Runtime optimization of the open source PLSNPAIRS project used for fMRI analysis:
- Used Hadoop MapReduce infrastructure to leverage multi-node clusters
- Developed and implemented a workload distribution algorithm and node evaluation techniques to fairly distribute the work across a heterogeneous cluster
- Integrated with work of other students leveraging GPU processing to reduce runtime on a per node basis
- Experimented with and evaluated various methods of precursor calculations to optimize search for the optimal predictability and reproducability (p, r)
- Implemented Gradient Descent for further search optimizations
The final report can be found here.
The final presentation can be found here.