Hi! I'm Jucheng (Jack) Shen, a senior at Rice University majoring in Computer Science, Mathematics, and Economics. I'm interested in robotics, reinforcement learning, and diffusion models. So far, my work has focused on making generative models efficient and understanding how they work, and I'm now exploring how these tools can help robots perceive, predict, and act in the world. I'm applying to PhD programs for Fall 2027.

I'm also collaborating with Jiachen Yao and Robert Joseph under Prof. Anima Anandkumar at Caltech on a project that began as a Caltech SURF in summer 2026.

Previously, I collaborated with Barbara Su under Prof. Anastasios Kyrillidis at Rice, worked with Boya Zeng under Prof. Zhuang Liu at Princeton, with Yeonju Ro under Prof. Atlas Wang at UT Austin, with Dr. Souvik Kundu at Intel Labs, and with Fanjiang Ye under Prof. Yuke Wang at Rice.


Publications * equal contribution

One shared model: with input injected it keeps some cells undecided; without input it commits to every cell.

One Model, Two Roles: Emergent Specialization in a Shared Recurrent Transformer

Jucheng Shen*, Barbara Su*, Anastasios Kyrillidis

NeurIPS 2026 | paper | code | blog

A controller reads the model's confidence and sets the decoding block size: large when confident, small when not.

Improving the Throughput of Diffusion-based Large Language Models via a Training-Free Confidence-Aware Calibration

Jucheng Shen, Gaurav Sarkar, Yeonju Ro, Sharath Nittur Sridhar, Zhangyang Wang, Aditya Akella, Souvik Kundu

Findings of ACL 2026 | paper | code | blog

Thresholds calibrated on the first input are reused for all later inputs, replacing one static cutoff.

Beyond Static Cutoffs: One-Shot Dynamic Thresholding for Diffusion Language Models

Jucheng Shen, Yeonju Ro

NeurIPS 2025 ER Workshop | paper | code

Over 300 controlled experiments pick the best design choices, which form i1, released with weights, code and data.

i1: A Simple and Fully Open Recipe for Strong Text-to-Image Models

Boya Zeng, Tianze Luo, Shu Pu, Jucheng Shen, Taiming Lu, Gabriel Herbert Sarch, Zhuang Liu

NeurIPS 2026 BeNTo Workshop | paper | code | data | project page

A low-resolution sketch sets the layout, then the high-resolution frame is refined tile by tile.

HyperGen: Empowering Ultra-high-resolution Video Generation Framework with Sketch-Tile Collaboration

Fanjiang Ye, Zepeng Zhao, Yi Mu, Jucheng Shen, Renjie Li, Kaijian Wang, Saurabh Agarwal, Myungjin Lee, Triston Cao, Aditya Akella, Arvind Krishnamurthy, T. S. Eugene Ng, Zhengzhong Tu, Yuke Wang

PPoPP 2027 | paper