It’s my fourth year as a PhD student at UW-Madison in the Department of Computer Science! I am fortunate to work with Fred Sala. I study data-driven methods for compute-efficient learning. My work been applied to several kinds of architectures, including multimodal models and diffusion LLMs.
I run MadML, a research reading seminar for students interested in all aspects of frontier machine learning research!
In 2025 I spent a wonderful summer with the Deep Learning Group at Microsoft Research, mentored by Xiaodong Liu and Lucas Liu, and continued working with them part-time through spring 2026. This summer I interned at AMD in San Jose with Vikram Appia.
Before starting my PhD, I obtained a master’s degree at SEAS in Harvard University. And even before that, I was a full-stack software engineer at Academia.edu and at Abbvie Stemcentrx. I got my bachelor’s degree in computer science from Caltech.
[For Students] Do you need help with research or advice? Schedule a time to talk!
Recent
[October 2026] Released Coral-KV, compact cross-layer prediction and residual coding for autoregressive video KV caches, done at AMD.
[September 2026] Released Register Tokens for Bounded-State Reasoning in Diffusion Language Models.
[2026] SlopCodeBench was accepted to the NeurIPS 2026 Datasets & Benchmarks Track.
[2026] Our paper Models Can Model, But Can’t Bind was accepted to COLM 2026.
[May 2026] Started my internship at AMD Research in San Jose with Vikram Appia.
[2026] Our paper OR-LLM-Bench appeared at the ICLR 2026 DATA-FM Workshop.
Research
Coral-KV: Compact Cross-Layer Prediction and Residual Coding for Autoregressive Video KV Caches.
Albert Ge, Akash Haridas, Utkarsh Saxena, Ruijing Yang, Ian Colbert, Frederic Sala, Vikram Appia, Emad Barsoum.
In submission to ICLR 2027.
Register Tokens for Bounded-State Reasoning in Diffusion Language Models.
Albert Ge, Chandan Singh, Yufan Zhuang, Xiaodong Liu, Jianfeng Gao, Frederic Sala.
In submission to ICLR 2027.
Models Can Model, But Can’t Bind: Structured Grounding in Text-to-Optimization.
Zhiqi Gao*, Albert Ge*, Alexander Michael Berenbeim, Nathaniel D. Bastian, Frederic Sala.
COLM 2026.
SlopCodeBench: Benchmarking How Coding Agents Degrade Over Long-Horizon Iterative Tasks.
Gabriel Orlanski, Devjeet Roy, Alexander Yun, Changho Shin, Alex Gu, Albert Ge, Dyah Adila, Frederic Sala, Aws Albarghouthi.
NeurIPS 2026 Datasets & Benchmarks Track.
Towards 131k-Context dLLMs.
Albert Ge, Chandan Singh, Dinghuai Zhang, Letian Peng, Yufan Zhuang, Ning Shang, Li Lyna Zhang, Liyuan Liu, Jianfeng Gao.
R&B: Domain Regrouping and Data Mixture Balancing for Efficient Foundation Model Training. Albert Ge, Tzu-Heng Huang, John Cooper, Avi Trost, Ziyi Chu, Satya Sai Srinath Namburi, Ziyang Cai, Kendall Park, Nicholas Roberts, Frederic Sala.
Everything Everywhere All at Once: LLMs can In-Context Learn Multiple Tasks in Superposition.
Zheyang Xiong, Ziyang Cai, John Cooper, Albert Ge, Vasilis Papageorgiou, Zack Sifakis, Angeliki Giannou, Ziqian Lin, Liu Yang, Saurabh Agarwal, Grigorios Chrysos, Samet Oymak, Kangwook Lee, Dimitris Papailiopoulos.
ICML 2025 (spotlight).
Workshops and Talks
ICLR 2026 DATA-FM - OR-LLM-Bench: A Pipeline for Scalable and Verifiable Text-to-Optimization Synthesis
ICML 2025 DIG-BUGS, DataWorld - R&B: Breaking the Data Mixing Bottleneck with Just 0.01% Overhead
MMLS 2024 - Ingredients for Transformer Length Generalization