👩‍🎓 5 Popular Online Courses on Machine Learning 😍

In the previous article we’ve highlighted 6 key characteristics you should consider when choosing an online course. Now we want to bring your attention to some of the most popular online courses on machine learning according to our opinion poll.

💡 Coursera — Machine Learning by Andrew NG (Stanford University)

This is a course that was based on the original Stanford class program and takes up to 11 weeks. It doesn’t require previous knowledge of the field and covers the most important areas, concepts and aspects of machine learning with a significant number of techniques and algorithms. Coursera has an option of applying for financial aid after what, if approved, you can take the course for free. You can also finish the course with no required payment, the only thing is that you won’t get a certificate if you do that.

1. Practical experience

The course author, Andrew Ng, is one of the best known AI experts in the world and the co-founder of Coursera. He has vast experience in the field of machine learning. He was the scientist who founded the Google Brain project where he concentrated on the development of the wide ranging deep learning algorithms. He also was the head of the Badu’s division which developed technologies in different areas including deep learning.

2. Teaching experience

Andrew Ng, among other things, has extensive experience when it comes to teaching due to being a Stanford University’s professor, a Stanford Artificial Intelligence Lab’s Director and a lead developer of Stanford University’s main MOOC platform where he was teaching as well.

Many of those who took his course describe Andrew as a dynamic and gifted instructor who inspires confidence, simplifies concepts and explains complicated subjects in a very clear way. His style of teaching ensures that you’ll be highly involved and never feel bored.

3. Practical aspect of the course

Although it’s a more theoretical course, Andrew explains many things by showing examples, engaging in a motivating discussion and taking through the real life assignments. He gives tips on when to apply various machine learning techniques and mentions the pitfalls that may occur. The course has a lot to do with changing the mindset of approaching the problem, but still leaves a lot more to learn when it comes to implementation.

4. Homework

The evaluation of how good you’re grasping the content of the lectures happens via various quizzes and assignments. The lectures are divided into several videos each and after every lecture and almost every video you have a quiz.

There are 8 assignments that come with the pre-setup environment and can be done just by following the lectures. They use the basic application of machine learning techniques and are very fun and interesting. To complete them you can use MATLAB, which is a paid software, or Octave, which is an open-source free software version.

You can also find the Python assignments for the machine learning class in this github repo.

5. Online Community

There are the community discussion forums where a great number of mentors are willing to help you out if you’re stuck. Most of those people are volunteers who’ve finished the course. There’s little chance though that Andrew will be able to answer you personally in the discussion.

6. Actuality

The course was released in 2011, so there’s a possibility that it’s a bit dated.

💡 Coursera — Applied Machine Learning in Python by Kevyn Collins-Thompson (University of Michigan)

This course takes less time, only a month, and it’s being taught in Python with the introduction of the scikit learn toolkit. It concentrates on the techniques and methods of applied machine learning and requires some prerequisite knowledge of Python programming. The course can be taken for free in which case you won’t get a certificate, or you can apply for financial aid.

1. Practical experience

Kevyn Collins-Thompson has significant experience in the industry due to being a software engineer, researcher (formly at Microsoft Research in the CLUES Group) and manager. His work aims at connecting people with information particularly for educational purposes by the means of specific effective and reliable systems, interfaces and algorithms the development of which is still ongoing and requires the combination of machine learning, natural language processing and human-computer interaction methods.

2. Teaching experience

Kevyn’s work as an Associate Professor in the School of Information and the College of Engineering at the University of Michigan, as well as the affiliate faculty member at the AI Lab and Michigan Institute for Data Science, reflects his broad teaching background. Many of his students refer to him as a clear instructor, when others have a hard time following his explanations. I personally believe this to be a very personal perspective of each student, which you can’t really rely on until you actually start the course and form your own opinion.

3. Practical aspect of the course

The course shows many practically used machine learning methods in Python with examples and the content of lectures and exercises seem to be working quite well to give students an idea of how to actually apply the machine learning theory. But it’s not the full picture. Due to being too packed up with material, the course still lacks practical parallels and real life problem examples to better display theoretical concepts. There’s little time to fully explore the given topics which makes it somewhat difficult for students to focus and completely grasp the presented methods and concepts. Some practical aspects are barely touched and there’s not much context when it comes to explaying of why some situation might happen or why there’s a need for a specific variable.

4. Homework

Every week in the in-browser notebooks you’re faced with quizzes and programming tasks. They play a significant role in strengthening the gained knowledge of concepts and practicing the usage of different methods. If you feel the “but” coming, then you’re correct. Some instructions aren’t very clear and the questions are tricky, all that might cause confusion instead of the consolidation of the learned topics. You can get stuck and spend a significant not actually applying your knowledge, but figuring out your assignment.

