Course Information

---

This course is designed as the first course for students who are interested in deep learning technology. It mainly focus on some basic components of modern deep neural networks, as well as the relevant applications from a practical perspective. While the advanced topics cover the cutting-edge technology in recent years, giving students the opportunity to know the trend of technology development. In particular, the topics include:

Convolutional Neural Network

Computer Vision Basic

Recurrent Neural Network

Natural Language Processing Basic

Attention Mechanism and Transformer

Pretrained Language Model

Self-supervised Learning

Unified Language and Vision Model

Prerequisites

Basic machine learning knowledge: It requires the students have basic knowledge of machine learning.

Proficiency in LaTex: All the reports need to be written by using LaTex. A template will be provided. If you are not familiar with LaTex, please learn from the tutorial in advance.

Proficiency in Python: All the assignments will be in Python, especially the PyTorch framework.

Proficiency in Openmmlab: All the assignments will be realized on the Openmmlab platform

Proficiency in GitHub: All the source codes need to be submitted in GitHub.

Textbooks

Recommended Books

Deep Learning, by Ian Goodfellow and Yoshua Bengio and Aaron Courville

Dive into Deep Learning, by Aston Zhang, Zachary C. Lipton, Mu Li, and Alexander J. Smola

Grading Policy

Assignments (30%)

Assignment 1 (15%) Image classification by using deep models

Assignment 2 (15%) Language processing and classification by using deep models

Midterm Exam (25%)

We will have 20~30 questions about the basic concept. The scope of the mid-term exam is from lecture 1 to lecture 9 (including Transformer).

Final Project (40%)

The final project is teamwork with no more than 3 teammates. You need to write a project report (max 5 pages not including references) for the final project. Here is the report template. You also need to prepare a “10 minutes” code explanation (online mode) to explain the core content of your code, its reasonableness, and the steps to run it, which is highly relevant to the quality and readability of your code. After the final project deadline, feel free to make your project open source.
【Note】If your team submits a project report after the submission deadline, the maximum project credit for all team members is 50%.

Code reproducibility (15%): Determine code quality and readability through visual inspection and code explanation. The code needs to ensure that it can be executed effectively and is consistent with the experimental results in the report (10%). The structure of the code and the functions implemented need to be clear and easy to understand (5%). Teams that choose the same topic are not allowed to plagiarize, otherwise, all members of the corresponding two teams will fail the course.

Project reprort (25%): Please follow the report template to organize your content and ensure the following items.

(1) Reference to prior work (3%): Cite existing related work and describe the lineage of technology development， as well as their relations to your algorithms.

(2) Technical correctness (4%): Present your algorithms or systems for your project. Provide key information for judging whether it is technically correct.

(3) Experimental validation (5%): Present your experimental setups, and experiments conducted. Discuss the motivation of the designed experimental setups and the corresponding evaluation metrics.

(4) Findings and Analysis (8%): Report the final results of your algorithms (performances and visualization results). Analyze and understand your system by inspecting key outputs and intermediate results (ablation study). Discuss how it works, when it succeeds and when it fails, and try to interpret why it works and why not.

(5) Clear in writing (2%): The report is written in a precise and concise manner so the report can be easily understood.

(6) Overall quality of report (3%): An overall evaluation of the final project report, including but not limited in writing, creativity, convincing experiments and analysis.

Originality and creativity (5%): Original and creative ideas are encouraged. If you try to leverage some techniques outside the scope of this course (2%) and correctly use them to improve your algorithms (3%), you will get extra scores.

Participation (5%)

Here are some ways to earn the participation credit. Note that the relevant scores are used to make up for grades.

Completing feedback surveys: We will send out two feedback surveys during the semester to improve the course. All students get 1.5% per survey (in total 1%) for completing the surveys.

Course and Teaching Evaluation (CTE): The school will send requests for CTE to all students. The CTE is worth 1% credit.