Systems Support for Intelligent Applications
This topic focuses on addressing the predictability, efficiency, and robustness challenges
raised by emerging real-time intelligent applications such as self-driving cars and robotics.
For instance, we develop new scheduling algorithms and runtime mechanisms that leverage
the application-level end-to-end requirements of intelligent systems, and software
frameworks that support the concurrent execution of multiple DNN tasks with deterministic
and provable timing guarantees on heterogeneous hardware platforms.
Predictable Parallel Real-Time Systems
In this topic, we aim to build predictable systems that can take advantage of modern
parallel hardware platforms (e.g., multi-core CPUs and GPUs), which are imperative to
the efficiency and usability of mission-critical applications such as avionics and
automotive systems. In particular, our focus has been on enabling predictable systems from
unpredictable commercial off-the-shelf hardware components as doing so greatly reduces
development cost and allows for timely and wide acceptance of new technology.
Smart Sensing and Processing in Dynamic Environment
For the robust and continuous operation of Cyber-Physical Systems (CPS) and the Internet of Things (IoT),
a system must be able to deal with uncertainties under various environmental conditions.
We have particularly focused on the dynamically changing availability of resources due to
extreme thermal conditions and the lack of a reliable power source.