Verifying Deep Neural Networks in Autonomous Cyber-Physical

Verifying Deep Neural Networks in Autonomous Cyber-Physical

Guest Speaker:
Taylor T. Johnson – 

ELectrical Eng. & Computer Science,

Vanderbilt University

Wednesday, November 18th, 2020
Zoom Webinar Link
2:00PM – 3:00PM

Webinar Link:
https://usc.zoom.us/webinar/register/WN_YSl0DRVOQJetWGNAACPOYQ

Abstract: The ongoing renaissance in artificial intelligence (AI) has led to the advent of machine learning (ML) methods deployed within components for sensing, actuation, and control in safety-critical cyber-physical systems (CPS). While such learning-enabled components (LECs) are enabling autonomy in systems like autonomous vehicles, swarm robots, and other CPS, as demonstrated in part through recent accidents in semi-autonomous/autonomous CPS and by adversarial ML attacks, ensuring such components operate reliably in all scenarios is extraordinarily challenging. We will discuss methods for assuring safety and security specifications in autonomous CPS using our NNV (Neural Network Verification) software tool (https://github.com/verivital/nnv) which has been applied to verify specifications for adaptive cruise control (ACC) and autonomous emergency braking (AEB) systems in motor vehicles. Next, we will present recent results on using NNV to prove robustness of neural networks used for perception tasks, such as image classification, applied to the VGG16/VGG19 networks that achieve high accuracy on ImageNet, as well as recent work on robustness of semantic segmentation. We will conclude with some architectural solutions to provide safety assurance in autonomous CPS at runtime, building on supervisory control with the Simplex architecture using real-time reachability, and will discuss future research directions for establishing trustworthy AI within CPS that we are exploring in a DARPA Assured Autonomy project.

Biography: Dr. Taylor T. Johnson, PE, is an Assistant Professor of Computer Engineering (CmpE), Computer Science (CS), and Electrical Engineering (EE) in the Department of Electrical Engineering and Computer Science (EECS) in the School of Engineering (VUSE) at Vanderbilt University (since August 2016), where he directs the Verification and Validation for Intelligent and Trustworthy Autonomy Laboratory (VeriVITAL) and is a Senior Research Scientist in the Institute for Software Integrated Systems (ISIS). Dr. Johnson was previously an Assistant Professor of Computer Science and Engineering (CSE) at the University of Texas at Arlington (September 2013 to August 2016). Dr. Johnson earned a PhD in Electrical and Computer Engineering (ECE) from the University of Illinois at Urbana-Champaign in 2013, where he worked in the Coordinated Science Laboratory with Prof. Sayan Mitra, and earlier earned an MSc in ECE at Illinois in 2010 and a BSEE from Rice University in 2008. Dr. Johnson has published over 90 papers on formal methods and their applications across cyber-physical systems (CPS) domains, such as power and energy, aerospace, automotive, transportation, biotechnology, and robotics, one of which was awarded an ACM Best Software Repeatability Award. Dr. Johnson is a 2018 and 2016 recipient of the AFOSR Young Investigator Program (YIP) award, a 2015 recipient of the National Science Foundation (NSF) Computer and Information Science and Engineering (CISE) Research Initiation Initiative (CRII), and his research is / has been supported by AFOSR, ARO, AFRL, DARPA, NSA, NSF, the MathWorks, NVIDIA, ONR, Toyota, and USDOT. Dr. Johnson is a member of AAAI, AAAS, ACM, AIAA, IEEE, and SAE, and is a Professional Engineer (PE) in Tennessee.

Host: Pierluigi Nuzzo
CCI: http://cci.usc.edu
MHI: http://mhi.usc.edu