7 Mistakes Beginner ML Students Make Every Year

Don’t study LLMs! You’re making a mistake!

I have received a lot of DMs from people asking me for advice on how to learn machine learning. So in this post, I thought I would talk through 7 of the top mistakes I see beginner machine learning students making every year. My goal for this post is if even one student reading manages to avoid any one of these 7 mistakes, then this post will have been absolutely worth making! So stick around, because the last mistake might be something you are doing already and think is actually good!

Jumping Straight to Neural Networks

Now, I know, I know, we all want to work straight away on the latest and greatest technologies, which all work with Neural Networks. In other words, we all want to get directly into Deep Learning. But this is really a mistake if you want to take studying Machine Learning seriously and get a job in AI. Machine Learning fundamentals are not 100% the same as Deep Learning fundamentals and are perhaps even more important.

You see, AI is just the most general term for decision-making algorithms and does not necessarily mean ML. In fact, as far as I know, AI itself is a bit difficult to define. But a subset of AI is then actually Machine Learning, and a subset of ML is Deep Learning, i.e., Machine Learning but with Neural Networks. A further subset of Deep Learning is then the hyped Generative AI field and an even smaller subset of Generative AI are LLMs like ChatGPT. So, if you are directly jumping to Neural Networks and LLMs, you will, for one, not be able to understand the origin and details of the technology, and more importantly, you will miss out on the majority of different classical ML techniques.

Also, not only are fundamentals of ML important in that regard, but you simply will need to know them when taking ML job or internship interviews. Very often, the first round of interviews for ML positions is rapid-fire ML questions. The fundamentals.

And if you are still (understandably) overwhelmed with learning Machine Learning, one of the best ways to avoid mistakes is to find someone with more experience than you to talk to. Someone who can mentor you and give you direct advice. And if you want to chat with me, feel free to just pick a slot in my calendar. But back to the general advice I give in this video of mistakes to avoid!

So, take your time, and don’t jump straight to Neural Networks.

Ignoring Algorithms and Data Structures

The same can be said for the next mistake beginner ML students make every year. Because of this mistake, you will very likely fail ML job interviews.

You might think, since you want to be an ML engineer or even researcher, why should you care too much about annoying algorithms and data structures?

Damn those Binary Decision Trees, Heaps, and sorting algorithms…

Why should I care about how to sort a list? I am developing artificial intelligence! Not implementing some user database backend or a new button to add an element to the shopping cart…

But I’m sorry to tell you that all the ML and DL knowledge has to come on top of algorithms and data structures. This is especially important to know if you are coming from a pure maths, data analysis, or other background.

Now, I will be honest: you will probably rarely need that knowledge in your real job, as do normal software engineers, but these coding problems will very often still be part of your ML job interviews. Perhaps you will not have 5 coding interviews in a row with LeetCode hard questions, but that just depends on the position and company you are applying for. These coding interviews are used to assess your problem-solving skills, just as it’s done for normal software engineers.

So, even though you want to work in ML, don’t ignore algorithms and data structures.

Ignoring the Fundamental Math

Now, you are right, in today’s time, you practically don’t need math for ML. But if you think that, you are also wrong.

Yes, all popular libraries take care of pretty much everything that involves math, like gradient descent. So, you don’t have to know how to compute the derivative of a function. You see tutorials on how to train a Neural Network in only a few lines of code and think that that is Machine Learning. That is what you will be doing at your job. But, if it were so easy, much more people would be doing it.

If you really want to have a deeper understanding of what is happening under the hood and if you have to debug a model trying to figure out why it is not learning, you need to understand the fundamental math. And this is even more true if you want to actually develop a new method, you then really have to know the math. E.g. why do people use the dot product to measure the similarity of two vectors? If you really took the time to understand why that is, you can understand when to apply it in other use cases.

I know math is scary and with all this I don’t want to discourage you from learning ML. It’s rather the opposite, I would love to inspire you to really learn Machine Learning and avoid mistakes that beginners very often make.

I really believe that math is not all too difficult if you tackle it step-by-step and understand the different, more advanced math concepts once you stumble across them. Curiosity should guide you to want to understand it. After giving it a shot a few times and not giving up, you will then see it’s not too scary, but again, rather the opposite. It’s very logical and simply fascinating.

So don’t ignore the fundamental math.

Having a Ridgid Mindset

Don’t study LLMs. Now, let me explain what I mean.

