Jialu Gao

Jialu Gao

Data Scientist at Microsoft AI

I work on large language model evaluation. My broader interests span trustworthy AI, generative models, computer vision, and robot learning.

I received my M.S. in Robotics from Carnegie Mellon University, where I worked with Prof. Fernando De la Torre. Before that, I earned my B.E. in Computer Science and Technology from Tsinghua University, where I worked with Huazhe Xu.

Google Scholar GitHub ORCID Email
LLM evaluation Generative AI Computer vision Robot learning
Jialu Gao standing by the ocean at sunset

Updates

Recent news

Joined Microsoft AI as a data scientist, focusing on evaluation for large language models.

Completed an M.S. in Robotics at Carnegie Mellon University.

ConBias was published at NeurIPS 2024.

Selected work

Research

My work explores how intelligent systems can generate, understand, and learn from visual and language data.

First page of the MonBench paper Findings of ACL 2026

LLM evaluation

Is Your Language Model Ready for Monetization Decisions?

Jialu Gao, Hua Li, Tracy Ortman, Yeyun Gong, Jian Jiao

MonBench evaluates the performance and robustness of large language models across realistic monetization tasks in search, advertising, and e-commerce systems.

Paper Microsoft Research
Teleportraits method and visual results ICCV 2025

Generative AI

Teleportraits: Training-Free People Insertion into Any Scene

Jialu Gao, K J Joseph, Fernando De la Torre

A training-free method for inserting people into new scenes while preserving identity, appearance, and realistic scene integration.

Paper arXiv Project
Overview of the ConBias framework NeurIPS 2024

Trustworthy AI

Visual Data Diagnosis and Debiasing with Concept Graphs

Rwiddhi Chakraborty, Yinong Oliver Wang, Jialu Gao, Runkai Zheng, Cheng Zhang, Fernando De la Torre

ConBias discovers visual shortcuts through concept graphs and generates targeted data to improve model robustness.

Paper Code
Robot learning tasks solved with generated visual goals NeurIPS 2023

Robot learning · Image generation

Can Pre-Trained Text-to-Image Models Generate Visual Goals for Reinforcement Learning?

Jialu Gao*, Kaizhe Hu*, Guowei Xu, Huazhe Xu

LfVoid turns language instructions into visual goals and uses them as reward signals for learning robotic manipulation policies.

Project Paper
Dual representation robot learning framework CoRL 2022

Human-robot interaction

A Dual Representation Framework for Robot Learning with Human Guidance

Ruohan Zhang*, Dhruva Bansal*, Yilun Hao*, Ayano Hiranaka, Jialu Gao, Chen Wang, Roberto Martin-Martin, Li Fei-Fei, Jiajun Wu

A framework that combines symbolic and continuous representations to make human guidance more effective for robot learning.

Paper Project
CBRAM device characteristics and measurements IEEE TED 2021

Neuromorphic computing

The Origin of CBRAM With High Linearity, ON/OFF Ratio, and State Number for Neuromorphic Computing

Yanming Liu, Jialu Gao, Fan Wu, He Tian, Tian-Ling Ren

An investigation of conductive-bridge memory behavior for reliable, high-density neuromorphic computing.

Paper

Background

Experience & education

2025 - Present

Microsoft AI

Data Scientist

Large language model evaluation and applied AI research.

2023 - 2025

Carnegie Mellon University

M.S. in Robotics

Research in generative models, computer vision, and robot learning.

2019 - 2023

Tsinghua University

B.E. in Computer Science and Technology

Coursework and research spanning AI, robotics, and computing systems.

Get in touch

Interested in my work?

I am always happy to connect with researchers and practitioners working on thoughtful, reliable AI. I usually respond within 1–2 days.

gaojialululu@gmail.com