There were videos that you post about Computer Science vs Computer Engineering in TUBerlin, where can I find them. I wanted to study CS at first but my NC is not enough, now i am thinking about Computer Engineering. But I want to focus on Software instead of Hardware or electrical side of things. But maybe it is better to get a job with knowing hardware side of things too.
I mean when I need your ideas most, they are gone! I remember watching them 1 year ago but I couldn't find it, do you have these videos somewhere, or at least can you answer if my ideas are right or not, or maybe are you happy with computer engineering bachelor?
I gave up learning deep learning many times. Then in 2021 i had painful divorce. It changed everything instead of going into the depression i channeled my energy into learning deep learning. Now i am able to train small llms and text to speech models from scratch on multiple gpus. I am starting a company in the next month!
Wow! Thank you for sharing this impressive story from being at a very low point in life to such an exciting time of starting a new company!! I can't imagine how tough a divorce can be... I am certain you had a hard time, but it seems you are doing much better now. I genuinely wish you all the best! Much love ❤️
Please share your roadmap, every time I try to learn ML/deep learning there are soo many concepts that I need to learn. which in the end I fail as I give up on something.
I took Deep Learning and Machine Learning subjects at the university last semester and now in the current semester. The Deep Learning professor would explain the theoretical process in plain English with graphical representations before even showing any math formulas. By that time, I get quite comfortable with what I'm learning or about to learn since I know the process and the end goal. The only thing I look for in the formulas is certain details or how exactly what is happening in each step if I'm too curious. On the other hand, my Machine Learning professor, if I attend the class, all I hear is math. When I go home, all I have to study from is the slides, which are again all math with no explanation in plain English. I ended up dropping the subject. It's crazy how a teacher can make someone love/hate a subject or make them feel good/bad at the subject
schools/colleges on purpose teach half knowledge, that whay I understood. The only way to learn is use chatGPT like tools and learn by your own step by step without relying on these misguiding education institutes
I think the perfect approach would be a combination of both. I think you should always start with the intuition, but not skip the math. But I also do agree that classical ML has more (more difficult) math than DL
@abhishekdhaka4833yes, the math is important (possibly the most important thing about machine learning). Sometimes though, people need a review. I for instance needed a review on the Chain rule from Calculus to understand back propagation in Machine Learning. Having slides with graphical representations help too, for instance, when we look at activation functions like sigmoid, tanh, ReLU, etc. it is good to actually get a graphical representation on how it works. From my current Deep Learning course that I am doing, I was having a hard time understanding convolutions at first until I watched Dr. Andrew Ng's Deep Learning videos. To be honest, I still have a lot of difficulty understanding CNNs, RNNs, etc. but I definitely feel that eventually I will have rock solid understanding of it.
The day I stopped hammering formulas into my head, I understood how amazing math actually is! My advice to my younger self would be to try to understand the intentions behind all those formulas and not to force it. Occasionally, you discover why you need it when you need it.
00:01 Understanding math the right way is crucial for learning machine learning. 01:39 Translate human ideas into math concepts 03:18 Understanding math derivation rules simplifies ML learning 05:02 Recognize patterns and practice is key to mastering ML. 06:40 Realizing that debugging is coding helps with implementation 08:18 Understanding large code bases with a simple strategy 09:56 Setting breakpoints in the main function for debugging is crucial in understanding ML implementations 11:41 Mastering ML takes time and dedication. Crafted by Merlin AI.
Its like you have opened a new window for me into ML world. I was stuck at figuring out which topic falls into which category, but after listening to you it all came together. Now i can organize my thoughts properly while thinking and learning. I can relax while reading math formulas and be content by just understanding the intent behind it. I can relax while going through derivations knowing that it is pure mathematical realm and i can tackle it separately. And finally now i am slowly beginning to understand how to break down my tasks and work aka apply first principles thinking to ML problems. In short, you have helped me form a initial mental schema for ML. Thanks a lot !
Wow, thank you so much for these kind words!! I am really happy to be able to at least somewhat provide value with my videos :) I wish you all the best and happy learning 💛
Well, I was broke with no marketable skill set and was already hungry and couldn’t afford an electric bill. So, I decided sitting for 5 years in the darkness working my way toward the light and finally reaching it was better than being a loser forever. But hey, what do I know 😂
You can do this for a few months, probably even years. I did that when I studied for university exams. Then you realize your body starts to weaken and your brain is not working as it should be (e.g. attention problems) and it takes months to fix that. I think you can do this realistically for a few weeks, however, it is not maintainable in a long term. What is much better is to create a robust long-term daily regime you can follow for a longer period of time in what you can maintain your health and productivity. I would advice this to a younger myself.
