW). We would like to continue current research and try out new possibilities, and here are some options for MSc projects:
- New search methods for evolution of digital circuits. Evolution of digital circuits can be a challenging tasks because of the complexity of the search landscape. However there have been advances in search methods in related fields, such as evolutionary robotics, which we think could be promising to try out also on evolvable hardware. The project would involve testing various state-of-the-art evolutionary search methods for evolution of digital circuits.
- EHW classifier systems. We have earlier developed an EHW classifier system similar to ensemble learning approaches. It is capable of doing high-speed classification and is simple to train on an FPGA. Topics for further improving the approach could be as follows:
- Improving the efficiency of the implementation, so that larger ensembles could fit in one FPGA
- Finding good application areas where high speed classification is required
- Improving training schemes, based on machine learning theory
- Creating a high speed dedicated training module on FPGA
The thesis work could either take a theoretical direction - in the machine learning methods, or a more practical direction - in optimizing the hardware of the current system.
Preconditions: Strong interest in machine learning, and evolutionary algorithms. For some of the topics good FPGA knowledge is required.