Executive Summary : | As Moore's law turns 57, high-performance computing has become increasingly prevalent, with multi-core processors and GPUs being used in data-centers and handheld mobile devices. This has led to the rise of machine learning methods in computer science and engineering research, which is influencing the design of algorithms, computer architectures, and formal methods. To address the challenges of massive datasets and dynamic updates in applications like social networks, it is crucial to develop dynamic data structures that can efficiently process massive amounts of data. This project aims to develop a code-library of data structures that supports real-time dynamic modifications scalable for massive datasets, harnesses the power of modern multi-core CPUs and GPUs, and can be made data-aware by learning the pattern of the given dataset. The real-time dynamic modifications will be enabled by concurrent operations in data structures, ensuring correctness and progress guarantee through lock-freedom. The project also aims to learn the patterns in the data using machine learning techniques and use that to optimize query processing. Traditional hierarchical-search-based data structures pose a barrier of O(log n) complexity for advanced queries like range search. The experiments in the proposed project will include benchmarking the implementations against existing state-of-the-art benchmarks, which will be operationalized and optimized for settings such as non-uniform-memory-access CPU and multi-GPU machines. |