I'm an 8th grade math teacher, and the first "secret" you talked about is my main point throughout the year. Math is just transltaing what we think to a more convenient way to write it. The problem is how most of us are being taught math since 1st grade, where they show us on side of an equation and tell us to solve it. so we are getting used to think the math must leads to a solution, instead of being a description. Math equation in schools should be called "fill in the blanks" and it should be taught in a same way a language is being tutored.
1. Think of mathematical formulations as being an idea from someone else, which is only being put in math terms as a tool 2. Math derivations are the continuous applications of rules 3. Debugging is a part of coding 4. Set a breakpoint in the code 5. Mastering ML is hard
This video is a breath of fresh air. I was frustrated because I couldn't understand how to use einops for 3 hours and I was only writing 1 line, I'm glad I stumbled upon this
Thanks for this (and also for the other videos) man! You can really see you're down to earth and you'va faced the same struggles everyone has in the journey towards becoming better ML scientists/practitioners!
I really appreciate that! We all have faced or will have to face a certain amount of struggles at some point haha. I'm trying to share what I have learned so that other people can hopefully avoid certain struggles or to at least somewhat help with overcoming them :)
I am 12th class student i have decided to start machine learning after my exams and in my Bachelor degree... UA-cam has recommend a great video to me...🎉
The thing that gets me about Deep Learning are RNNs, then going through LSTMs and GRUs. I try to memorize what happens throughout the entire RNN with activation layers and backwards propagation through time. It is just a lot of over-convoluted complex stuff. I just started learning Machine Learning and Deep Learning in August of 2023. I am just struggling so much with it, but I love it. I love watching lectures on Transformer sequences with encoder and decoder even though at first a lot of it sounds like Ancient Greek to me. I will keep trying
Yeah, those can be tricky to understand. But don't get hung up on them if you don't understand a specific detail. Continue with looking at the whole lecture on the topic and revisit the detail you didn't understand once you understood everything else.
This was actually helpful! Particularly your insights around getting thrown math formulas and being expected to understand it. It totally resonated that you need to understand the intuition before you can hope to understand the formula. Thank you so much!
1. Math intuition 2.Foundation of math toolkits 3 learn how to debug 4 overview of the code base and foundation of code 5. As a PhD student now, I have to say it takes a lot of time and loads of stuff or even new things to keep up with everyday. Try to enjoy it
As an software engineer by profession and being coding for more than 10 years, I 100% agree with your point. I have never met anyone who can just write a function and run it successfully everytime. Debugging is so integral to coding that I don't even treat it as a separate thing.
Hello! I really admire the way you teach something. Maybe we can have a sample tutorial of a ML / DL project with the standard structure for industry level
Thank you :) Really happy to hear that! Building a project at an industry level is quite challenging in itself and tricky to put into one video haha I have a video on how build out a project of reimplementing a paper. Perhaps that might be helpful :)
Of course. It's more important to understand the concepts behind math than to try to understand how to interpret equations. Mathematics is just a very specialized language, nothing more. It's a kind of shorthand for high-level concepts. This is why, as a student, I eventually figured out that it was best for me to just sit back during class and listen and truly understand the ideas presented instead of record accurate notes.
Hello Boris sir, thank you for your youtube videos. Moreover, I want to ask whether buying M4max apple with 128 gb laptop for machine learning, AI and Data science or buy M4 max with 36 gb and use cloud for higher data computation. It would be our pleasure to have on best laptop idea for these categories.
My man you and I think the same. Trust me I struggled with implementing a simple summation in my first code back in the days when I was doing my masters thesis. I was looking at a simple formula i.e. L1 = exp(2*pi*1575.42e6*time + doppler) and I literally had no idea to work implement this to generate a GPS signal in MATLAB and Python.
bro, colleges, shcools sucks at teaching maths, I feel sometimes its on purpose. Math is most important tool, but if you learn you would be much more powerful than people who want to have power in the society
Yeah, I enjoy math so much and am really sad when others hate it because they had a bad teacher. Luckily there are amazing videos on youtube that can assist the learning process ☺️
Great video! I am a college student and has already join some ML/DL courses. But I always think that the college resource cannot compare to some big companies like OpenAI, Google, Meta and so on. Lots of ML techniques have been mature by them. What could we do as just a college student? Thank you !
