Bangjie Sun

Bangjie SUN @ NUS

Graduating Ph.D. Student, National University of Singapore

Mobile AI for safety- and trust-critical real-world problems

About Me

I am a graduating Ph.D. student at the National University of Singapore (NUS), advised by Prof. Mun Choon Chan (NUS) and Prof. Jun Han (KAIST). My research sits at the intersection of mobile sensing and applied artificial intelligence, with a focus on safety- and trust-critical real-world problems. At its core, I build ubiquitous and physically grounded provenance systems — deployable on commodity devices — to verify the origin, authenticity, ownership, and transformation history of physical products and digital content.

Looking ahead, I aim to extend provenance in two directions: toward accountable human-AI workflows, where provenance supports auditable collaboration between humans and AI; and toward trustworthy physical AI systems, where provenance grounds the actions and decisions of AI agents operating in the physical world.

Beyond research, I enjoy traveling and photography, and I am a photography enthusiast.

I am on the academic job market and currently seeking faculty or postdoctoral positions.

Research Interests

My research lies at the intersection of mobile sensing and applied AI for safety- and trust-critical real-world problems. My core focus is on building ubiquitous and physically grounded provenance systems for physical and digital artifacts using commodity devices.

I study how multiple forms of evidence, including visible physical signals, sensor fingerprints, computational forensics, and digital records, can be combined to support robust provenance. This perspective is grounded in a simple premise: no single provenance signal is sufficient on its own, especially in adversarial settings.

A key goal of my work is to make provenance recovery and verification practical in everyday life. Rather than relying on specialized infrastructure, I build deployable systems on smartphones and other widely available devices so that trustworthy verification becomes accessible at the point of need.

Looking ahead, I aim to advance hybrid provenance as a foundation for artifact verification, accountable human-AI collaboration, and physical AI. In this vision, provenance supports not only authenticity and attribution, but also auditable records of how artifacts and decisions are created, transformed, and acted upon.

Mobile Sensing & Applied AI Primary Security of Mobile AI Systems Secondary

Recent News

2026

May

I was selected as the ACM MobiSys Rising Star 2026 for research on building widely usable provenance systems using mobile devices.
Our paper "What Do Neighbors Know? Open-World Semantic Inference Attack on Intermediate Representations" was accepted to NetAISys'26 co-located with MobiSys'26 in Cambridge.
Our poster "Poster: Discovering Unanticipated Semantic Leakage from Intermediate Representations" was accepted to ACM MobiSys'26.

Apr

Our paper "RD-PHash: A Robustness Enhancement for DCT-Based Perceptual Hashing against Adversarial Bit-Flipping Attacks" was accepted to ArtSec'26 co-located with IEEE S&P.

Mar

Our paper "WRATH: Turning Watermark Robustness Against Itself via a Watermark-Agnostic Black-Box Invalidation Attack" was accepted to IEEE S&P Cycle 2. Acceptance rate: 12.6% (135 out of 1070).

2025

Aug

Received the NUS SOC Dean's Graduate Research Excellence Award. [slides]

Jun

Our poster “CAMPrints: Leveraging the “Fingerprints” of Digital Cameras to Combat Image Theft” won the Best Poster Award at ACM MobiSys'25.

Apr

Our paper “CAMPrints: Leveraging the “Fingerprints” of Digital Cameras to Combat Image Theft” was accepted to ACM MobiSys'25.

Connect