Generalization & Adaptation under Distribution Shift
How can models remain effective when domains, tasks, and environments differ from their training conditions?
About
Shaobo Han is a Senior Researcher in the Machine Learning
Department at NEC Laboratories America. His research
focuses on generalization, adaptation, and efficient AI.
He works on foundation-model adaptation, small multimodal
models, probabilistic machine learning, and AI for
physical sensing systems. His work connects advances in
machine learning to real-world applications, including
large-scale sensing using deployed telecommunication
fiber networks.
Education
Ph.D. in Electrical and Computer Engineering
Duke University
M.S. in Statistical Science
Duke University
M.Eng. in Signal and Information Processing
University of Chinese Academy of Sciences
Research Interests
Generalization and Adaptation
Foundation-Model Adaptation and Post-Training
Multimodal Learning and Reasoning
Probabilistic and Generative Models
AI for Physical Sensing Systems
Research
Generalization, Adaptation & Efficiency | Foundation Models, Multimodal AI & Physical Sensing
Efficient Foundation-Model Adaptation & Post-Training
How can knowledge from pretrained and specialized models be adapted, compressed, and consolidated efficiently and reliably?
Small Multimodal Models: From Recognition to Reasoning
How far can small multimodal models progress from open-vocabulary recognition toward understanding and reasoning?
Sensing AI for Real-World Physical Systems
Machine learning for optical fiber sensing over deployed telecommunication networks.
Selected Publications
Uni-LoRA: One Vector is All You Need
Reparameterization and randomized parameter sharing for extreme parameter-efficient fine-tuning.
VB-LoRA: Extreme Parameter Efficient Fine-Tuning with Vector Banks
Vector-bank parameter sharing for extreme parameter-efficient fine-tuning.
Uncertainty-Aware Knowledge Distillation for Multimodal Large Language Models
Uncertainty-aware knowledge distillation that adaptively balances data supervision and teacher guidance.
CLAP-S: Support Set Based Adaptation for Downstream Fiber-Optic Acoustic Recognition
Combines support-set exemplar memory with parametric adaptation for pretrained audio-language models under severe domain shifts.
Learning Transferable Reward for Query Object Localization with Policy Adaptation
Transferable reward learning with test-time policy adaptation to unseen target classes.
Variational Gaussian Copula Inference
Variational inference for flexible dependence modeling with Gaussian copulas.
Deep Learning-Based Intrusion Detection and Impulsive Event Classification for Distributed Acoustic Sensing Across Telecom Networks
Event recognition for distributed acoustic sensing, studying generalization under different sensing conditions and validated through field experiments.
Manhole Localization and Condition Diagnostics in Telecom Networks Using Distributed Acoustic and Temperature Sensing
Structured prediction for infrastructure localization and condition diagnostics using spatial and temporal sensing context.
Selected Talks & Tutorials
Machine Learning for Optical Fiber Sensing: Generalization, Adaptation, and Real-World Deployment
Invited Tutorial at the 30th International Conference on Optical Fiber Sensors (OFS30).
Raleigh, NC · Nov 30, 2026
Extreme Parameter-Efficient Fine-Tuning of Foundation Models
Invited Talk at the POSTECH–KAIST–NEC Joint Research Workshop.
Seoul, South Korea · Jul 9, 2026
Generative AI for Distributed Acoustic Sensing over Telecom Networks
Invited Speaker and Panelist at the OECC/PSC 2025 Workshop.
Sapporo, Japan · Jun 29, 2025