That is very true. The college courses can't really keep up because everything is moving so fast. It is very hard for them to update the whole curriculum of a course each semester. Nevertheless, there should exist seminars at college where you discuss recent research in smaller groups. That could be one option. Otherwise, it is up to yourself to keep up :/ College teaches the fundamentals of ML and DL, which you definitely need. But beyond that, you need to keep up on your own, which college will hopefully have prepared you for by providing a good fundamental understanding. That said, it's just moving so fast, you can't feasibly keep up with everything. Try to find a domain that interests you where you can dive deeper and also work on respective projects. I hope this somewhat helps :)
Thanks for your reply. Selecting a specific field seems to be a good choice. However, most of the domains have been prosperous. Many online courses cover diverse topics and I do not know how I should start.I want to know how you can find your way. Thank you again!
@@chiahungchiang9506 Oh my, not sure if I can help you there haha. I agree there are a lot of interesting fields. Explore as much as you can, see what you can see yourself actively working on, and perhaps consider the what the labs at your college offer. It doesn't really make sense to go hard on e.g. RL if none of you college labs has expertise in RL. I mean, you can, but you won't be able to get the best support.
I think it is somewhat a blessing if your college doesn't keep up with the latest hype. It is hard to know if the hype is useful after 5 years. I studied a lot likelihood functions (following George Casella's "Statistical Inference")though nowadays there isn't much hype around them but understanding MLE gave me a solid foundation
thanks for your advice, even though i'm still not knowing how can i improved my knowledge in the coding during my 5 year studying in computer science major but here are all the advice i need to find my own path in the world coding. you are right to get mastering in certain area you need to sacrifice your hard work and time to get through over it! starting now again i will try more harder
So so so grateful for this video. I have an AI Application I want to design and was worried about the math peice until I watched the Harvard Deep learning video and now you reaffirmed what I understood there.
@9:46. I agree that memorizing or coding opimtization is overkill. I wanted to learn everything by scratch and learnt that's a fool's errand. I learnt that the hard way. In deep learning, learning back propagation for FNNs, CNNs, RNNs are good ideas. But it is better to concentrate on the matrix algebra aspects of the forward prop than focus on the optimization part since most of the back prop is efficiently done in Tensorflow or PyTorch or for any ML algorithm in Scikit-learn. The mathematical derivation of any algorithm is more important than the cumbersome calculations. I took the Google's Tensorflow certification exam last year. What I learnt from the exam is the core idea of fitting the inputs and outputs in the correct matrix or numbers or any data augmentation or manipulation that better fits the model than raw data. To sum up, the practical aspects of the ML is all about how we frame the problem and how efficiently we solve the problem for real life use and debug our model when it goes crazy or throws up errors.
I'm enjoying your video thus far. Some constructive criticism: The summation code shown at 02:25 is wrong. It modifies the loop index variable, "i" and the "range(n)" function stops before 10, giving only 0-9. Your code has a value of "10" in "i" after the loop completes. However, if you change the variable "i" to "x" on lines 3 and 4 and increment the value passed to "range()", "i" ends up with the value "10". An equivalent one-liner: "print(sum(1 for i in range(n+1) if i != 4))". Although my advanced math is VERY rusty, I'm certain that answer is incorrect. You must add the value of the loop index variable with each iteration, NOT 1. That is: "print(sum(i for i in range(n+1) if i != 4))".
You speak some straight up truth in this video, these are super helpful insights into how to learn. One thing I would say is for programming you kind of go against the super good advice you give for learning maths. When trying to understand a big codebase, the debugger will not help -- the debugger is there as a magnifying glass into all the details. But just like your example in maths, code is just human ideas expressed a certain way, so understand the higher level ideas of functions or modules, and go deeper as your understanding improves. The debugger is great for when you are debugging the behaviour of your code, or if you are very familiar with the code someone else wrote.
