Efficient computing systems for deep learning and combinatorial optimization

基于机器学习的人工智能已经在几个重要领域找到了应用，包括模式和语音识别、机器人和自动驾驶。特别是，组合优化是许多智能应用的关键，包括交通控制和规划、高效物流规划和智能电网。然而，机器学习系统需要处理大量的数据，这些数据需要大量的计算、内存存储和它们之间的通信资源。这一问题对传统的计算体系结构提出了重大挑战。该项目将与华为加拿大公司合作，为机器学习和人工智能开发高效的计算系统。为此，我们将追求几个新的研究方向，包括：a.使用近似和超紧凑组件的硬件高效神经网络；b.用于处理大数据的内存或近内存处理；以及c.一种基于自然计算原理的新型计算机芯片，用于高效的组合优化。该项目得到了华为加拿大公司的全力支持，该公司目前在加拿大多个城市拥有800多名员工，通过亚利桑那大学的联合研究计划。加拿大一直是机器学习研究的大本营；然而，相关行业一直落后于国际竞争对手。该项目的成果将有助于填补这一空白，并推动加拿大行业在竞争中处于更强大的地位。项目成果将导致更环保的信息和通信技术(信通技术)系统，并有助于解决减少信通技术行业碳足迹方面的关键挑战。这项研究采用了一种真正的创新方法，为高素质人才(HQP)提供了极好的培训机会，他们将配备加拿大工业高科技部门高度要求的技能。因此，该项目的预期结果将对经济的长期增长产生深远影响，并将对加拿大的工业和经济具有重要意义。

Machine learning-based artificial intelligence has found applications in several important areas, including pattern and voice recognition, robotics and autonomous driving. In particular, combinatorial optimization is key to many intelligent applications, including traffic control and planning, efficient logistic planning and smart power grids. However, machine learning systems need to process a huge amount of data that requires significant resources in computing, memory storage and the communications between them. This issue presents a significant challenge for conventional computing architectures. In collaboration with Huawei Canada, this project will develop efficient computing systems for machine learning and artificial intelligence. To this end, we will pursue several new directions of research, including: A. Hardware-efficient neural networks using approximate and ultra-compact components; B. In- or near-memory processing for handling big data; and C. a new type of computer chips based on the principles of natural computing for efficient combinatorial optimization. This project is fully supported by Huawei Canada, who currently has over 800 employees with offices in multiple Canadian cities, through a joint research program at the U of A. Canada has been known as a stronghold for machine learning research; however, the related industry has been lagging behind international competitors. The outcomes of this project will help fill this gap and promote the Canadian industry to a stronger position in the competition. The project outcomes will lead to more environmentally friendly Information and Communication Technology (ICT) systems and help resolve key challenges in reducing the carbon footprint in the ICT industry. This research takes a truly innovative approach and provides excellent training opportunities for highly qualified personnel (HQP), who will be equipped with skills highly demanded by the high-tech sectors in Canadian industry. The expected outcomes of this project will, therefore, have a profound impact on the long-term growth of the economy and will be of significant industrial and economic importance to Canada.