Zhijia Zhao

Bitstream programs rely on bitwise operations (e.g., AND, SHIFT, and XOR) to process long sequences of bits. They are widely used in multimedia processing and bitmap indexing. When carrying dependences, the whole bitstream traversal becomes serial. In this work, we propose bit-level dependency analysis and modeling techniques to reduce bitstream programs to finite-state machines (FSMs), thus enabling the use of FSM parallelization techniques to parallelize bitstream programs.

Semi-structured data, like JSON and its variations, is the de facto standard for information exchange on the web and the basis for modern NoSQL document stores (e.g., MongoDB and Firebase). With nested structures, it is difficult to process such data in a scalable way. To address this, we leverage the automata theory and parallelization techniques to automatically generate data-parallel code with small memory footprint.

Many real-world graphs are large and irregular by nature. These properties make them difficult to take advantages of modern parallel architectures like GPUs, which are mainly designed for regular computations with relatively limited memory capacity. We address these challenges with novel graph transformations that can ``regularize'' the computations and minimize the CPU-GPU data transfer.
For compute-intensive iterative queries over a streaming graph, it is critical to evaluate the queries continuously and incrementally for best efficiency. However, the existing incremental graph processing requires a priori knowledge of the query (e.g., the source vertex of a vertex-specific query). We present a principled solution to generalizing the incremental processing using graph triangle inequalities, such that queries, without their a priori knowledge, can also be evaluated incrementally.

Mobile systems (e.g., Android) expose changes on the device configuration (e.g., orientation, screen size, or keyboard availability) to the apps at runtime to enable rich user experiences. When handled inappropriately, these changes can cause unresponsiveness, state loss, or even app crash. In this work, we analyze the common causes to various runtime change issues and design novel runtime change handling solutions to address them.