Thank you! I definitely see your point! Perhaps I should have better described the scenario where the debugger is very useful to understand a code base or model. When I e.g. wanted to better understand a certain Transformer model, I understood the high level idea, but then wanted to understand the details. I then imported the huggingface implementation, set my breakpoint, and went into the code. I got to see the data preprocessing, data flow through the model, and so on. That‘s where I fell in love with using the debugger to understand code/ large code bases. But yes, you will most likely not want to have a certain high level idea of what you are looking at, but I still believe the debugger is one of the best ways of learning to understand code :) I hope this makes sense ☺️
Hey Boris, Can you help me with my little question..The question I had was, what topics of ML should I know to get my first job next year? And BTW, I'm way too fascinated by your knowledge, and believe me when I say you have helped my love for ML grow by 100x. Nothing's able to demotivate me anymore, and I'm maintaining my pace. I'm following your roadmap too.
Thinking of different degrees. I saw and experienced statistics, computer science, physics, and cognitive science and contemplated the curricula at my university (Helsinki). Many times I felt these fields to go deep in their own direction, if one wants to go primarily towards ML or AI. Then the internet is full of learn X quickly types of content. However, what I get from your experience is, that perhaps your/some ML curricula is streamlined in a way to get optimal amount of depth and breadth specifically for ML. For me the big question is, how deep into math one should go, because prioritization gives you other skills too...
Bro you are making me get there.Everyday I spend 3+ hours coding and learning something new of machine learning.So far I can develop Linear Regression models,Logistic Regression and Decision Trees. I am a final year student for Bsc in Mathematical Sciences and multi-tasking.Soon going for an honors degree to advance.I do work on projects too,I hope that in the next 6+ months I will be here to tell you I got a job in machine learning,now its 02:00 (South Africa) and here I am on Yt for ml ,passion!!!
Amazing video. This video comes at the right time because as a college student, who is about to make an ML course in the next semester, I was a bit anxious as well. Thank you for the guide!
Hello I would like to enter the realm of Machine learning, but as you know AI is becoming better and better everyday, what should I do to be a Machine Learning Engineer...what's would I study?
Hey boris I really like your advice on paper implementation but I am really confused on which paper to start implementing first. Could u please recommend papers which a beginner should focus on implementing
Watch the video "Let's build GPT: from scratch, in code, spelled out." Karpathy walks through implementing the transformer architecture from the paper Attention is All You Need. This is probably the best paper to try to implement from scratch given the ubiquity of transformers
really really good. I think you have explained this very well. I think I knew a lot of this from my own experiences of learning different things over my life but you explained this well and I found your words truly encouraging. Thanks.
The code at 2:24 is very wrong and bad at the same time A better version would be : ----------------------- n = 10 for i in range(n): if i == 4: continue
Yeah hahaha I looked at the edit so many times, but never realized there was a bug in the example code until I rewatched it when it was already live 🥲 Correct (and better) would be n = 10 j = 0 for i in range(n): if i == 4: continue j += i
What is your opinion on fully remote ML engineering positions?, it is realistic for someone that lives for example, in Latam to look for a job at this field?
There definitely are companies that are remote work heavy (Weights and biases, Huggingface, AirBnB) and there I don‘t see much reason why it shouldn‘t work (given you have the skills they require). :)
proofs are there so you understand the thinking behind the end result. It isn't there so you memorize it, its there so you understand it and can make the concept yours to be adapted for your use. Most concepts already are as generalized as can be but having it as a tool set rather than as a rule is the main idea.
Hi 👋, great video! I’m a bachelor college student and I wanna start with research (as long term goal) in the field of ml. I learned the basics and did a few projects. As I said, my long-term goal is to provide truly new insights in this field. Do you have any advice to get there and maybe how to find a person that could help on my way? Are there any contact persons or institutions in German universities?
I would say, your best bet is to look around your college (or other ones if you want to relocate) and find departments that do research in a domain that interests you. Get to know other PhD students or even professors that can be your advisors. Your bachelor thesis can be the first step of doing research. In your masters, you can then work as a part-time student researcher and/ or on cool projects as part of your program (that give credit points). Either way, it always comes down to how much effort you put into your work, but I can highly recommend to find people who are ahead of you who can mentor/ advise you. At college, you should be bale to find such people :) Good luck and have fun learning! 💛
Boris i really like u and connect with ur style. U simply state ur experience and how u r improving and i somehow keep that as a benchmark and learn from that
not quite sure what you mean or how that would work tbh. They are fundamentally different models. Perhaps you could use the features your resnet produces as input to one of the classical models (like a SVM) but not sure if that would work.
