Seminar: Property Inference for DNN

Topic: Property Inference for DNN

Time: 17h00, Thursday, August 4th, 2022

Speaker: Le Cong Thanh, Research Engineer – Singapore Management University

📌 Location: Online on MS Teams at this link


Deep Neural Networks (DNNs) have recently emerged as a powerful framework for solving complex real-world problems, including safety-critical tasks such as autonomous driving, finance, and medical diagnosis. Despite their popularity, it has been shown that DNNs can be vulnerable and unreliable. Ensuring the reliability and trustworthiness of DNNs thus becomes an increasingly challenging and essential task. Towards this, researchers recently have developed DNNs analyses, e.g., verification or testing, to provide insights into the behaviors of DNNs.
In this talk, we will study property inference – a new research direction on DNNs analysis. First, the presenter will introduce the property inference for deep neural networks. Then, he will present the first work on the research direction, Prophecy, which automatically infers the formal properties of feed-forward neural networks and gets introduced at ASE 2019. Finally, GNN-Infer, a new property inference technique introduced at ICSE 2022, will be presented towards discovering formal properties for graph neural networks.

Speaker Bio:

Thanh Le-Cong is a research engineer at Singapore Management University. He will join The University of Melbourne as a Ph.D. student next year. He received a B.E. in Information Technology from the Hanoi University of Science and Technology in 2021. His research interests span software engineering and artificial intelligence. He is currently focusing on automated debugging for software and artificial intelligent systems.

Slides: DNNInfer

Recording:  to be updated …