Hi, I am Kai Huang. I received my Ph.D. in Electrical and Computer Engineering from University of Pittsburgh and my B.E. from University of Science and Technology of China (USTC). I have broad interests in Efficient AI and AI for Systems.

Here is my CV. My research aims to enable efficient inference and fine-tuning of AI models, which can be particularly useful for applications on resource-constrained devices, e.g., STM32 boards, Raspberry Pi, and Nvidia Jetson. My early study focused on AI for wireless communication, where I designed modular and lightweight AI models to enhance the performance of wireless communication systems (e.g., WiFi and backscatter). Later I started to work on On-Device AI, where I optimized the inference and training cost of AI models on weak embedded devices. I also have some research experience in Trustworthy AI. Overall my research involves not only traditional AI models (e.g., MLP, CNN, and LSTM), but also large generative models, such as large language models, diffusion models, and large multimodal models.

Recently I made Generative AI Tutorial and Roadmap, which is a detailed learning guide based on my own experience for generative AI research, including a curated list of state-of-the-art research articles, projects, open-sourced code repositories, and other related research resources. Check this out if you are interested.

Latest News

• [6/2024] Our paper “Perceptual-Centric Image Super-Resolution using Heterogeneous Processors on Mobile Devices” got conditionally accepted to MobiCom 2024!
• [5/2024] Our tutorial for Generative AI Research is released!
• [1/2024] Our paper “Towards Green AI in Fine-tuning Large Language Models via Adaptive Backpropagation” got accepted to ICLR 2024!

Publications

[arXiv] Achieving Sparse Activation in Small Language Models
Jifeng Song, Kai Huang, Xiangyu Yin, Boyuan Yang, Wei Gao
arXiv preprint arXiv:2406.06562
[paper] [code]

[arXiv] FreezeAsGuard: Mitigating Illegal Adaptation of Diffusion Models via Selective Tensor Freezing
Kai Huang, Wei Gao
arXiv preprint arXiv:2405.17472
[paper] [code] [dataset]

[arXiv] Modality Plug-and-Play: Elastic Modality Adaptation in Multimodal LLMs for Embodied AI
Kai Huang, Boyuan Yang, Wei Gao
arXiv preprint arXiv:2312.07886
[paper] [code]

[MobiCom’24] Perceptual-Centric Image Super-Resolution using Heterogeneous Processors on Mobile Devices
Kai Huang, Xiangyu Yin, Tao Gu, Wei Gao
The 30th Annual International Conference On Mobile Computing And Networking
Acceptance Rate: 103/494=20.9%
[paper]

[ICLR’24] Towards Green AI in Fine-tuning Large Language Models via Adaptive Backpropagation
Kai Huang, Hanyun Yin, Heng Huang, Wei Gao
The Twelfth International Conference on Learning Representations
Acceptance Rate: 2250/7304=30.8%
[paper] [talk] [code]

[MobiSys’23] ElasticTrainer: Speeding Up On-Device Training with Runtime Elastic Tensor Selection
Kai Huang, Boyuan Yang, Wei Gao
Proceedings of the 21st International Conference on Mobile Systems, Applications, and Services
Acceptance Rate: 41/198=20.7%
🎉 Received ACM Artifact Available, Functional, Reusable, Results Replicated Badges (4/17)
[paper] [slides] [teaser] [code]

[MobiSys’23] PTEase: Objective Airway Examination for Pulmonary Telemedicine using Commodity Smartphones
Xiangyu Yin, Kai Huang, Erick Forno, Wei Chen, Heng Huang, Wei Gao
Proceedings of the 21st International Conference on Mobile Systems, Applications, and Services
Acceptance Rate: 41/198=20.7%
[paper] [teaser] [code]

[MobiCom’22] Real-time Neural Network Inference on Extremely Weak Devices: Agile Offloading with Explainable AI
Kai Huang, Wei Gao
The 28th Annual International Conference On Mobile Computing And Networking
Acceptance Rate: 56/314=17.8%
[paper] [slides] [talk] [code]

[IoTDI’22] RAScatter: Achieving Energy-Efficient Backscatter Readers via AI-Assisted Power Adaptation
Kai Huang, Ruirong Chen, Wei Gao
The 7th ACM/IEEE Conference on Internet of Things Design and Implementation
Acceptance Rate: 12/36=33.3%
[paper] [slides] [code]

[SenSys’22] AiFi: AI-Enabled WiFi Interference Cancellation with Commodity PHY-Layer Information
Ruirong Chen, Kai Huang, Wei Gao
Proceedings of the 20th ACM Conference on Embedded Networked Sensor Systems
Acceptance Rate: 52/209=24.8%
[paper] [slides] [code]

[CML-IOT’22] Out-Clinic Pulmonary Disease Evaluation via Acoustic Sensing and Multi-Task Learning on Commodity Smartphones
Xiangyu Yin, Kai Huang, Erick Forno, Wei Chen, Heng Huang, Wei Gao
The Fourth Workshop on Continual and Multimodal Learning for Internet of Things
🎉 Best Paper Award
[paper]

[IPSN’22] FaceListener: Recognizing Human Facial Expressions via Acoustic Sensing on Commodity Headphones
Xingzhe Song, Kai Huang, Wei Gao
21st ACM/IEEE International Conference on Information Processing in Sensor Networks
Acceptance Rate: 38/126=30.2%
[paper]

[ASPLOS’22] Eavesdropping user credentials via GPU side channels on smartphones
Boyuan Yang, Ruirong Chen, Kai Huang, Jun Yang, Wei Gao
Proceedings of the 27th ACM International Conference on Architectural Support for Programming Languages and Operating Systems
Acceptance Rate: 80/397=20.2%
[paper] [talk] [code]

[MobiSys’20] MagHacker: eavesdropping on stylus pen writing via magnetic sensing from commodity mobile devices
Yihao Liu*, Kai Huang*, Xingzhe Song, Boyuan Yang, Wei Gao
* indicates equal contributions
Proceedings of the 18th International Conference on Mobile Systems, Applications, and Services
Acceptance Rate: 34/175=19.4%
[paper]

Education

Ph.D., Electrical and Computer Engineering, University of Pittsburgh, Sept. 2019 - May 2024

B.E., Electronic Information Engineering, University of Science and Technology of China (USTC), Sept. 2015 - July 2019

Visitors