This course gives you an overview on the emerging area of "Efficient ML," i.e., building effective ML models under various resource constraints (e.g., active GPU RAM, total training FLOPs, inference latency). The main focus is on the algorithmic advances for efficient ML, while we still discuss relevant issues on the system/hardware side.

No formal requirements, but I expect you to have some elementary knowledge on machine learning and deep learning.
If you took EECE454 (or AIGS515/538 or CSED490X/Y), you should be okay.
Feel free to contact me if you are worried :)

Homework Assignments: 30%
In-class Presentation: 30%
Final Report: 30%
Participation: 10%

Sze, Chen, Yang, and Emer, "Efficient Processing of Deep Neural Networks," Morgan Claypool, 2020.
Menghani, and Singh, "Efficient Deep Learning," (not-published-yet), 2022.

W1: Introduction to EfficientML & Basics of Deep Learning
W2-3: Compute and Memory in Deep Learning, Scaling Law
W4-5: Network Architectures / NAS
W6: Weight Initialization, Transfer Learning, and Meta-Learning
W7: AutoML, Hyperparameter Tuning
W8: Data and Model Parallelism in Deep Learning
W9-11: Model Compression
W12: Efficient Training
W13: TinyML
W14-15: Student Presentations
W16: Final Report

- Final Report: A short survey of a small sub-field of EfficientML, that is most relevant to your research or interest.
- Student Presentation: A 20-minute-long talk about a paper published in NeurIPS, ICML, ICLR or MLSys (2020-2022), or a more recent preprint.
- Homework Assignments: Involves coding and math; get ready for it!
- Participation: Not about attendance (you are all professionals), but about an active discussion.