My name is Zecheng (Victor) Wang. I am a second-year PhD student at Boston University (as of Fall 2026). My research interests broadly center on machine learning, world models, and sequential decision-making. I am particularly interested in how models can efficiently recover structural and causal patterns from observations, represent multiple plausible futures, and support robust decision-making in autonomous systems. My current work investigates energy-based world models and iterative inference, including the longer-term possibility of using inference over learned energy landscapes for unsupervised thinking without supervised Chain-of-Thought traces.

At Boston University, I have also worked on safety-critical autonomous driving, where I explored higher-fidelity vehicle simulation and developed a BeamNG benchmark for evaluating agile evasion, stabilization, and recovery during imminent collision events. For BabyVLM-V2, I developed a reusable visual data-mining pipeline supporting multiple benchmark tasks and constructed the Picture Vocabulary dataset. From October 2024 to June 2025, I was a Research Intern at Microsoft Research with Zhengyuan Yang, investigating in-context learning in vision-language-action models for in-distribution and out-of-distribution robotics tasks.

Before that, I working as Research Assistant with Dr. Keith Ross at NYU Abu Dhabi. Our research focused on Deep Reinforcement Learning topics such as pre-training effects on RL, cross-environment learning, as well as using scalable architectures in RL. See our latest topic here.

I graduated from New York University Shanghai. During my undergraduate years, I worked with Dr. Yik-Cheung Tam on Natural language processing topics such as task-oriented dialogue systems and retrieval-augmented language modeling. See our paper at ICASSP 2023.

During my study away program at NYU New York, I joined AI for Scientific Research Group as machine learning specialist working on solving scientific problems with ML solutions. Later I become the ML Lead for the group responsible for multiple teams of the group. Currently, I act as an ML advisor for the group.

I conducted my capstone project with Professor Keith Ross on modeling simultaneous machine translation with Reinforcement Learning framework. See the report.

Latest CV here.

Latest News (Starting from 2024)

• 5.2026 - Our paper Making Sense of Touch from the Child's View for Contrastive Learning was accepted at ICDL 2026.
• 5.2026 - We submitted CrashCourse: Towards Agility in Safety-Critical Driving to CoRL 2026.
• 2.2026 - Our paper BabyVLM-V2: Toward Developmentally Grounded Pretraining and Benchmarking of Vision Foundation Models was accepted at CVPR 2026.
• 6.2025 - I completed my research internship at Microsoft Research, where I worked on in-context learning in vision-language-action models.
• 9.2024 - I started my PhD in Computer Science at Boston University.
• 1.2024 - Our paper Pre-training with Synthetic Data Helps Offline Reinforcement Learning was accepted at ICLR 2024.