How do I get started with machine learning? This video is a three-month guide which will help you go from an absolute beginner to proficient in the art of machine learning. This video describes that this is for those who can only dedicate part-time to learning, either have a job or B have to go to school.
If you are able to go full time, you can complete this in half the time. When you will start learning AI today this is an open source for you which will help you out in your learning. Let's start by looking at the hiring page for the deep mind: the worlds leading institution when it comes to publishing state-of-the-art AI algorithms.
If we look under the research engineer position, well see a description for what it takes to be a part of the team responsible for Alpha go, the deep Q network and Wavenet, each of which revolutionized the field of AI. The minimum qualifications are having a bachelors in computer science or a related field. Strong knowledge of Python, machine learning and algorithmic design and no need to have a Ph.D. or have published a ton of papers even under the preferred qualifications.
If we were to create a pie chart to try and understand the necessary topics for machine learning it would look something like this thirty-five percent of it would consist of linear algebra, twenty-five percent would include probability theory and statistics, fifteen percent would be calculus, another fifteen percent would be algorithms and complexity, and the last ten percent would be dedicated to data preprocessing knowledge. We can build our own curriculum around this chart.
This curriculum is divided up into three months:
The first month consists of math and algorithmic complexity. The second month consists of machine learning and the third month consists of the most popular subset of machine learning, Deep learning.
But before we begin I want to first say that one of your daily tasks should be to keep up with the field. So here are the best resources to do that, First is the youtube channel you will get the content of artificial intelligence, here nonstop every single week. Next is the machine learning subreddit. Its a place where people working both in academia and in business on ML. Twitter is also a really underutilized resource, as Twitter acts as a learning feed. If you use it the right way, follow really smart people and scrolling down your feed unlike most social platforms won't drain you, in fact, it will enrich you with knowledge.
Machine Learning researchers love to use Twitter there are some really good academic debates that happen there as well. Some examples of researchers, you should follow are Ilya Sutskever, Trent McConaughey, Andrej Carpathia, Andrew Trask, Petera, Biel, Chris Ola, and Nando de Freitas.
There are many more that are linked in the description. Hacker news is another must-have. The audience is well versed in the technical details of the topics mentioned, so you know that the only the best arguments rise to the top.
All of the resources that are suggested to you are going to be video based. You will get much better-using videos and the short form is written content on the web and by using a plain textbook that just lists equation.
Now on to the first month learning math well start with the portion that makes up the most of machine learnings linear algebra. One of the most popular courses on MIT OpenCourseWare
website is the linear algebra course taught by Gilbert strong. He's got a kind of infectious enthusiasm for this stuff and there are 35 video lectures available on YouTube. As a playlist watch each one at 2x speed, if possible 3x, if you can handle it take the time to handwrite your notes as you watch.
Not necessarily for your record but just ingrain what you're learning into your head that is much more efficient. Studies show that taking notes by hand really does enhance your brain's ability to retain the concepts of you're learning. Then you can move on to calculus 3 blue, 1 brown has an incredible playlist called the essence of calculus.
When it comes to probability and statistics, EDX
has a really good course called the science of uncertainty taught by MIT. Youll also finds an awesome course on algorithmic design and analysis on EDX taught by UPenn. Dedicate a week to each one of these courses.
Complete one project at the end of every week preferably, harder ones at the end of the course. Each of these subjects has a one-page cheat sheet, you can find on the web anyway don't feel bad if you complete a course faster than what the authors suggest.
Learning is just about downloading data into your brain you can use tools to accelerate that process. Now we're ready for month two machine learning, I've got three playlists for you to watch in order learn Python. For data science, the math of intelligence and intro to tensorFlow then check out Udacity
free intro to machine learning course.
Here is a great compilation of ideas in the form of a GitHub link to help you. Through this, you will easily develop a sense, of when to use a specific machine learning model and how it would work for an application. You are suggested to pick and choose two projects a week and building each from scratch. This will give you hands-on experience with the pillars of machine learning including optimization data preprocessing. Types of learning data set splitting and model evaluation this is all about turning sponge mode into the code and a lot of the art of data science lies in the dozens of micro decisions. If you can code this from scratch, literally everything else becomes easier.
Alright on to the third month, deep learning the specific algorithm called a neural network when given vast quantities of data and vast amounts of computing power outperform almost everything. Check out the intro to deep learning course on YouTube.
Then check out deep learning course 36 hours of high-quality lessons for free for the last two weeks just implement deep learning app, after deep learning app over this time period you should have at least 5 to 10 projects related to machine learning and deep learning.
On your GitHub, you'll be surprised how fast you can go from learning about simple perceptrons to state-of-the-art models. The difference being usually just a few minor tweaks in the architecture the deep learning space as of now is rich with technical literature and some of the best stuff is free.
At the end of these three months, you should feel confident enough to take your knowledge and either work in the field like a machine learning engineer. B start a consulting firm or C start your own product based AI company. Just figure things out as you go, stop telling yourself that you can't learn it or you're not smart enough. Remember, learning is a life long process no matter how old you are or where you're from you can get to the point where you can make an impact in this field.