Okay, what I mean is I want to use a ResNet50 to extract features of Images,then use those features to train a classifier like random forest or SVM,so when am requesting,is how do I go about it,or even point me to some resources that I can use,thank you
I don't know any particular resource for this approach :/ But I think what you just described is all I can think of. You have a (pretrained) ResNet feature extractor and feed those features as a feature vector into let's say a SVM. There are libraries with which you can very easily try out this idea and perform some experiments! Sounds interesting! Good luck 😊
Ask this exact thing to chat gpt.ask gpt to break it down into approachable and practical step.basically prompt chat gpt until it gives you step by step description. First understand the problem yourselve.....u will be better able to explain it to gpt.good luck.
Nice video, agree with the points. I think your last point is particularly important. Patience! It takes time to learn complex things. We live in a world of (learn/do) xyz quickly, so expectations need to be managed.
Regarding the Rule nr 1: as much as I liked and respected my math teacher in high school, that was exactly problem. She would write some formula on the blackboard and start from there. Then I realized that function in programming languages is exactly the same thing as a function in math, or that many of the formulas were used to solve some real life problems, or were probably inspired by real world problems. At least that's what I ignorantly assume: the problem in real world was the first, then the "solution", the formula was invented, it didn't happen in vacuum.
Needed this so much. I have been sitting and solving issues for the entire day. I am working at a startup.Was just so much frustrated. Can anyone suggest a good community where we can post our queries and get a bit sonner replies?
Yes, sadly those type of questions are still going to be used for ML engineering interviews. Perhaps no system design in the end but rather „ML design“
Uhm... I'm not sure if I can give you a good answer. It varies a ton. It depends on you skill level and how much time you have (if you need to get it done in 1 week or 1 year). So sorry for the unsatisfying answer 🥲
Pretty good points. A couple comments from a 16+ year software engineer with a Master's in CS: - Understanding that a sum corresponds to a for loop is great, but in practice, it should still almost always be thought of as a function. You should think in self-documenting ways, and this will also lead to easier optimization later. For example, libraries that can vectorize functions like sum can perform operations on large datasets much faster than a for loop would. - Good advice for examining ML codebases, but more generally _follow the data flow._ Knowing how data flows through a system makes understanding the pieces along the way much easier. This is also a good way to estimate code quality. If the data flow of the system is easy to follow, it might be good code. If it's hard to follow, it's almost certainly poorly structured code, and changing or adding anything will be a struggle without refactoring first.
Things i wish i was told about math... "see in pictures." I know that this serves geometry more (i heard it from dr takashi lecture in south africa). And little by little, i am understanding the things i learned years ago as an undergrad of physics. I think the fact that it is a language is rarely spoken about.
Best advise is actually the last one. I hate how the first few are not so useful because it's already obvious that the key to understanding machine learning is understanding the math and the implementation. could use some discussion about the best beginning topics such as understanding of some functions that are widely used
The last one was also really important to me. It's often difficult to intuitively asses what is obvious to others when one falls into the "curse of knowledge" where those things are obvious and trivial to oneself. The other 4 tips are there to help people be less scared of math and code (which many are) and realize that it still requires practice, but is less "impossible" than you might initially think (before knowing these tips). I hope this makes sense :)
Great video! Can you link us to a good tutorial that explains how to take advantage of the "breakpoint/debug" method you described to better understand large codebases? I currently just insert print statements everywhere to better understand large codebases, but I'm eager to use the debugger if it is a more efficient method. Perhaps you can make a video of you explaining how you use the debugger with a simple Python script? Also, there are different ways of inserting break points: (1) in VS Code, or (2) in any general IDE, you can import the 'pdb' module and insert a breakpoint manually?
Very nice explanation. Very genuine in terms of setting expectation and explaining the reality. I am no AI guy, no interest in doing so, but really captivated by his explanation and understanding, even though he is very younger than me. :D
There indeed are a lot of similar projects on the internet. One tip I like to give is to not necessarily reinvent the wheel, but try to develop a "standard" model that works on your own custom data. E.g. when looking at LLMs, you could want to fine-tune a model on your native language where you might need to collect your own data! As long as you are building things yourself, show how you are solving real problems and learn a lot, you have a decent project! I hope this somewhat helps :)
hey your suggestions are really useful unlike the roadmap of other creators thanks a lot for this. I am really confused that should i start learning about deep learning and neural networks or I should focus on the supervised learning algos like regression, classificatin and xgboost stuff that is used in data science as my goal is pursue carrier in data science ?
