TinyML and Efficient Deep Learning Computing
6.S965 • Fall 2022 • MIT
Have you found it difficult to deploy neural networks on resource-constrained hardware? Have you ever found it too slow to train neural networks? This course is a deep dive into efficient machine learning techniques that enable powerful deep learning applications on resource-constrained devices. Topics cover efficient inference techniques, including model compression, pruning, quantization, neural architecture search, and distillation; and efficient training techniques, including distributed training, gradient compression and on-device transfer learning; followed by application-specific model optimization techniques for video, point cloud, generative model, NLP and LLM; it will cover futuristic research on quantum machine learning. Students will get hands-on experience implementing deep learning applications on mobile devices with an open-ended design project related to efficient AI computing.
- Time: Tuesday/Thursday 3:30-5:00 pm Eastern Time
- Location: 36-156
- Office Hour: Thursday 5:00-6:00 pm Eastern Time, 38-344 Meeting Room
- Discussion: Piazza
- Homework submission: Canvas
- Online lectures: The lectures will be streamed on YouTube.
- Resources: MIT HAN Lab, HAN Lab Github, TinyML, MCUNet, OFA
- Contact: Students should ask all course-related questions on Piazza. For external inquiries, personal matters, or emergencies, you can email us at [email protected].

- Instructor Song Han
- Email: [email protected]

- TA Zhijian Liu
- Email: [email protected]

- TA Yujun Lin
- Email: [email protected]
Announcements
Dec 5, 2022 | Lab 3 NAS Solution and Lab 4 Deployment on MCU Solution are released! |
Nov 8, 2022 | Lab 4 Deployment on MCU is released! Due date is Nov. 22. |
Oct 27, 2022 | Lab 2 Quantization Solution is released! |
Oct 25, 2022 | Midterm Feedback Form is out! Due Nov. 1 11:59 pm ET. Please let us know what you consider to be the high and low-points of the subject to date. |
Oct 25, 2022 | Lab 3 NAS is released! Due date is Nov. 10. |