7 Mistakes Beginner ML Students Make Every Year
Become better at machine learning in 5 min/ week 👉🏻 newsletter.borismeinardus.com/
In this video, I share 7 common mistakes beginner ML students make every year! I myself made some of these and have personally seen many others doing them as well. I hope this video can help you catch yourself making any of these mistakes so that you can avoid them!
Enjoy 💛
👉🏻 Book a one-on-one call
calendly.com/boris-meinardus/...
⬇️ Follow me on my other socials and feel free to Email questions! ⬇️
⚫⚪ Medium: / boris.meinardus
🐦 Twitter: / borismeinardus
================== Timestamps ================
00:00 - Intro
00:33 - Jumping straight to Neural Networks
02:32 - Ignoring Algorithms and Data structures
03:57 - Ignoring the Fundamental Math
05:41 - Having a Ridgid Mindset
07:24 - Overthinking
09:01 - Playing Single-Player Mode
10:33 - Doing too many Projects
=============================================
#ai #machinelearning #mistakes
🚀 There is so much more to explore in ML. Feel free to grab my FREE cheat sheet of different ML domains and open challenges: boris-meinardus.ck.page/2f5e05bb1f
Math isn't hard. You just have to find a teacher who explains it well. The more logical something is, the less likely it is for the experts to have above average levels of communication. You're probably not dumb, you just haven't found someone who explains it in the way your brain works. Math is a universal truth. Teaching is hard since people see the world and absorb information differently.
very well said! Keeping that in mind will help you stay on track and not giving up ☺️
I would argue that the very disjointed common standards for mathematical notation can be hard. A good example is newton vs leibnez notation for derivatives, or how difficult it is to type on a computer (can't type delta on my keyboard!). The concepts are ultimately easy, but the notation is messy. I kinda think we should update math notation to something that is typed, similar to programming functions. Sum(2, 3) = 5, for example. Less symbols, more consistency. If we taught it in this format it'd be a lot easier to learn imho. It would also simplify order of operations and make it translate to the digital world. It would also simplify learning trig and calculus due to prior familiarity with the format trig identities arrive in. Imagine the quadratic equation: 0 = sum(prod(A, pow(X, 2)), prod(B, X), C)
Mathematics is complex. Absolutely no one with deep knowledge in mathematics (not at the engineering level) would say that they are not. If you think that, you don't know enough mathematics. You are probably referring to entry-level mathematics, which is essentially the mathematics taught in engineering. In that case, you are correct. Otherwise, no. I would like to see any average person even try to understand the formulation of complex mathematical problems. In any case, it is one thing to understand mathematics and quite another thing to discover and/or solve mathematical problems.
@@raul36 I absolutely agree with you. "Real/ pure math" is insane. To be fair, the math used for engineering but also most math used for AI are quite simple compared to "real math" where you have one proof that spans 20 pages. But as far as my experience goes, the math for AI is not too complex. Just my opinion, of course and if you are not even remotely into numbers and letters, you will have a hard time and it probably isn't for you. But yeah, thanks for sharing! Perhaps I should have made it clearer! :)
@@borismeinardus Don't worry, it was a great video. If you allow me to clarify something else, as I said, understanding mathematics is not really the complicated part. That is to say, it is relatively easy to understand the basic fundamentals of what an integral or a derivative is, but I would dare to say that a person who is not a genius would never have thought of formalizing those ideas. It is very easy to say that something is simple when it is already solved. Otherwise, I would bet all my money that 95% of the people in the world would never have been able to even formalize ideas as relatively simple as the derivative.
Hi , I made one of those mistakes i.e. directly start deep learning. Now I want to dive into the classic Machine learning algorithms. Thank you for this video
Thanks for sharing! Although a typical mistake, it wasn't time wasted! You started to get a feel for machine (deep) learning and will possibly have a better start when learning classical machine learning
I made the same mistake
Thank you so much Boris. You give some of the best ML advice I ever seen on YT. As a quick joke, I am low-key happy about the LLM hype, because the other fields of ML might become slightly more accessible. 😅
I appreciate that! 💛💛 Haha yeah, you do have a point there 😆
Amazing points to remember, Thank you so much, Boris
Really glad you liked it!
Very insightful. Thanks.
I’m really glad you liked it 🤗
Your content is really underrated... hope you keep on making amazing videos like these ! 😊
Thank you so much! This really means a lot to me :) I'll do my best to not disappoint you and every one else!
This is so great for me! Thank you ML Guru bro!
My pleasure! I really appreciate the support 🤩
I guess recreating research paper is most essential and crutial. You are absolutely right. Been missing this technique.
Thank you for making this video, super underrated channel.
Thank you for the support! ☺️
Man, I have been listening to many ML guys. but i dont think they tell the things the right way, I was figuring out the next steps in ML journey, and you made it clear for me,. Sooo happy. blessings
Thank you! I'm really glad I could help a bit along the way!
I was just about to scrap the stanford ML specialization to jump to LLMs. Thanks so much I needed to hear this!
Exploring things where you want to get to is fun! But if you want to take learning ML seriously you need to learn those fundamentals :) Keep it up and have fun ☺️
This is 100% accurate. Because of LLMs being my research topic I went through many papers recently, therefore I can bring another reason why you should not restrict your study to LLMs: there is a need (hence an excitement) for other approches to enhance LLMs abilities. Few-shot learning, meta-learning, multimodal setup, knowledge graphs to name a few. So the more you explore the field, the better prepared you will be for the future of LLMs :)
Exactly! Well said! 🚀
you mean in-context learning ryt?
@@shilashm5691 ICL is also an interesting perspective for LLMs post-training, although it is different from few-shot learning mainly because it doesn't involve parameter updates. This is well explained in GPT-3 paper for instance.
@@barbaragendron2836 But, as far as i know ICL also can be used for few-shot learning, we create a prompt with few instance of examples for some task nd input it to the LLM's. Take a look at Meta learning and multi-task learning cpurse from stanford
@@barbaragendron2836 We can also both fine-tuning with few examples of task and in-context learning with few examples in prompts as "Few Shot Learning". Correct me if im wrong.
Hey Boris! I appreciate your channel and I would love to see some actual building of ML projects! I feel that KZhead is slowly turning into a "give generic advice" platform, mostly because those kind of videos are just so easy to make. I don't want to criticize what videos you make, all I'm saying is that I crave videos with practical substance =D
Hey! Yeah, I see what you mean. The issue I have with such specific projects/ tutorials is that they are very specific. There are other great channels that make ML coding tutorials (such as Nicholas Renotte or Patrick Loeber) so I feel like it would be nothing all too new. That said, my next video will be about my favourite ML project to work on, but it's again rather a breakdown of how to approach that project. I still think it is very useful and I would have loved to have such a video when I got started. That's why I'm making the video :) So yeah, thank you so much for taking the time to give a recommendation! I will still always keep that in mind! 💛
Simply superb
If I may say, I'd add another mistake. As you said people tend to create just a notebook and put all your code inside. Generally, this code has the data as input and a model or metrics as outputs. I suggest makingake this code as a pipeline. Think about how you could make this reproducible as a manufacturing mat. Modularize your code and break it into small pieces independently.
That's indeed a good practise to get your code from exploration to an actual pipeline for longer training! Thank you for sharing :)
MLOps ? I am a beginner don't know much, so pardon me if I am wrong 😅
@@Param3021 It's related of course but I like to mention this because in the real world, the projects don't finish when the model is trained. There are several steps to put this in production and for me, a model that doesn't go to production is the same that you don't have a model.
