Repository files navigation

This learning schedule is sorted out forreseachers orwhoever interested in machine learning / deep learning / computer vision. It's also most welcomed forinstructors to refer to this schedule to train beginner students. Each course listed below takes abouttwo weeks to finish if you are fully dedicated to it.

The skills you need to develop amachine learning / deep learning / computer vision project include:

• Coding skills. Coding is not the objective, but the tool. Without the tool, nothing can be built.
• Mathematics. Math is the foundation of machine learning that you can't evade, among which statistics and probability are the most important.
• Machine learning algorithms, which is the focus of most research or competitions.
• Paper reading and writing, which involves English proficiency and professionalisim. You may not necesssarily publish a paper on a journal or a conference. However, to keep up with others' work and report your work, you have to be familiar with how to read and write a paper.

These courses are for beginners. It's suggested to follow the order below. Try to finish the coding and mathematics homework in each course. Everyone should have covered the following staffs before your REAL journey to machine learning. It does not mean you have to take the following courses, we recommend you to do so; but if you are not totally fresh, you should know the contents.

• Andrew Ng (Stanford)'s Machine LearningHomepage video link:YoutubeBilibili

Quick intro to machine learning, no math but a very good overview about what ML is and what ML does. We recommend approx. 2 weeks for study. The MATLAB assignments are not longer recommended.

• Feifei Li (Stanford)'s Computer Vision (CS231n)Homepage video link:YoutubeBilibili

Quick intro to deep learning, with emphasis on computer vision. You could do some interesting staffs after taking this course (and code practice!), e.g., face recognition, pedestrian detection and segmentation, medical image diagnosis, image generation.

After finishing the above courses, it's highly suggested to join some simple competitions before you keep going onKaggle, a well-known data science competition website. You can refer to others' code for inspiration. Free GPU resources are also available on Kaggle. For beginners, theGetting Started category is the best place to obtain project experience and practice coding skills. The following two competitions are good basic options.

• Digit Recognizer A classification task based on hand-written digit images. A convolutional neural network might be involved. For this competition, we provide somereference code with different mahcine learning computing package.
• Titanic: Machine Learning from Disaster A classification task based on structured data.

We recommend ML freshman should know the following packages:

• scipy-family (numpy, scipy, pandas, matplotlib, etc.)
• scikit-learn (for out-of-box ML tools, models, metrics, etc.)
• one of deep learning packages: PyTorch or Keras (other DL packages are recommended for beginners!)

Now we need to expand our sight to the current research topics inmachine learning / deep learning / computer vision.

• Hsuan-Tien Lin (NTU)'s Machine Learning FoundationsHomepage video link:YoutubeBilibili

• Hsuan-Tien Lin (NTU)'s Machine Learning TechniquesHomepage video link:YoutubeBilibili

• Hung-yi Lee (NTU)'s Deep LearningHomepage video link:YoutubeBilibili

By far, you should be familiar with the basic concepts ofmachine learning / deep learning / computer vision. You might need to participate in a real project in a lab at school (choose a reputed lab carefully) or in a IT company. You may also consider join a more advanced competition onKaggle.

Here, we provide aPyTorch coding template in python for developing a real project.

Don't rush to dig into these advanced courses. These courses are more specific for certain topics. Only after you have several project experiences, can these advanced courses help you build up a systematic sense of these topics.

• Yida Xu (UTS)'s Probabilities and Machine Learning video link:YoutubeBilibili
• Hung-yi Lee (NTU)'s GAN 2018YouTube
• Hung-yi Lee (NTU)'s Next Step of Machine LearningYouTube
• CS 294-131: Trustworthy Deep LearningHomepage
• CMU 10-708 PGM (19) by Eric XingHomepage
• Berkely Deep RL BootcampHomepage
• CS294-158 Deep Unsupervised Learning Spring 2019Homepage
• Udacity's Cuda (Homepage)
• Cousera's Programming LanguageHomepage
• Udacity's Design of Computer ProgramsHomepage

At this point, you have mastered the basic skill and knowledge required formachine learning / deep learning / computer vision research. But there are still so much unknown placed waiting for you to explore. What you learn here merely provides you withthe way leading to those places. Begin you adventure now! And enjoy the beauty of maching learning!Ads

• git (best tutorial I've read in Chinese, haven't found a counterpart in English)

• How to use remote server
• How to set up a brand new machine, from installing Linux to installing Pytorch （Here is a rough instruction in Chinese) (Here is the packages list required to set up a system for machine learning research)
• Advanced coding courses
• Papers reading and writing
• Code convention

Any advice or comments to improve this learning schedule is most welcomed.

• Xiaoyang Huang

• Jiancheng Yang who provides the primary study route and first start this project.
• Linguo Li who provides the MNIST reference code and packages list.