CENG5030 slogan 

CENG5030 Energy Efficient (Deep Neural Network) Computing – Spring 2021


Lecture: M 12:30-14:15 Venue: zoom
W 16:30-18:15 Venue: zoom
Course Instructor: Prof. Bei Yu byu@cse.cuhk.edu.hk
Course Tutors: Qi Sun qsun@cse.cuhk.edu.hk


Announcements

  • Feb. 24, 2021: Course schedule is updated.

  • Jan. 18, 2021: Grading policy is updated (in-class discussion and decent project are encouraged).

  • Dec. 01, 2020: Course webpage is built up and the teaching schedule is online.

Description:

This course is an intensive and research oriented course, discussing some practical and fundamental techniques, skills, and tools for deep neural network acceleration, in particular inference acceleration. The students selecting this course are assumed to have solid DNN experience and some programming skills (e.g, C/C).

Course Requirements

Schedule

Week Date Topic Remark
1 Jan. 11 L01 Introduction (slides)
Jan. 13 Lab01 PyTorch (report, code) Due: Jan. 27
2 Jan. 18 Lab02 Training (report, code) Due: Feb. 02
Jan. 20 L02 CNN Training (slides)
3 Jan. 25 L03 Pruning (slides)
Jan. 27 L04 Accurate Speed-up I (slides)
4 Feb. 01 continue on L04
Feb. 03 Lab03 GEMM (report, code) Due: Feb. 17
5 Feb. 08 L05 Quantization (slides)
Feb. 10 L06 Binary/Tenary Networks (slides)
6 Feb. 15 n/a Lunar New Year Holiday
Feb. 17 n/a Lunar New Year Holiday
7 Feb. 22 Lab04 Distiller (report, slides, code) Due: Mar. 08
Feb. 24 under reschedule
8 Mar. 01 Lab05 MNN (report, code)
Mar. 03 L07 Decomposition (slides) 16:30–18:45
9 Mar. 08 L08 Accurate Speed-up II (slides)
Mar. 10 continue on L08 16:30–18:45
10 Mar. 15 Lab06 TensorRT (report, slides, code)
Mar. 17 Midterm Presentations
11 Mar. 22 Midterm Presentations
Mar. 24 Lab07 TVM (report, code)
12 Mar. 29 n/a Reading Week
Wed. 31 n/a Reading Week
13 Apr. 05 n/a The day following Ching Ming Festival
Apr. 07 n/a Reading Week
14 Apr. 12 L09 TVM (slides)
Apr. 14 n/a
15 Apr. 19 L10 NAS (slides)
Apr. 21 Final Presentations
16 Apr. 26 n/a
Apr. 28 Final Presentations Final report Due: May 9th