Home | People | Research | Publication | About | Artifact

Big data and Machine Learning Processing

The continuous growth of big data regime and machine learning applications has posed an overwhelming increase in computing demand. How to enable high-performance/low-power/real-time processing of such emerging workloads, without sacrificing the quality of service (QoS), has, therefore, become a crucial question to computing industry. We adopt a software-hardware codesign approach to tackle this problem:

Secure/Dependable Computing Systems

Cyber-security in the digital age is a first-class concern. The ever-increasing use of digital devices, unfortunately, is facing significant challenges, due to the serious effects of security vulnerabilities, which can lead to medical data leakage, adversarial control of nuclear power plants or car hijacking. This challenge is even more pronounced in battlefields when IoT systems are more likely to be targeted by cyber-criminals or hostiles. To identify vulnerabilities in digital systems, we adopt a fuzzing-based approach to:

Low-Power Computing

Power efficiency has become a top priority for both high-performance computing systems (e.g., data center) and resource-constrained embedded systems (e.g., mobile/IoT device). The adverse effects of high-power consumption include an enormous emission of environmentally hostile carbon oxide, system reliability and lifetime degradation due to power-induced thermal effects, and operational cost of cloud and IoT service. To tackle this problem, our work adopts a cross-layer approach: