Shuai Zhang

Researcher and team lead at Qualcomm AI Research working on Edge AI, on-device agentic pipelines, computer vision, and numerical optimization. Organizer of LPCVC and co-organizer of the ECV Workshop at CVPR-2025 & CVPR-2026.

Edge AI Multimodal Perception Agentic Pipeline Optimization and Quantization Foundation Models

1,528+

Citations

h-index

i10-index

Yrs @ Qualcomm

Publications Scholar LinkedIn CV

Recent Work

arXiv · 2026 New

Latent Bridge: Feature Delta Prediction for Efficient Dual-System VLA Model Inference

Vision-Language-Action · Efficient Inference · Robotics

arXiv · 2026 New

ForeSea: AI Forensic Search with Multi-modal Queries for Video Surveillance

Multi-modal · Video Retrieval · Surveillance AI

arXiv · 2026 New

DiSa: Saliency-Aware Foreground-Background Disentangled Framework for Open-Vocabulary Semantic Segmentation

Open-Vocabulary Segmentation · Vision-Language · Saliency

arXiv · 2025 SEMA: A Scalable and Efficient Mamba-like Attention via Token Localization and Averaging

Efficient Attention · ViT · State Space Models

AAAI · 2024

MobileInst: Video Instance Segmentation on the Mobile

Mobile AI · Instance Segmentation · Real-time

All publications

Research Areas

Agentic AI

VLA models & on-device agents

Computer Vision

Detection, segmentation, tracking

Edge AI

Quantization, pruning, efficiency

Optimization

Non-convex & compressed sensing

Highlighted Earlier Work

ICLR 2019

Understanding Straight-Through Estimator in Training Activation Quantized Neural Nets

KDD 2020 Most Influential

AutoShuffleNet: Learning Permutation Matrices via Lipschitz Continuous Penalty in DNNs

Math. Programming 2018 Minimization of Transformed L1 Penalty: Theory, DCA & Compressed Sensing