SPX: Collaborative Research: Ula! - An Integrated Deep Neural Network (DNN) Acceleration Framework with Enhanced Unsupervised Learning Capability

点击翻译按钮获取中文摘要

In light of very recent revolutions of unsupervised learning algorithms (e.g., generative adversarial networks and dual-learning) and the emergence of their applications, three PIs/co-PI from Duke and UCSB form a team to design Ula! - an integrated DNN acceleration framework with enhanced unsupervised learning capability. The project revolutionizes the DNN research by introducing an integrated unsupervised learning computation framework with three vertically-integrated components from the aspects of software (algorithm), hardware (computing), and application (realization). The project echoes the call from the BRAIN Initiative (2013) and the Nanotechnology-Inspired Grand Challenge for Future Computing (2015) from the White House. The research outcomes will benefit both Computational Intelligence (CI) and Computer Architecture (CA) industries at large by introducing a synergy between computing paradigm and artificial intelligence (AI). The corresponding education components  enhance existing curricula and pedagogy by introducing interdisciplinary modules on the software/hardware co-design for AI with creative teaching practices, and give special attentions to women and underrepresented minority groups.The project performs three tasks: (1) At the software level, a generalized hierarchical decision-making (GHDM) system is designed to efficiently execute the state-of-the-art unsupervised learning and reinforcement learning processes with substantially reduced computation cost; (2) At the hardware level, a novel DNN computing paradigm is designed with enhanced unsupervised learning supports, based on the novelties in near data computing, GPU architecture, and FGPA + heterogeneous platforms; (3) At the application level, the usage of Ula! is exploited in scenarios that can greatly benefit from unsupervised learning and reinforcement learning. The developed techniques are also demonstrated and evaluated on three representative computing platforms: GPU, FPGA, and emerging nanoscale computing systems, respectively.