Online Community

As has been mentioned when covering the other Coursera course, there’re discussion forums where you can seek assistance.

Actuality

There are some things that are a little dated (like Python modules in the notebooks), but overall — great material. Here you can find the course materials.

💡 edX — Machine Learning by John W. Paisley (Columbia University)

This course is recommended for those who aren’t just good at math, but great at it, ’cause otherwise you won’t get far. So, as you could’ve gathered, there are certain prerequisites, such as strong knowledge of linear algebra, calculus, statistics, probability, well, a sufficient math background. And, of course, you need to know how to code. The course takes 12 weeks and covers a lot of machine learning algorithms and techniques from the mathematical perspective and in great detail, which means that you are most likely to forget a lot of the material and will need to revisit it. Although, you must also know that the course offers nothing on the neural networks. So, if you’re up for the challenge, this one might be for you. It’s free, and in case you want a certificate, you have a chance to pay for it.

1. Practical experience

John Paisley has done a significant amount of research on the statistical machine learning. His work was mostly devoted to the development of probabilistic models, Bayesian models, and posterior inference techniques.

2. Teaching experience

John’s teaching experienced is reflected by his work as an Assistant Professor at Columbia University. He also got a “Distinguished Faculty Teaching Award, Columbia Engineering Alumni Association, 2017”. Many of his students talks highly of him. Professor is noted to have precise language and explain difficult concepts quite clearly, especially given the complexity of the material. He tries to help his students understand the instructions, although for some his presentation might seem very matter-of-fact, which makes the course he’s teaching to be perceived as monotonous. Well, different people require different approaches. For some John’s lack of emotion may be a problem, but if entertainment is not what you’re looking in a course, it shouldn’t bother you.

3. Practical aspect of the course

The course is mostly theoretical as it explains in great detail the math behind machine learning techniques, and there aren’t enough examples. In any case you can get a lot out of this course and the assignments will give you an opportunity to learn how to implement some of the learned algorithms in real life situations.

4. Homework

To get a certificate you need to complete all of the quizzes and assignments. Every week you’ll be faced with a quiz (11 of those), 4 coding projects, for which you can use what’s more comfortable for you, whether it’s Octave, MATLAB or Python, and, of course, a final exam. The questions in the quizzes are hard and quite tricky. Also, you’ll be doing not enough coding, so it might be a good idea to find task by yourself to get more practice.

5. Online Community

Edx has discussion forums where you can ask questions or start discussions and the course TAs, other learners and participants will help you out.

6. Actuality

The course material is solid.

💡 Udacity — Intro to Machine Learning

The course takes up to 10 weeks, but it’s mostly self-paced, so you can take as long as you need to figure everything out. It’s a pretty good start for beginners who have just some basic programming skills. The topics and math here is quite easy, so obviously you won’t dive too deep into the algorithms and implementations, and get only the basic understanding of how things work. But it’s not such a bad thing. If you’re just finding your way to machine learning, then this would be a great starting point to get into the data analysis with Python and get familiar with the sklearn package. The course is free.

1. Practical experience

Katie Malone is a data scientist who works at Civis Analytics leading the research and development departments. She builds general-purpose tools to help solve challenges related to companies science consulting engagements. She also participates in the blue-skies research and hosts podcasts on machine learning and data science.

Sebastian Thrun has gathered a lot of work experience and accomplishments. He’s founded and continue running Udacity, is a CEO of the Kitty Hawk Corporation, worked at Google X where he led a team of computer scientists on the projects like Google Glass and a self-driving car. He also laid a hand on various projects related to AI tech working on healthcare, automated homes and developing robotic systems, among other things. Sebastian gained recognition as the 5th Most Creative Person in Business by Fast Company, one of the 50 Smartest People in Tech by Fortune, and was highlighted in 50 Best Inventions of 2010 by Time.

2. Teaching experience

Sebastian gained his teaching experience by previously working at Carnegie Mellon University, and continuing his work as a Research Professor at Stanford University and Georgia Tech. He was the first one who got the inaugural Smithsonian American Ingenuity Award for Education.

Together with his co-instructor, Katie, they are making the course that much more fun, simple and involving, explaining the material pretty clear.

3. Practical aspect of the course

This is a very practical 15 lessons course for a beginner with great Python examples and not very great math. You won’t get a lot on the mathematics behind the machine learning algorithms, but you’ll gain a good understanding of how everything works explained on the interesting use cases and find out how to use practical tools and deal with the real-world difficulties.

4. Homework

You’ll have to go through various quizzes that follow each lesson. The questions there are quite easy, but their clearness leaves much to be desired. The mini-projects that you’ll also have to deal with aren’t that much harder, and some of them would benefit from additional guidance. The final project involves real-world data. There you need to detect persons of interest in the Enron scandal, which is quite interesting.