Indeed, this is a good suggestion; however, in R&D testing phases, a more practical and time saving approach might involve using a monolithic strategy.
yes yes like we started doing 10 years ago
Great piece of advice
Thanks 😊
Mal wieder ein sehr gelungenes Video 💪🏼
Danke 🙏🏼☺️
I would argue that a bit of iteration between classical ML and DL can keep things exciting. I did my PhD on a sequence generation problem that was pretty much entirely dealt with using HMMs, which became the first ML algo I studied. The second was GANs, which at the time was considered the most challenging DL method to work with (possibly still is!), and I ultimately got my PhD for showing they performed better in the problem I was considering than the existing approaches. Since then, I've gone back and forth between DL and classical ML algorithms and have done postdocs on projects involving various applications of ML and now teach the subject at Master's level - from K-NN to the most obscure SSL computer vision methods. I think I would have lost interest with just learning classical ML before I got to DL - learning about them in parallel helped keep it exciting AND helped me link things together. Completely agree with all your other points though! Fundamental maths and computer science concepts are everything and knowledge about advancements in LLMs is nothing - if you know higher order Markov chains then LLMs are better equipped to work on them than anyone who doesn't.
I've been aware of the fundamentals of machine learning for around a decade now, but I just recently got into it and took the plunge. I have no PHD, or higher education that would suggest knowledge of machine learning, but I feel I've learned a lot pouring over research papers and applying them in code over the last six months. Strictly speaking, odds are that I shouldn't be able to get into an engineering position this way, but stranger things have happened. I'm not really sure how to explain it, but in the last two months or so that I've been properly implementing things, particularly research papers without existing code implementations, I've really felt that it's all started to make sense. Who knows, maybe I'll luck out.
Thank you Boris. What tool do you use to make your animated diagrams?
Amazing video!!! I'm a recent Electrical Engineer BS grad but I really want to get into this! I will be starting my MS in CS this Fall but want to hyper focus on the pre-requisite math, data structures, and algorithms that will give me a solid foundation. I have some experience from my undergrad but not enough to feel super confident. Does anyone have any recommendations on online resources to get fully caught up on these core foundational classes. Particularly a class that is teaching these topics with the intention of being applied to AI/ML?
Great video and I feel you are underrated. Thanks for existing on KZhead. Could you tell the average salary and the salary range in the ML field in Germany?
Thank you!!! I will do my very best to not disappoint! So, regarding the salary ranges, it's difficult to say of course. It depends on the city, level, and company you are applying to. Most non Big Tech companies might go up to 90k/ year, but for big tech that is rather the lower end for entry/ early level salaries. I recommend to visit levels.fyi, there you can very nicely filter through the jobs yourself and have a great look! Happy Learning!
I'm still watching just want to say you have one beautiful and kind soul I bove you so much you're great, Boris
love*
Thank you very much 😊 I'm trying my best to provide value and really appreciate kind comments like yours!
Can you provide us with a video explaining the basic mathematics in this field?🙃
Actualy when someone use the Term AI you knows he is highly incompetent or is working in the marketing department. 😅 Good video, like in all disciplines start from the foundation otherwise your growing potential and flexibility is limited very fast.
yes haha I have to struggle with when to use AI and when not. 😅 But most people are looking for „AI stuff“ and since I want to reach those people, I somehow have to offer them the term Thank you for the kinds words ☺️
Good to remember that different programming languages have different cultures. I have been learning Julia and that has helped me a lot because I have been able to get help from Discord channels and forums. Also that culture steers you to certain path, i.e, Julia community often solves problems with a certain way.
I honestly don‘t have experience with Julia, but I can very much imagine that the communities differ! So many C++ fans for example hate on python haha
@@borismeinardus You can see the difference more clearly when you ask ChatGPT do easy task in C, C++, Python, Java or R. If I do statistics, I often ask the code in R and then translate it to other languages because if you ask it in Python, it might write the code like a web developer
Your videos help me in decision making. I worked on few supervised learning projects as an Engineer and realised that i can't be top 1% engineer. I'm looking for AI PM roles where i can seemlessly discuss with ML Engineers without having to worry about bug fixes or days of uncertainty. May I know what challenges do you face when working with product managers ?
exciting! Hmm, I personally have not yet worked with PMs, sorry :/
Thanks, Boris. I found your video very insightful. However, I have another question. Is it important to secure an internship or industry-related work experience in machine learning? I'm a final year student and most of my experience so far has revolved around research activities at the university. This includes participating in an undergraduate research program, working as a research assistant, and doing a few of personal ML projects. I would really appreciate your insights on this matter.