Thank you! So, in any case you should learn the fundamental ML techniques (ideally before delving into deep learning). Now, when it comes to data science, then you would probably want to indeed focus a bit more on the classical algorithms and xgboost, since when it comes to DS, your main job will often be to work with tabular data and analyse that so that you can then potentially apply a rather simple ML algorithm (e.g. xgboost). P.S. I will soon publish a video on the different ML jobs. Perhaps that might give you some more insights :)
@@borismeinardus your response is highly appreciated! so what i got from your response is that i should focus majorly on all the classical algorithms and statstics stuff which is majorly used by data scientists and should avoid jumping on neural nets and deep learning. waiting for the video.......
@@kishantripathi4521 It‘s good to also learn about the recent DL developments, but you will very likely be working with techniques like xgboost, yes :)
Hey! Your videos are really awesome to watch and helpful. Also, How to start reading research paper? Should I start from reading basic papers and increase the difficulty? Can you give more insights on this?
Thank you! So, yeah, in general, I would suggest to work your way up in complexity. But papers are inherently a bit more difficult to get used to. I would recommend to first start watching UA-cam videos on paper explanations and then read the same paper yourself. That should be a much nicer start! From there on, it just comes down to practice. After a few paper, you get used to the lingo and learn to read them like a normal report or news article :)
hey @Boris Meinardus given your expertise in ML what papers should i read if i wanna build a product / a tool like say sora or any generative ai any advise would be much appreciated!
There are a lot of details in those state-of-the-art models but they all still build on the diffusion model. I would recommend you start by studying (Variational) Autoencoders -> GANs -> diffusion models (and possibly autoregressive image generation models like Google's Parti model)
Become better at machine learning in 5 min/ week 👉🏻 borismeinardus.substack.com/
There were videos that you post about Computer Science vs Computer Engineering in TUBerlin, where can I find them. I wanted to study CS at first but my NC is not enough, now i am thinking about Computer Engineering. But I want to focus on Software instead of Hardware or electrical side of things. But maybe it is better to get a job with knowing hardware side of things too.
I mean when I need your ideas most, they are gone! I remember watching them 1 year ago but I couldn't find it, do you have these videos somewhere, or at least can you answer if my ideas are right or not, or maybe are you happy with computer engineering bachelor?
Lol hopefully it is not ...
I gave up learning deep learning many times. Then in 2021 i had painful divorce. It changed everything instead of going into the depression i channeled my energy into learning deep learning. Now i am able to train small llms and text to speech models from scratch on multiple gpus. I am starting a company in the next month!
Wow! Thank you for sharing this impressive story from being at a very low point in life to such an exciting time of starting a new company!!
I can't imagine how tough a divorce can be...
I am certain you had a hard time, but it seems you are doing much better now.
I genuinely wish you all the best! Much love ❤️
I am at the stage where you where at 2021 . Any tips
I am at the stage where you where at 2021 . Any tips
Please share your roadmap, every time I try to learn ML/deep learning there are soo many concepts that I need to learn. which in the end I fail as I give up on something.
@@jayanthsattineni2151 i am trying to share the roadmap but could not see the replied comment not sure why
I took Deep Learning and Machine Learning subjects at the university last semester and now in the current semester. The Deep Learning professor would explain the theoretical process in plain English with graphical representations before even showing any math formulas. By that time, I get quite comfortable with what I'm learning or about to learn since I know the process and the end goal. The only thing I look for in the formulas is certain details or how exactly what is happening in each step if I'm too curious. On the other hand, my Machine Learning professor, if I attend the class, all I hear is math. When I go home, all I have to study from is the slides, which are again all math with no explanation in plain English. I ended up dropping the subject. It's crazy how a teacher can make someone love/hate a subject or make them feel good/bad at the subject
schools/colleges on purpose teach half knowledge, that whay I understood. The only way to learn is use chatGPT like tools and learn by your own step by step without relying on these misguiding education institutes
I think the perfect approach would be a combination of both. I think you should always start with the intuition, but not skip the math.
