Grigorios Mingas

Research Interests

• Acceleration of computationally intensive Markov Chain Monte Carlo (MCMC) and Sequential Monte Carlo (SMC) methods using FPGAs (and GPGPUs): Many popular MCMC and SMC methods present inherent parallelism (e.g. Population-based MCMC uses multiple parallel chains). In my PhD, I am developing custom FPGA-based architectures and GPGPU implementations which can deliver one to two orders of magnitude higher sampling thoughput compared to a desktop processor.

• Precision optimization of MCMC and SMC methods: Reconfigurable hardware allows the use of custom arithmetic precision in any part of the implementation. By reducing precision, arithmetic operators become cheaper and more parallel modules can fit into the same FPGA. Nevertheless, when implementing MCMC and SMC samplers in custom precision, the sampling accuracy and/or the mixing speed can be affected. I am investigating ways to optimally choose precision in this setting. Results show that custom-precision floating point FPGA implemenations can offer up to a 5x speedup compared to double-precision floating point FPGA implemenations, while guaranteeing a bounded bias in the MCMC estimate.

• Development of software tools to facilitate the use of FPGAs as hardware accelerators for MCMC.

• Simultaneous Localization and Mapping (SLAM) using FPGAs (older work): SLAM algorithms are used by mobile robots which need to build a map of the environment and locate themselves in the map. By implementing SMG-SLAM (scan matching SLAM based on a genetic algorithm) on an FPGA, speed and energy savings can be significant. This work is part of the PANDORA rescue robot project of the Aristotle University of Thessaloniki.

Publications

• Grigorios Mingas, Farhan Rahman and Christos-Savvas Bouganis. "On Optimizing the Arithmetic Precision of MCMC Algorithms". FCCM 2013 (accepted). Available on request.
  

• Grigorios Mingas and Christos-Savvas Bouganis. "A Custom Precision Based Architecture for Accelerating Parallel Tempering MCMC on FPGAs without Introducing Sampling Error". In Proceedings of the 20th Annual International IEEE Symposium on Field-Programmable Custom Computing Machines (FCCM), pp.153-156, 2012. link
  

• Grigorios Mingas and Christos-Savvas Bouganis. "Parallel Tempering MCMC Acceleration Using Reconfigurable Hardware". In Proceedings of the 8th International Symposium on Applied Reconfigurable Computing (ARC), pp.227-238, 2012. link
  

• Grigorios Mingas, Emmanouil Tsardoulias and Loukas Petrou. "An FPGA implementation of the SMG-SLAM algorithm". Microprocessors and Microsystems, Volume 36, Issue 3, pp.190-204, May 2012. link

Education