Thank you! So, it is difficult to say because it really depends on you specific case and university experience and projects. But in general, I would always say it is very useful to have some industry/ research lab experience. You can always write me an email and I'll do my best to get back to you, but if you want to ask more complex questions and get some better mentoring feel free to simply book a 1-on-1 call on my calendar :) calendly.com/boris-meinardus/consulting You can also subscribe to my free weekly newsletter where I share even more advice on getting into ML newsletter.borismeinardus.com I hope this helps a bit 😊💛
thanks for advice , i am programmer starting to learn ml instead directly llm ... thanks for advice 😊
I‘m glad it was helpful ☺️
Hi Boris, I’m focusing on computer vision can you suggest any courses or learning materials that focuses on computer vision. I want to create a model that detects if something got rotated, flipped, etc.
Did you find material to study from? Oh and, might I ask what is your main focus which makes you study this? :D
One last question, do you recommend to do ML crash course by Google before Andrew Ng's course. And a humble request , please make begginer friendly projects on ML as hands on project base learning really helps to understand the topic.Btw thanks for helping the community with your valuable knowledge 😊
Disclaimer: I have not completed the ML crash course by Google, but have skimmed through it a while back. I think they complement each other but you don't necessarily need both, i.e. I think you don't need the ML crash course before Andrew Ng's course. Regarding the hands-on projects, I'll have to see whether I can provide more value with such videos as there are already a lot of those on YT. That said, a while ago I did actually make a project with 2 parts :) Here is part 1: kzhead.info/sun/ndyLdZGeZ2uhdXk/bejne.html Thank you for the support!! I really appreciate it 💛 Happy Learning!
Hi Boris! Thank you for your awesome contents. I really want to have a chat with you regarding directions about my ML career. But in the call booking link, there is only the option of paying with paypal. There is no option for payment with cards. Unfortunately, paypal isn’t allowed in my country. Can you provide any alternatives?
Hey! Thank you for kinds words 😊 I see. I went ahead and created an alternative link: calendly.com/boris-meinardus/consulting-2 Looking forward to our chat! 😊
Should I study Design and Analysis of Algorithms then? I like the book “The Art of Computer Programming” by Donald Knuth, I’m hoping to start doing some problems with it when I get into uni
If you enjoy coding and algorithms, you will have a good time studying algorithms and data structures, which are part of the toolkit a ML engineer needs! At least for the ML engineering interviews :)
For Design and analysis of algorithms, I’d recommend “Introduction to Algorithms “ by Cormen, Leiserson, Rivest and Stein.. Popularly known as CLRS..
Hi Boris. Where do you think I should see research papers from? Most of them are so hard to digest.
I think when it comes to reading papers there are a few tips one can give, but you are literally reading about the current (or past) state of the art research. It will be difficult in the beginning. But if you just keep going, google ever thing you don‘t understand. Ask ChatGPT or Microsoft Copilot, Perplexity, or any of those LLMs, you will learn so much and get used to it much quicker than you think :) Just keep going and, most importantly, enjoy it! every google search you do is a new thing you learn! That‘s at least how I see it ☺️💛
@@borismeinardus thanks!
We need separate video about learning math the right way.
Haha that's really a tough one
@@borismeinardus I mean math for machine learning. How did you learn it? Should we take separate math courses before jumping into machine learning? Should we learn math while by googling while learning machine learning? This is something important not clear for most new comers.
@@borismeinardus I mean math for machine learning. How did you learn it? Should we take separate math courses before jumping into machine learning? Should we learn math while by googling while learning machine learning? This is something important not clear for most new comers.