But I also do agree that classical ML has more (more difficult) math than DL
@abhishekdhaka4833yes, the math is important (possibly the most important thing about machine learning). Sometimes though, people need a review. I for instance needed a review on the Chain rule from Calculus to understand back propagation in Machine Learning. Having slides with graphical representations help too, for instance, when we look at activation functions like sigmoid, tanh, ReLU, etc. it is good to actually get a graphical representation on how it works.
From my current Deep Learning course that I am doing, I was having a hard time understanding convolutions at first until I watched Dr. Andrew Ng's Deep Learning videos. To be honest, I still have a lot of difficulty understanding CNNs, RNNs, etc. but I definitely feel that eventually I will have rock solid understanding of it.
@abhishekdhaka4833 I think you've missed the point of this persons comment.
Try UA-cam university. You will be surprised
"Writing code is not actually coding, debugging is coding." perfect bro
The day I stopped hammering formulas into my head, I understood how amazing math actually is! My advice to my younger self would be to try to understand the intentions behind all those formulas and not to force it. Occasionally, you discover why you need it when you need it.
💯💯💯
But how do you understand the intentions? Where do you get the resource?
agreed
the one thing that helped me with math is having a bad memory. Means i can't remember formulas so i have to try to understand them lol
such a classic comment. ❤@@vladnedelea05
00:01 Understanding math the right way is crucial for learning machine learning.
01:39 Translate human ideas into math concepts
03:18 Understanding math derivation rules simplifies ML learning
05:02 Recognize patterns and practice is key to mastering ML.
06:40 Realizing that debugging is coding helps with implementation
08:18 Understanding large code bases with a simple strategy
09:56 Setting breakpoints in the main function for debugging is crucial in understanding ML implementations
11:41 Mastering ML takes time and dedication.
Crafted by Merlin AI.
Goddamn bravo
Oh damm 😯
How did you learn to do this?
Thanks brother
Its like you have opened a new window for me into ML world. I was stuck at figuring out which topic falls into which category, but after listening to you it all came together. Now i can organize my thoughts properly while thinking and learning. I can relax while reading math formulas and be content by just understanding the intent behind it. I can relax while going through derivations knowing that it is pure mathematical realm and i can tackle it separately. And finally now i am slowly beginning to understand how to break down my tasks and work aka apply first principles thinking to ML problems. In short, you have helped me form a initial mental schema for ML. Thanks a lot !
Wow, thank you so much for these kind words!!
I am really happy to be able to at least somewhat provide value with my videos :)
I wish you all the best and happy learning 💛
@Neuralbench : Great comment , very illustrative and cogent, will follow you down that long dusty road...
He forgot rule 1: talent for sitting and focusing for days without getting up for food or daylight
that‘s not just a talent, that‘s a superpower, lol
Do it before you have kids
@@Stan_sprinkleyou probably can still do it when you have kids (take them to the library with you).
Well, I was broke with no marketable skill set and was already hungry and couldn’t afford an electric bill. So, I decided sitting for 5 years in the darkness working my way toward the light and finally reaching it was better than being a loser forever. But hey, what do I know 😂
You can do this for a few months, probably even years. I did that when I studied for university exams. Then you realize your body starts to weaken and your brain is not working as it should be (e.g. attention problems) and it takes months to fix that. I think you can do this realistically for a few weeks, however, it is not maintainable in a long term. What is much better is to create a robust long-term daily regime you can follow for a longer period of time in what you can maintain your health and productivity. I would advice this to a younger myself.
I'm an 8th grade math teacher, and the first "secret" you talked about is my main point throughout the year. Math is just transltaing what we think to a more convenient way to write it. The problem is how most of us are being taught math since 1st grade, where they show us on side of an equation and tell us to solve it. so we are getting used to think the math must leads to a solution, instead of being a description. Math equation in schools should be called "fill in the blanks" and it should be taught in a same way a language is being tutored.
1. Think of mathematical formulations as being an idea from someone else, which is only being put in math terms as a tool
2. Math derivations are the continuous applications of rules
3. Debugging is a part of coding
4. Set a breakpoint in the code
5. Mastering ML is hard
This video is a breath of fresh air. I was frustrated because I couldn't understand how to use einops for 3 hours and I was only writing 1 line, I'm glad I stumbled upon this
Thanks for this (and also for the other videos) man! You can really see you're down to earth and you'va faced the same struggles everyone has in the journey towards becoming better ML scientists/practitioners!
