Large Model Compression & Acceleration
Pruning, quantization, distillation, sparsity, low-rank adaptation, and deployment-aware acceleration for foundation models, vision backbones, and edge AI systems.
East China Normal University
EIC Lab
Edge Intelligent Computing Lab
We build compact, fast, and reliable AI systems for edge intelligence, spanning large model compression, efficient image restoration and generation, intelligent agents, and AI4Science.
4
Research directions
ECNU
East China Normal University
Edge + Cloud
Efficient intelligent systems
Research directions
Edge intelligence from algorithms to systems
Efficient Image Restoration & Generation
Compact networks for super-resolution, dehazing, enhancement, denoising, restoration, and controllable generation under real compute budgets.
Intelligent Agents
Multimodal reasoning, tool use, planning, and evaluation pipelines that make agents more reliable, inspectable, and efficient for real-world workflows.
AI4Science
Efficient learning systems for scientific data, simulation surrogates, discovery workflows, and domain models where accuracy and compute both matter.
Open source
Representative projects
Agents · Multimodal reasoning
Vision-R1
A multimodal reasoning project focused on incentivizing reasoning capabilities in vision-language models.
Repository
Compression · Interpretability
KSE
Exploiting kernel sparsity and entropy for interpretable CNN compression, accepted by CVPR 2019.
Repository
Compression · Pruning
Generative Adversarial Learning for CNN Pruning
Structured neural network pruning for compact and deployable convolutional models.
Repository
Compression · Dynamic pruning
Global & Dynamic Filter Pruning
Accelerating convolutional networks through global and dynamic filter pruning, from IJCAI 2018.
Repository
Compression · Low-rank
LRDKT
Holistic CNN compression via low-rank decomposition with knowledge transfer, published in TPAMI.
Repository
Compression · Sparsity
SSR
Structure-sparsity regularized filter pruning for compact ConvNets, published in TNNLS.
Repository
Image · Dehazing
AECR-Net
Contrastive learning for compact single image dehazing, connected to the low-level vision direction on Shaohui Lin's publication list.
RepositoryWorking style
From publishable ideas to usable releases
EIC Lab at East China Normal University emphasizes reproducible code, practical acceleration, and systems that can be tested beyond benchmark tables. The site is structured so new papers, datasets, demos, and leaderboards can be added as the lab grows.
People
Advisor and team
Shaohui Lin
Research interests include model compression and acceleration, efficient image restoration, generation, multimodal intelligent systems, and edge intelligent computing.
Students, interns, and collaborators
Add member cards here with research topics, personal pages, and selected publications.
Join us
Build efficient AI with real impact
We welcome students and collaborators interested in efficient foundation models, low-level vision, agents, scientific AI, and deployable edge intelligence.