Multimodal LLMs & Agents
Multimodal large language models for visual understanding, and multimodal agents with visual memory for photo album applications.
COMPUTER SCIENCE · BOSTON UNIVERSITY
Aoming Liu
Efficient AI & Multimodal LLMs
I am a Ph.D. student in Computer Science at Boston University, advised by Prof. Bryan Plummer and Prof. Boqing Gong. I also work with Prof. Kate Saenko. I expect to graduate in 2027.
My research lies at the intersection of Efficient AI and Multimodal LLMs. I study how to make multimodal models and agents more efficient during both training and inference, and how to help them adapt to new data and generalize across changing distributions with minimal additional computation and supervision.
I received my M.Sc. from ETH Zurich and my B.E. from Xi’an Jiaotong University. My industry experience includes research internships at Adobe, AMD, and OPPO, as well as full-time work at TuSimple on perception for autonomous driving.
ON THE JOB MARKET
I am seeking full-time Research Scientist or Machine Learning Engineer positions in industry, starting in summer or fall 2027. Let’s get in touch.
Research interests
Efficient training, inference, and adaptationEfficient training & inference
Data-efficient pretraining and post-training, and computation-efficient inference through operation pruning and model compression.
Adaptation & generalization
Adapting to new data and generalizing across domains and time with limited resources, including through model weight averaging.
Publications
Google ScholarICLR2026
Noise-Aware Generalization: Robustness to In-Domain Noise and Out-of-Domain Generalization
Learning to generalize when training data contains both domain shifts and noisy labels.
EMNLP2025
Scaling Up Temporal Domain Generalization via Temporal Experts Averaging
Combining temporal experts to improve generalization to future distributions.
ICCV2025
BabyVLM: Data-Efficient Pretraining of VLMs Inspired by Infant Learning
Data-efficient vision-language pretraining inspired by how infants learn.
ECCV2024
PanoFree: Tuning-Free Holistic Multi-view Image Generation with Cross-view Self-Guidance
Consistent multi-view and panoramic image generation without additional model tuning.
ICCV2021
Direct Differentiable Augmentation Search
Efficiently searching augmentation policies for image classification and object detection.
AAAI2021
Neural Architecture Search as Sparse Supernet
Discovering compact neural architectures through sparsity-constrained supernet search.
* Equal contribution
Working papers
IN SUBMISSION
Fine-grained Token Allocation Via Operation Pruning for Efficient MLLMs
Allocating computation at the operation level to make multimodal models more efficient.
IN PREPARATION
AlbumEMM: Benchmarking Event Modeling and Memory Construction from Discrete Photo Collections
A benchmark for event modeling and memory construction from photo collections, exploring multimodal agents with visual memory for photo album applications.
Internships & work experience
May 2026 – Present
Adobe Research Internship
Exploring multimodal agent systems with visual memory for photo album applications.
Oct 2025 – May 2026
AMD Research Internship
Post-training multimodal LLMs for visual document understanding tasks, including the construction of a 10M-scale instruction-tuning dataset.
Jun – Aug 2023
OPPO Research Institute Research Internship
Developed a training-free panorama generation algorithm, leading to PanoFree at ECCV 2024.
TuSimple
Full-timeSep 2021 – Aug 2022
Worked on 3D perception for autonomous driving, including LiDAR-based velocity estimation, model training, compression, and deployment.
Research InternshipMar – Nov 2020
Developed automated augmentation search for object detection, leading to work published at ICCV 2021.
Education
Boston University
Ph.D. in Computer Science
2022 – 2027 (expected)
Boston, Massachusetts
ETH Zurich
M.Sc. in Robotics, System and Control
2018 – 2021
Zurich, Switzerland
Xi’an Jiaotong University
B.E. in Electrical Engineering and Automation
2014 – 2018
GPA: 3.98/4.00 · Rank: 3/340
Academic service
Reviewer for CVPR, ICCV, ECCV, ICLR, NeurIPS, and AAAI.
Let’s connect.
I welcome conversations about Research Scientist and Machine Learning Engineer opportunities in industry.
amliu@bu.edu