Learning AI in 2022
The list of the 5 learning resources I use in my day-to-day to study AI and stay up to date with the latest news in the field.
Starting to learn Artificial Intelligence can be overwhelming and frustrating due to the immense variety of topics and concepts that are floating on the internet. Moreover, most people think of AI as a difficult, mysterious, impossible-to-understand technology, when the truth is that apart from a basic understanding of math — literally sum, product, and a bit of derivatives — you can easily start learning it from scratch.
In this article, I will share my 5 learning resources to start learning AI in 2022 and a few tips at the end that will help you in your journey (Spoiler alert: There are no courses!).
Note: These resources are not just for the people that are beginning to learn but also for those who want to keep on studying new topics and go deeper into them
Table of contents
Resources list
Tips
Resources list
1. YouTube
The Google platform brings together millions of people who share their knowledge and can teach you anything you can imagine, and of course, there is no exception with Artificial Intelligence. If you want to get an intuition about complex topics, learn the main AI programming python libraries, or be aware of the latest updates in the world, I recommend you to visit the following channels:
Mainly dedicated to news and explanations of complicated topics about AI, but you can also find some coding tutorials.
One of the best things about this channel is the incredible animations that the creator uses to make the explanations easy-understandable.
Although this is a Spanish channel, you can turn on the YouTube automatically translated subtitles that, actually, are generated using AI.
This channel gathers a vast amount of Machine Learning content, but I especially recommend you the list by Andrew NG, who is a famous professor and researcher in the Artificial Intelligence field that has created a full series of videos explaining the theoretical concepts behind AI.
This series of videos belong to a specialization AI course in Coursera that you can watch for free if you select the option “Audit course”.
Both channels are focused on coding tutorials and paper reviews that are very good to achieve a better understanding.
2. Podcasts
Becoming more popular recently, one of the biggest advantages of audio resources is that you can listen to them on your way to work while taking a walk or doing exercise.
The Machine Learning Guide Podcast is very good and I highly recommend it. Its speaker, Tyler Renelle, explains the concepts of ML and AI in an appropriate and clear way, using examples to illustrate his explanations throughout.
One key point of this podcast is that talks about the simplest shallow algorithms for Machine Learning, which are other important powerful tools.
3. Books
The following two books are focused on the simple ML algorithms and mathematical notions that are required to understand better the more complex models.
Clarification AI and ML: Artificial Intelligence and Machine Learning are not different fields of study, in fact, ML is classified within AI. This means that AI is like a huge world and ML is one of its continents. Moreover, Deep Learning — for which AI is known and where the Neural Networks and complex models belong — would be like a country within the ML continent.
The majority of the biggest AI models nowadays are based on ideas and simpler ML models that were developed decades ago. However, what is more important is that many of the actual problems can be solved using those techniques, requiring lower costs and less time, and thus knowing and understanding them is vital to developing the optimal solution.
4. Articles and papers
While articles are easy-to-read informative content created for a wide public, papers are first-hand information for the scientific community. Knowing how to combine them and extract the most valuable pieces of content is a time saver in your research.
Although papers tend to use difficult notation and they are hard to understand, eventually, with time and practice, you will be able to read them. Try to read papers in which you are familiar with the topic and usually it is better if you have read some informative articles or watched some YouTube video before.
One of the main paper repositories on the internet from where you can download them completely free is the website arxiv.com
On the other hand, for articles, I highly recommend reading the Medium publication Towards Data Science, where many researchers and science popularizers publish their content.
5. Practice! Practice! Practice!
And last but not least, there is the key point that will allow you to learn more than any of the previous resources: The Practice.
Theoretical knowledge is very important, but without the application it is useless. Therefore, in addition to all the previous content, I recommend you start developing your own projects and see firsthand how to plan and build one.
To do this, I suggest you use Kaggle. Kaggle is a platform where you can find thousands of databases and projects related to AI. In addition, it has some basic projects to start and learn from scratch.
Another interesting platform for finding databases is the UCI Machine Learning Repository
Tips
- Try to follow a chronological order
AI was born around the 50' and since then has experimented a vertiginous growth. To understand the latest advances, it is necessary to look at the past and understand the simple models, from the Neuron to the Multi-Layer Perceptron and the Backpropagation algorithm, each discovery has been based on the previous ones (as in all fields of science). To increase the speed of your learning it is better to follow an order that starts from the simplest bases and reaches the most complex concepts of today. - Focus on one topic and, after you fully understand it, move to another
As you may have noticed, AI is a huge field, and trying to cover it in just a couple of bites is a big mistake. After laying the groundwork, you should choose a field of study and focus on a specific topic, e.g. Image generation with GAN, Image classifiers, Sentimental analysis, Text comprehension… There are many open study fronts and each of them needs a time of specialization and research. If you just read something about GAN and then switch to NLP, then prediction models, afterward move to image classifiers… you will end up with a very shallow understanding of each field but no valuable information. - Break things into simpler concepts and start from the simplest one
This is a very common study practice that makes it easier for you to digest complex concepts. For instance, if you are trying to understand an Image Classifier first you will need to know how images are represented within a computer, then how the Convolutional Layers work and the other ones used to manipulate images, and finally how the Cross-Entropy loss function works and why it is used. - Do not just focus on the latest supermodels, remember that many problems don’t need such a complex solution
As Tyler Renelle says in one of his podcasts: “Deep Learning is like a bazooka, you can go rabbit hunting with it and get it, but there are other ways cheaper and easier to do it”. If you want to solve a problem for a company or a client, you must find the most optimal solution as an AI expert, and that includes also low costs, time-saving and easy implementation. In addition, the companies that run the latest projects don’t publish all the information about them, so other businesses cannot copy them.
