Jiachen Sun
Hi 👋, I'm a senior machine learning scientist at LinkedIn's Core AI team (Dec 2024 – present). My work centers on large language models, agentic reinforcement learning, fine‑tuning and benchmarking. I'm broadly interested in computer vision, vision–language models, and adversarial & trustworthy ML for real‑world LLM applications.
I earned my Ph.D. in Computer Science & Engineering at the University of Michigan, advised by Prof. Z. Morley Mao. Prior to that, I received a B.S.E. in Information Engineering from Shanghai Jiao Tong University (graduated with honor, top 5%) and studied at the University of Washington.
You can reach me at jcsun0806@gmail.com. This webpage was 99.9% generated by agentic AI.
Research
I love building reliable and efficient machine‑learning systems that can be deployed in the wild. My work spans from pre- and post-training large language models to designing multimodal perception systems for autonomous vehicles and robotics. I’m particularly excited about:
- Large Language Models & Agents: I am currently focusing on agentic reinforcement learning for LinkedIn flagship applications.
- Vision–Language Models: building VLM agents and hardening them against visual prompt injection.
- MultiModal Perception: fusing camera, LiDAR and other sensors for robust 3D understanding and collaborative vehicular perception.
- Adversarial & Trustworthy ML: certified robustness and adversarial attacks/defenses for real-world deep learning models.
What's New
Honors & Awards
2024
Spotlight Bonus & Outstanding Employee Award, TikTok
2023
Rackham Student Travel Grant
2022
SRML Workshop Best Paper Award
2020
ACM MobiCom & MobiSys Travel Grant (Virtual)
2017–2018
Outstanding Research Scholarship of SJTU (top 1%)
2017
Outstanding Winner & INFORMS Award in Mathematical Contest in Modeling
2016–2017
Academic Excellence Scholarship of SJTU, A Class (top 1%)
2014–2016
Academic & Honor Scholarships of Zhiyuan College and SJTU
Selected Publications
See full list at Google Scholar.
Label‑Free Coreset Selection with Proxy Training Dynamics
In Proceedings of ICLR 2025 (Acceptance Rate 32.0%)
Cocoon: Robust Multi‑Modal Perception with Uncertainty‑Aware Sensor Fusion
In Proceedings of ICLR 2025 (Acceptance Rate 32.0%)
On Data Fabrication in Collaborative Vehicular Perception: Attacks and Countermeasures
In Proceedings of USENIX Security 2024
Data Condensation with Hierarchical Memory Network
In Proceedings of ECCV 2024
Dolphins: Multimodal Language Model for Driving
In Proceedings of ECCV 2024 (Acceptance Rate 27.9%)
Boosting Collaborative Vehicular Perception on the Edge with Vehicle‑to‑Vehicle Communication
In Proceedings of ACM SenSys 2024
SmoothVLM: Safeguarding Vision–Language Models Against Patched Visual Prompt Injectors
arXiv preprint arXiv:2405.10529 (2024)
CALICO: Self‑Supervised Camera–LiDAR Contrastive Pre‑training for BEV Perception
In Proceedings of ICLR 2024 (Acceptance Rate 26.0%)
VPA: Fully Test‑Time Visual Prompt Adaptation
In Proceedings of ACM MultiMedia 2023 (Acceptance Rate 29.2%)
A Critical Revisit of Adversarial Robustness in 3D Point Cloud Recognition with Diffusion‑Driven Purification
In Proceedings of ICML 2023 (Acceptance Rate 27.9%)
On Adversarial Robustness of Trajectory Prediction for Autonomous Vehicles
In Proceedings of CVPR 2022 (Acceptance Rate 25.3%)
A Spectral View of Randomized Smoothing under Common Corruptions: Benchmarking and Improving Certified Robustness
In Proceedings of ECCV 2022 (Acceptance Rate 28.0%)
Adversarially Robust 3D Point Cloud Recognition Using Self‑Supervisions
In Proceedings of NeurIPS 2021 (Acceptance Rate 26.0%)
On Adversarial Robustness of 3D Point Cloud Classification under Adaptive Attacks
In Proceedings of BMVC 2021 (Acceptance Rate 35.9%)
EMP: Edge‑Assisted Multi‑Vehicle Perception
In Proceedings of ACM MobiCom 2021 (Acceptance Rate 20.0%)
Automatic Discovery of Denial‑of‑Service Vulnerabilities in Connected Vehicle Network Stack
In Proceedings of USENIX Security 2021 (Acceptance Rate 19.0%)
Towards Robust LiDAR‑based Perception in Autonomous Driving: General Black‑box Adversarial Sensor Attack and Countermeasures
In Proceedings of USENIX Security 2020 (Acceptance Rate 13.3%)
MPBond: Efficient Network‑Level Collaboration Among Personal Mobile Devices
In Proceedings of ACM MobiSys 2020 (Acceptance Rate 19.4%)
Detecting Anomaly in Large‑Scale Network Using Mobile Crowdsourcing
In Proceedings of IEEE INFOCOM 2019 (Acceptance Rate 19.7%)
Ghostbuster: Detecting the Presence of Hidden Eavesdroppers
In Proceedings of ACM MobiCom 2018 (Acceptance Rate 22.4%)