My understanding is that there are no entry level MLE jobs. As a baseline you are expected to have senior level experience in Software Design and System Design on top of all the ML theory, methodolgy and tools. The likley path is to either be a SDE or Data Engineer first then transion into ML after you have enough experince working with data at scale and desiging systems at scale.
It really depends on the company, the role, and your experience straight out of college. I e.g. know a guy who got an ML engineering position at Apple straight out of college (masters).
Hey I have completed the math course for machine learning provided by DeeplearningAi, do I need to deep dive more into, or it's enough to start and to build the foundation
I don't quite know what exactly the specific course covers, but if you have had topics like probability theory, linear algebra, multivariate calculus (not even all too in depth), you should be good to go to continue on building more practical experience in ML! Every time you later come across some math you don't understand or have not seen before (which is pretty much 100% guaranteed) you will have to look it up and thus continuously learn more and more :)
Hey what do you think about the application of machine learning in the cybersecurity domain?
Detecting fraud and nowadays detecting AI generated content is a very important application of AI!
@@borismeinardus I'm also interested in things like malware detection with ml
Is google's machine learning crash course enough to get started? Im planning on doing Andrew Ng's course next
Yes, it is!
Also I am a good maths student and have currently completed the ml specialization course by Andrew ng where would you recommend I go from here?
I think you should get to reading papers. There you will get a feel for how comfortable you are with understanding the theory. If you feel comfortable with reading papers, I would say you get your hands dirty and work on projects. My favourite one is reimplementing a paper and recreating it's results :) (I will have a video in 2 weeks covering exactly that! Might be useful to you)
Hi Boris, I haven't come across such videos of implementing research papers and reproduce results. I think this would be very helpful in understanding what critical thinking skills are needed to read papers and contribute. Good idea, bro!
@@shaasvatbharat video on that coming soon ;)
excellent video. However if you are a developer and want to apply existing models to whatever you are doing, the best way to learn is doing. If you are a researcher then indeed....
Thanks for sharing! In any case, be it as a developer or researcher, the best case is just putting in the time and learning by doing. As Andrej Karpathy also said - If you put in 10,000 hours, you will be very good at whatever you are doing.
I start my Computer Science Degree with a major in ML at UQ this year- any tips?
- Don't get frustrated when you don't understand something. Take a break then revisit it. Look for explanations on the internet. Ask friends. - Be curious. When you see something new, try to get into the mindset of "wow, that might be interesting", "learning this will make me an even better ML engineer/ researcher" - Get to reading papers and working on projects sooner than you think. By reading papers (and following the tips above) you will more than by just going through curated content. By coding you will learn how to apply the theory and really get an intuition for it, how it works and why. - Enjoy the learning itself! It will never stop :) Happy Learning!
@@borismeinardus thank you so much =) I'll see you in 3 years when I'm looking for a job and need advice to not fail my next ML interview
@@Uncreeperble I'll be waiting! 🚀
Can you please make a vid on maths for ml, all maths in detail with if possible free sources on internet (Books and vids)
Hey! I give some of my favourite in my video on how I would learn ML in 2024 if I could start over :) But a new one I came across is the following book called "Mathematical Introduction to Deep Learning: Methods, Implementations, and Theory" arxiv.org/abs/2310.20360
@@borismeinardus thank you for your reply also to mention I have started ml Done python now am doing stat 110 from Harvard on KZhead I know about linear algebra, multivariate calculus are there any other topics in math that I need to know??
@@Lostverseplays Amazing! You are on a really good track! Probability theory is another big part of ML so stat 110 is a good place to be! Have a look at the topics in the book I mentioned and see if there are any big topics that you might have missed. But as one of my main recommendations, I always say to not get too hung up on the complex stuff and details. Once you get to reading papers and working on more complex projects you will have a reason to revisit them and really understand them. If you put in the hours, just continue doing what you are doing, you will eventually understand much more than trying to understand something difficult, failing, and giving up. Happy Learning!
@@borismeinardus Thank you, the last four lines are really good !!
