Hank Hoffmann: Possible BA/MS or BS/MS projects:

• Applied machine learning:
  • Classify pairings of architectures and applications according to the hardware resource which would most benefit the application if added/least affect it if removed. Some background in computer architecture and ML classifiers is beneficial.
  • Use reinforcement learning techniques to optimize energy efficiency of resource constrained applications. For example, learn to reduce processor speed for memory bound applications. Background in systems programming and reinforcement learning is helpful.
• Parallel Computing:
  • Use fine-grain parallelism to extract parallel speedups for traditionally serial applications. For example, speedup dijkstra's algorithm by separating memory operations from compute operations and executing each on separate cores. Some experience in parallel programming is required. Background in compilers is beneficial.
  • Speedup sparse matrix computations with adaptive prefetching -- put underutilized cores to work by rearranging memory dynamically to make sparse computations more efficient. Background in parallel computing is required, experience with sparse matrix computation is required.
  • Implement a parallelized version of a serial library for digital signal processing. Speedup serial DSP codes by parallelizing expensive, commonly used functions (like FFT) to produce a faster library which can be used by programmers with no knowledge of parallelism. Parallel computing experience is required, some background in signal processing or linear algebra is desired.
• Power aware computing:
  • Compare a software power managment system to a hardware-based implementation using Intel's SandyBridge architecture. Background in systems programming is beneficial.
  • Port an existing power management system to a new platform. Some choices are: mobile phone, distributed system, webserver. This process involves a significant modeling effort to determine power/performance characteristics of new platform and applications. System programming background is beneficial.
  • Implement software-based energy control system. Can we manage energy consumption the same way we manage time in CPUs? System programming background is beneficial.
• Reliability:
  • Use high-level application feedback to detect misbehaving applications and systems. Requires instrumenting and profiling a number of applications, then injecting various faults into the application and building a system to detect these fault through high-level changes in application behavior.