Seminars

While conventional reinforcement learning focuses on designing agents that can perform one task, meta-learning aims, instead, to solve the problem of designing agents that can generalize to different tasks (e.g., environments, obstacles, and goals) that were not considered during the design or the training of these agents. In this spirit, we consider the problem of training a provably safe Neural Network (NN) controller for uncertain nonlinear dynamical systems that can generalize to new tasks that were not present in the training data while preserving strong safety and correctness…

Sensing and actuation systems are entrusted with increasing intelligence to perceive the environment and react to it. Their reliability often relies on the trustworthiness of sensors. As process automation and robotics keep evolving, sensing methods such as pressure/temperature/motion sensing are extensively used in conventional systems and rapidly emerging applications. This talk aims to investigate the threats incurred by the out-of-band signals and discuss the low-cost defense methods against physical injection attacks on sensors. Dr. Hei will present the results from…

Traditionally, in healthcare, records are kept with pencil and paper. Now we have healthcare IT, and with the introduction of wearable technology, we can automatically record health data in clinical settings and at a patient’s home. In sport, wearable devices are removing the need for video analysis and making skill development automated and portable.  These advances have also opened up a market of intelligent feedback systems that provide quality insights directly to the user. This presentation focuses on how I advance research in healthcare and athletic performance by designing wearable…

Neural networks have become ubiquitous when it comes to learning enabled cyber-physical systems, like autonomous cars and closed loop medical devices.  Such safety critical applications, lead to a compelling use case for formal verification approaches.  In this talk I will be presenting  an approach for verification of neural network controllers for closed loop dynamical systems.  Given a neural network and a set of possible inputs to the network described by polyhedral constraints, the aim would be to compute a safe over-approximation of the set of possible output values.  I would present…

Onboard resources continue to be stretched to maximize performance while accommodating more advanced algorithms. Computation, communication, and control resources must be carefully allocated to achieve mission objectives. Traditionally, this allocation is fixed and designed for worst-case anticipated conditions despite the dynamic environment in which it operates. To address this, I have developed a strategy for co-designing single and multi-agent controllers that co-regulate cyber and physical effectors corresponding to system needs and performance. Single-agent controllers dynamically…