😊😊😊
How to find people for accountibility or to work with?😊
If you are at college, look at people that are hard working and enjoy ML or perhaps join some cool club. You can also join online communities, e.g. on discord.
I have a bachelor in Computer Science and one year of experience in software engineering. Do you think getting a masters in Computer Science will make it easier to land a ML job?
Short answer: yes. Long answer: Depends on what opportunities you have. If you are a SWE working closely with ML engineers or on ML projects at your current company, you probably don‘t need it. But if you don‘t have such an opportunity and can‘t work on personal ML projects, a masters would probably make sense :)
Thank you! @@borismeinardus
Strong in fundamental math, ok in coding still not getting any job 😢😢
Hey why don't you start a ML course , that would be really helpful 🙂 , I know it's time consuming , you can just guide us using others course
I have definitely thought of creating a course. But as you mention yourself, it is indeed very time consuming 😬 But if enough people show interest, I will definitely do my best to put something together! 💛
I am looking for good tutorial to learn basic ML to Advance ML with mathematics
Mistake--not watching KZhead videos by "'The data janitor"' Mistake--Either go top down or bottom up or sideways( and if your primary concern is money , don't come in ML...not that it is wrong... it's just that you're not going to make it) Mistake-- Believing (ok he has mentioned in the video,, as I am typing this lol..I guess mistake 5) Edit: This comment is for Indian viewers. Like most of the jobs are out of India, and in even in IITs, nits and iiits there is nothing serious going on ( research/ industry). Now you might land up a job simply answering how linear regression works or you might not get a job even after understanding everything in the paper "'attention is all you need"'. There's no framework and luck plays a major role.
Thank you for sharing!!
🔥
🚀🚀
Hi I really liked your video and recommended to my friends. Can you please add a video on how to start learning python
LLM’s are like the DOS operating system, it will be left in the dust.
Hi boris -- Give me some good resources for learning machine learning and ai in reply
Hey! I give some of my favourite in my video on how I would learn ML in 2024 if I could start over :) But a new one I came across is the following book called "Mathematical Introduction to Deep Learning: Methods, Implementations, and Theory" arxiv.org/abs/2310.20360
Please suggest some good resources for learning python
I really like www.youtube.com/@patloeber But another great full course to get started with is this one kzhead.info/sun/ktmKoq5oi52Qn5s/bejne.html by Programming with Mosh :)
is python strictly necessary or can you go along with a different language@@borismeinardus
TRUST ME YOU WORTH MADE
Not 100% sure what you mean but thank you! 💛
Full stack web developer vs data engineer more job opening more package future growth
Both are pretty different but if I really had to choose one, I'd go with data engineering :)
s sound is too loud
Totally agree with this video; I've had many students who hated ML because their professors taught many maths... Also, in case anyone would like to understand how simple is to write the code for any Machine Learning model from Scikit-Learn library: kzhead.info/sun/mal7qKqPrKaeZ5E/bejne.html
Hi, is there someone starting like me and what to learn together? Please leave a reply, people from all the countries are encouraged to join me and make a new friend and learn together... I'm from INDIA
Learn math and code it in Python
hallo, I saw you and your sister in the cinema watching The Boy and the Heron, i feel that we have a connection, we can be really good together. Please answer me, i love you...
Yeah, I indeed was there with my girlfriend :) The movie was amazing, but also very confusing haha You can of course politely come up to me and I would be very happy to say hi 😊 I am also flattered by your kind words but I am sure there are other great guys that are into AI and Anime 😊
😂😂😂😂😂😂@@borismeinardus
This video is not making any sense to me. I am proponent of the "10000 hours rule". There is no perfect path to learn AI. Try the path which interests you the most. Andrew Ng suggests that an AI student should try to read and understand and deply AI research paper in various ways. Simply, there exists any foolproof path. There is no one path.
I read your medium article this video came into my feeds on KZhead by "KZhead recommendation " nice article and video :) I recommend don't jump to hype ! what you think
Reimplementing the paper
Ahh, cool 🤩 Thank you and welcome aboard!