Thesis abstract:

Deep learning, nowadays considered as a core technology of today’s Fourth Industrial Revolution, has emerged as a critical component in numerous applications, ranging from computer vision, natural language processing to robotics and healthcare. Despite the remarkable success of deep learning, the neural networks' adaptation, implementation and deployment across variable neural architecture, framework and hardware often face challenges in terms of efficiency and effectiveness. My research encompasses various layers of the deep learning algorithm and system stack, including generic neural network enhancement algorithms, GPU kernel development, compiler and framework tuning and design.