Priyanka Verma
Poornima University (India)
Stress Testing AI in Real Time Systems - When Milliseconds matters
In automotive embedded systems, milliseconds are not margins—they are safety boundaries. I have seen firsthand how even a small delay in a control loop or perception module can alter system behavior in ways that traditional functional testing does not reveal. As AI components increasingly power automotive systems such as ADAS, driver monitoring, and predictive braking, I believe our testing approach must evolve beyond correctness and accuracy.
In this talk, I will share my experience as Head of Research, where I encountered an AI-enabled automotive module that performed accurately under standard validation but exhibited rare and irregular latency spikes under specific processor loads, interrupt patterns, and background diagnostic tasks. These timing deviations were invisible in unit and integration tests but surfaced only under carefully designed stress scenarios.
I will present the practical stress-testing strategy I applied—combining load testing, spike testing, soak testing, and fault injection with resource throttling to simulate real-world automotive constraints. I will also explain how I used AI-based anomaly supervision to learn normal inference timing behavior and proactively detect deviations. Key metrics such as p95/p99 inference latency, jitter, CPU/GPU saturation thresholds, fallback activation frequency, and decision stability under degraded conditions will be discussed from a real engineering perspective.
Note for reviewers: This will be my first international presentation, and if selected, I will do my very best to clearly explain my research and project so that attendees can understand both the technical challenges and the practical solutions. My goal is to show that in AI-enabled automotive systems, reliability is not only about accuracy—it is about deterministic performance under stress, where every millisecond truly matters.
Comprar Tickets
Dr. Priyanka Verma is a Professor in the Department of Computer Science and Engineering at Poornima University, Jaipur, India, with over 20 years of distinguished experience in academia, research, and institutional leadership. She holds a Ph.D. in Computer Science and is an active researcher in Artificial Intelligence, Machine Learning, Web Mining, Internet of Things (IoT), Data Analytics and stress testing methodologies for intelligent systems. Her research work is published in reputed international journals and conference proceedings indexed in Scopus and Web of Science, including publications with Taylor & Francis, Springer, Elsevier, and IEEE. She is the author of books and book chapters with leading international publishers such as CRC Press, Routledge (Taylor & Francis), and Wiley. Prof. Verma has received several prestigious recognitions, including the Adityashree Award (2025), Rising Contributor Award (2025), Satyaa Stree Samman (2025), ERDA Teaching Excellence Award (2024), and the I2OR International Teaching Excellence Award (2023). Her book on Web Mining was featured in a world record initiative recognized by the OMG Book of Records, and she has been invited as a guest on the internationally viewed Professor Kev Show, reflecting her global academic outreach. She has served as Guest Editor, Conference Co-Chair, Editorial Board Member, and Reviewer for numerous international journals and conferences, and is an active member of ACM, ERDA, I2OR, IAASSE, and other professional bodies. She has delivered keynote lectures, acted as a resource person in faculty development programs, guided Ph.D. scholars and postgraduate dissertations, and contributed to funded and collaborative research projects. Her current research focuses on presenting and implementing a practical stress-testing strategy for intelligent and embedded AI systems by combining load testing, spike testing, soak testing, and fault injection with resource throttling to simulate real-world automotive constraints.
