Liangze Jiang

I am a PhD student at , advised by Damien Teney and Caglar Gulcehre. I am also a Research Assistant at Idiap Research Institute.

Previously, I was a research scientist intern at Meta Superintelligence Labs (FAIR) working with Kartik Ahuja and David Lopez-Paz, and was a student researcher at Google Research hosted by Claudiu Musat. I received my MSc from EPFL and BEng from UESTC.

I aim to build models that learn data-efficiently and generalize beyond their training distributions. I see these as one problem: learning from less data means generalizing more from each example. At Meta, I studied data efficiency for LLM pretraining, exploring what squeezes the most juice from every token and how it scales. My published work investigates:

  • 🧱 Data efficiency via pre-pretraining: warming up language (and vision) models with procedural data (ICML'26 Oral, CVPR'26).
  • 🧭 Learning to adapt: meta-learning to optimize a model's inductive biases (CVPR'25 Oral) or to explore as language agents (ICLR'26).
  • 🔗 Generalization under distribution shift: how data and algorithms together shape OOD robustness (ICML'25, ICLR'24, ICLR'23).

 profile photo

Selected Research (* denotes equal contribution; full list on Google Scholar)

Procedural Pretraining teaser
Procedural Pretraining: Warming Up Language Models with Abstract Data
Liangze Jiang*, Zachary Shinnick*, Anton van den Hengel, Hemanth Saratchandran, Damien Teney
ICML 2026Oral
Project arXiv Code
Meta-RL teaser
Meta-RL Induces Exploration in Language Agents
Yulun Jiang*, Liangze Jiang*, Damien Teney, Michael Moor, Maria Brbic
ICLR 2026
arXiv Code
Procedural warm-up for ViTs teaser
Can You Learn to See Without Images? Procedural Warm-Up for Vision Transformers
Zachary Shinnick, Liangze Jiang, Hemanth Saratchandran, Damien Teney, Anton van den Hengel
CVPR 2026
arXiv Code
OOD-Chameleon teaser
OOD-Chameleon: Is Algorithm Selection for OOD Generalization Learnable?
Liangze Jiang, Damien Teney
ICML 2025
OpenReview arXiv Code
Simplicity bias teaser
Do We Always Need the Simplicity Bias? Looking for Optimal Inductive Biases in the Wild
Damien Teney, Liangze Jiang, Florin Gogianu, Ehsan Abbasnejad
CVPR 2025Oral
arXiv
OOD diversification teaser
Unraveling the Key Components of OOD Generalization via Diversification
Harold Benoit*, Liangze Jiang*, Andrei Atanov*, OÄŸuzhan Fatih Kar, Mattia Rigotti, Amir Zamir
ICLR 2024
OpenReview arXiv
FedTHE teaser
Test-Time Robust Personalization for Federated Learning
Liangze Jiang*, Tao Lin*
ICLR 2023
OpenReview arXiv Code

Misc

My Chinese name is 姜良泽, pronounced as "Jiang Liangze" in Pinyin.


Last updated in September 2026. Template is here.