dongyyyyy/ca_seminar

Computer Architecture Seminar (Graduate Course) @ Hallym University

Computer Architecture Seminar (Graduate Course) @ Hallym University

컴퓨터 구조 세미나: 한림대학교 컴퓨터공학과 대학원, Spring 2019

최신 인공지능 기술인 딥러닝의 가속기 구조(Accelerator Architecture)에 대한 수업 및 프로젝트를 진행합니다.

• Instructor : Prof. Jeong-Gun Lee
  • Email: [email protected]
• Time : 5:00 PM - 8:00 PM, every Thuesday
• Place : 1219

This class is for learning recent advanced deep learning algorithm and computer system architecture for accelerating the deep learning algorithm. From the half point of the course, each student is expected to conduct an individual project on object detection. Particularly the topic of people counting or crowd counting is recommended. Please check the following Github sites

• https://github.com/search?q=people+counting
• https://github.com/search?q=crowd+counting

and the following youtube.

• https://www.youtube.com/watch?v=u_XOczajfRI
• https://www.youtube.com/watch?v=j2f4QGVSJCk
• https://www.youtube.com/watch?v=la50CZ9-NIo : Source is available.

When you choose a paper, you try to select a paper whose idea has been implemented by open source software. So you can test the idea on your machine.

Requirement

• Every students have to do paper presentations
• Every students have to do class project

---

Weekly Plan

• 1st Week (2019-02-26): Introduction to class

• 2nd Week (2019-03-05): Basics of Deep Learning

  • MIT 6.S191: Introduction to Deep Learning: From 6:34
  • Student Presentation
    • Dongyoung Kim, The Effect of regularization and identity mapping on the performance of activation functions

• 3rd Week (2019-03-12): Running Deep Learning & How the hardware for the accelerator can be optimized.

  • MIT 6.S191: Convolutional Neural Networks
  • Deep Learning with Python, TensorFlow, and Keras tutorial
  • Student Presentation (2)
    • Yunhee Woo, End-to-end crowd counting via joint learning local and global count, End-to-end people detection in crowded scenes
      • Related Youtube Movie Clip

• 4th Week (2019-03-19): Topic

  • Introducing TensorFlow 2.0 and its high-level APIs (TF Dev Summit '19)
  • TensorFlow Datasets (TF Dev Summit '19)
  • Efficient Methods and Hardware for Deep Learning Part I

• 5th Week (2019-03-26): Topic

  • Efficient Methods and Hardware for Deep Learning Part II

• 6th Week (2019-04-02): Topic

  • Efficient Methods and Hardware for Deep Learning Part III
  • Student Presentation (2)
    • JaeMin Jeong, Multi-Column Convolutional Neural Network [paper]

• 7th Week (2019-04-09): Topic

  • C4W1L01 Computer Vision
  • Student Presentation (2)
    • Jinwoo Lee, Evaluating the Energy Efficiency of Deep Convolutional Neural Networks on CPU and GPU

• 8th Week(2019-04-16): Midterm.

• 9th Week: Topic

  • Video
  • Student Presentation (2)
    • Student 1,
    • Student 2,

• 10th Week: Topic

  • Video
  • Student Presentation (2)
    • Student 1, DongYoung Kim, 'YOLOv3 : An Incremental Improvement' [paper]
    • Student 2,

• 11th-14th: Student Project using Deep Learning and its accelerator

  • You can use high performance GPU (1080Ti or 2080Ti RTX) for training
  • You can use Raspberry Pi and Intel Movidius Compute Stick for embedded smart vision processing
    • https://www.movidius.com/
    • Deep Learning with Movidius NCS and Raspberry Pi3B+ (pt.1) VM and NCSDK setup

• 15th: Final Presentation for Your Project

---

References

• Stanford Deep Learning Class
• http://eyeriss.mit.edu/tutorial.html