5. Online Community

Udacity has a discussion forum and a Udacity Community aimed at collaborative learning, it’s a network of students that help each other learn and overcome challenges.

6. Actuality

Although the course has been released quite a while ago, it’s still considered a great intro for the beginners with not very strong math background.

💡 Fast.ai — Practical Deep Learning for Coders, v3, 2019 edition by Jeremy Howard, Rachel Thomas and Sylvain Gugger

Fast.ai organization founded by Jeremy Howard and Rachel Thomas at the time have four courses on practical machine learning and deep learning, focusing on making AI tools and techniques highly available for everyone interested. This course is a great resource taught in Python that takes 7 weeks and requires some previous programming experience and basic math knowledge. It uses an amazing fastai library, integrated with all the common best machine learning and deep learning practices, and PyTorch.

1. Practical experience

The authors of the course have incredible passion toward their project and gave it their vision and experience. The team currently consists of Jeremy Howard, Rachel Thomas and Sylvain Gugger.

Jeremy Howard is an outstanding practitioner and developer. After 8 years in management consulting he created two quite successful startups: FastMail.FM and Optimal Decisions Group (ODG), both of which he sold (the first one to Opera Software and the second — to Lexis-Nexis). Then he worked at the data science platform Kaggle as the President, Chief Scientist, and was the leading participant in international machine learning competitions. After that Jeremy founded a machine learning company — Enlitic — that worked on applying deep learning tools and techniques to medicine. This startup was named one of the world’s top 50 smartest companies by MIT Tech Review. He also was a co-founder and a researcher at fast.ai. Jeremy has made contributions to various open source projects and continues working with many startups.

Rachel Thomas is a co-founder of fast.ai, where he also keeps a column — ask-a-data-scientist. She worked as a data scientist and a backend engineer at Uber, a full-stack software instructor at Hackbright and as a quantitative analyst in energy trading. Her writing has made the front page of Hacker News and has been featured in newsletters from O’Reilly, Fortune, Mattermark, Metis, and others. Rachel was named one of “20 Incredible Women Advancing AI Research” by Forbes. She participates at various conferences and was a lead speaker at JupyterCon and PyBay.

Sylvain Gugger is currently a Research Scientist at fast.ai, who wrote several textbooks in French on the undergraduate math that were published by Dunod editions. He writes a blog to explain the concepts he’s learned going into machine learning. For fast.ai he reviews the latest papers and looks at what could be used, while he also helps Jeremy develop new functionality in the library and prepare the next course.

2. Teaching experience

Jeremy Howard a great educator and an excellent communicator. He is a Distinguished Research Scientist at the University of San Francisco and the youngest faculty member who teaches data science at Singularity University.

Rachel Thomas is a teacher in Data Science program, an Assistant Professor and a Researcher at the University of San Francisco Data Institute.

Sylvain Gugger is a hardworking Math and Computer Science teacher passionate about data science and especially about the field of deep learning. He taught in CPGE in France for 7 years, where he covered general undergrad math and introduction to programming in Python. He was the first one to whom Jeremy has offered to work at fast.ai full-time, because of his great communication skills, patience, cleverness and the ability to understand complex concepts and distill them down to their essence.

3. Practical aspect of the course

The course is very practical and aimed to immediately engage students and make them as productive as possible. The practical aspects and techniques of deep learning are being covered by Jeremy during the first part of the course, turning to the theoretical explanations to make application more understandable. When you get to the second half, Jeremy immerses you deeper and deeper into the theory without any pressure and confusion. The code-first approach with great examples and explanations on the real-world problems make the course that much more useful and help understand what’s possible and how easy it can be.

4. Homework

Each of the lessons has assignments where you’ll have to deal with various tasks, like natural language processing, computer vision, model building using PyTorch and fastai, and others. And since fastai was included to the Google Colaboratory, which is a free Jupyter notebook environment, you’ll be able to write and execute your code there for free.

5. Online Community

Fast.ai has an amazingly active online community consisting of students and practitioners that are always ready to help and share knowledge and experience on the fast.ai forums, which is now has 21k users.

6. Actuality

The course was released in January 2019, it’s the third iteration, so the material is really new. Furthermore, it contains some applications that no previous intro course has covered before and the techniques you’d be surprised by.

🖐 Conclusion

So, we’ve conducted reviews on 5 highly popular online courses on machine learning based on the main characteristics we’ve listed in the previous article. What do you think about these courses and our assessment of their effectiveness? Share you thoughts in the comments below, and if you have suggestions about other courses, we’ll be happy to hear them.