I am sure everyone is seeing the hype around LLMs. ChatGPT is the root of this hype. I’m not saying ChatGPT was bad; absolutely not. It brought a lot of attention to the world of AI and excelled in research and development. Perhaps even too much, but that is a whole other complex story. But all of a sudden, every beginner ML student wants to eventually work on LLMs, ignoring other fields that might interest them even more or might perhaps give them a better chance of getting a job because of lower competition.

It’s like wanting to open another Italian restaurant around the corner, where there already are 4 others. Perhaps you have better chances of success if you start an Indian restaurant and perhaps you enjoy it even more!

There are so many different fields of study in Machine Learning. RL, evolutionary learning, graph neural networks for chemistry, biology (think of alpha fold), or street networks. There is Audio processing for generating voice, understanding voice, and translating in real-time, or there is simply applying machine learning to specific domains, for example, medicine and so much more.

In the end, if you are inspired by a topic and, most importantly, know how to ask and answer the right questions, you will become very good or even an expert in that field. Being stubborn and only following the current hype can work, but in most cases, it is not the best option. Open your eyes to the vast amount of different options you have with ML and find something unique to you and your interests.

Don’t have a rigid mindset.

Overthinking

“Where do I start?” Or “What’s the best way to learn ML?!”

These are the easiest but also most difficult questions to answer.

In most cases, they imply that there is one way of learning ML. And if you ask different people, they will most likely say different things. So beginner ML students think those recommendations are mutually exclusive and then get confused. The right answer is, there is no single right or wrong way to learn ML or a place to start.

The best time to plant a tree was 20 years ago. The second best time is now.

Learning ML takes time and will be challenging. If you focus too much on math, you will later struggle with coding. If you focus on coding too much, you will later struggle with the math.

“Should I learn PyTorch or TensorFlow?” is another of those questions. Although I generally recommend PyTorch, it really is not that important. In my opinion, the single best answer to these questions is to just get started and always let curiosity guide you. If you encounter something new and you don’t understand it, look it up. You can find everything on the internet. Yes, I would, of course, recommend starting with the fundamentals of math and Python, but in what order is honestly different for every person. Just don’t get hung up on them. Again, if you are curious about how and why things work, you will find joy in researching the new math you encounter further down the line. The single most important mindset you need is curiosity. So, just get started.

Don’t overthink it.

Playing Single-Player Mode

This is one mistake I also made for a very long time. I always studied and coded on my own, in the dark, in my room. But there was a time when I did work on a project with a very good friend of mine, and those long nights of debugging and studying were really the best times for learning anything.

If you ever feel like studying ML is draining, ask yourself, “How could this be more fun”. One answer could be to simply listen to music you like while studying, but one even more significant way of making learning ML more enjoyable is by doing it with a motivated friend. The motivated part is important; you have to be on the same page when it comes to putting in the time and effort. Then, when studying together, you can ask the other person for help if you don’t understand something, you can quiz each other, work on projects, and suffer together.

Just having someone silently studying beside you is such a boost in productivity and fun. I really mean it. Having someone as an accountability buddy not only helps you stay accountable but also ideally increases throughput. By that I mean you might not struggle with a math concept for so long as when you were alone or you simply have to double the coding throughput when working on an ML project together if not more because you are boosting your productivity by having more fun at what you are doing.

There are so many people interested in learning ML like never before. Look for friends to join the ride.

Don’t play single-player mode.

Doing too Many Projects

Now this final mistake is something I see very often. Beginner or even intermediate ML students have like 10 ML projects and ask themselves why they are not treated as well as someone with perhaps only one ML project.

The thing is, in the beginning, playing around with simple ML projects is a great way to get started with ML and have fun. But having 10 different Jupyter Notebooks where you import an existing dataset, pass it into an imported ML model, and train it, does not really count as an ML project.

Again, I think those simpler projects are a great way of having fun at the beginning of your journey and feeling like a real ML engineer or researcher but don’t make the mistake of thinking that just having more and more of such GitHub repositories will make you stand out more.

At some point, quality is definitely greater than quantity. Once you feel decently comfortable, try to really challenge yourself with a single but difficult project.

This will not only be more impressive on your resume but also teach you way more than having many little projects. One such project, in fact, my favorite project, is reimplementing a research paper and recreating its results.

So don’t do too many projects. Do the right ones.

Conclusion

So, if you don’t want to make these very common mistakes and would rather know how I would recommend properly learning ML, then you might want to read this post next, where I share how I would learn ML in 2024 if I could start over.

Thanks for reading! Ba-bye 👋

P.S.: If you like this content and paper reviews, you can also have a look at my YouTube channel, where I post similar content but with more neat visuals!