Modern computing workloads are becoming increasingly diverse and complex, making it difficult for traditional computing systems to deliver high performance and efficiency across different applications. While general-purpose systems offer flexibility, specialized systems achieve greater efficiency by optimizing for specific workloads. However, their effectiveness can diminish as workload characteristics and hardware conditions change, leading to inefficient resource utilization, unnecessary data movement, and redundant computation. These challenges motivate a fundamental question: How can computing systems retain the efficiency of specialization while intelligently adapting to changing computational demands?

This proposed research explores the design of intelligent, self-adaptive computing systems that dynamically align computation, data organization, and hardware resources with workload needs. By combining hardware-software co-design with lightweight, workload-aware intelligence, the goal is to enable systems to make effective execution decisions without introducing significant adaptation overhead. Spanning specialized accelerators, distributed platforms, and heterogeneous computing environments, this research establishes general principles for building more efficient, scalable, and resource-conscious computing systems.