In general I am interested in ways that make parallel programming more accessible to the programmer yet allowing to exploit the full potential of current multi/many-core architectures.
Briefly :
Parallel Programming
High Performance for Java Applications
Automatic Parallelization (more info) using Thread-Level Speculation (more info)
Transactional Memory (more info), mainly in software
Optimizing Compilers (more info), mainly for parallelization
Computer Architecture
Association Rules (more info)