I really appreciate that!
We all have faced or will have to face a certain amount of struggles at some point haha. I'm trying to share what I have learned so that other people can hopefully avoid certain struggles or to at least somewhat help with overcoming them :)
@@borismeinardus i m learning ML , currently learning python libraries as pandas , hope u will keep guiding us , thanku
Thanks for my birthday present. It’s nice to know that still there are people who believe in a system. Much appreciated.
I am 12th class student i have decided to start machine learning after my exams and in my Bachelor degree... UA-cam has recommend a great video to me...🎉
🤩🤩 Best of luck
The thing that gets me about Deep Learning are RNNs, then going through LSTMs and GRUs. I try to memorize what happens throughout the entire RNN with activation layers and backwards propagation through time. It is just a lot of over-convoluted complex stuff. I just started learning Machine Learning and Deep Learning in August of 2023.
I am just struggling so much with it, but I love it. I love watching lectures on Transformer sequences with encoder and decoder even though at first a lot of it sounds like Ancient Greek to me. I will keep trying
Yeah, those can be tricky to understand. But don't get hung up on them if you don't understand a specific detail. Continue with looking at the whole lecture on the topic and revisit the detail you didn't understand once you understood everything else.
try stat quest
This was actually helpful! Particularly your insights around getting thrown math formulas and being expected to understand it. It totally resonated that you need to understand the intuition before you can hope to understand the formula. Thank you so much!
1. Math intuition 2.Foundation of math toolkits 3 learn how to debug 4 overview of the code base and foundation of code 5. As a PhD student now, I have to say it takes a lot of time and loads of stuff or even new things to keep up with everyday. Try to enjoy it
Code is not really an issue for me, the math is my frustration.
I just started few weeks ago, I felt very lost, but I am still learning and reading things, thank you so much for this amazing video my friend!
How is it going?
¡Gracias!
I am reading it mate! It was actually a learning update. This year will be an intense core year of studying and personal projects.
As an software engineer by profession and being coding for more than 10 years, I 100% agree with your point. I have never met anyone who can just write a function and run it successfully everytime. Debugging is so integral to coding that I don't even treat it as a separate thing.
Hello!
I really admire the way you teach something. Maybe we can have a sample tutorial of a ML / DL project with the standard structure for industry level
Thank you :) Really happy to hear that!
Building a project at an industry level is quite challenging in itself and tricky to put into one video haha
I have a video on how build out a project of reimplementing a paper. Perhaps that might be helpful :)
Brothers who stuck in this type of problem typically math related problems i suggest u to follow NPTL IITM deep learning course..
i am beginner can you guide me please in college
Link pls
Thank you for this video clip. I am embarking on machine learning by being in charge of a team that know this already, to complete a project.
Of course. It's more important to understand the concepts behind math than to try to understand how to interpret equations. Mathematics is just a very specialized language, nothing more. It's a kind of shorthand for high-level concepts. This is why, as a student, I eventually figured out that it was best for me to just sit back during class and listen and truly understand the ideas presented instead of record accurate notes.
Thanks a lot for sharing the truth of learning machine learning. It was not less than a pain killer.
🙏🏼💛🙏🏼
Hello Boris sir, thank you for your youtube videos. Moreover, I want to ask whether buying M4max apple with 128 gb laptop for machine learning, AI and Data science or buy M4 max with 36 gb and use cloud for higher data computation. It would be our pleasure to have on best laptop idea for these categories.
My man you and I think the same. Trust me I struggled with implementing a simple summation in my first code back in the days when I was doing my masters thesis.
I was looking at a simple formula i.e. L1 = exp(2*pi*1575.42e6*time + doppler) and I literally had no idea to work implement this to generate a GPS signal in MATLAB and Python.
bro, colleges, shcools sucks at teaching maths, I feel sometimes its on purpose. Math is most important tool, but if you learn you would be much more powerful than people who want to have power in the society
@@ravirajac I can't agree more
Yeah, I enjoy math so much and am really sad when others hate it because they had a bad teacher. Luckily there are amazing videos on youtube that can assist the learning process ☺️
Great video!
I am a college student and has already join some ML/DL courses. But I always think that the college resource cannot compare to some big companies like OpenAI, Google, Meta and so on. Lots of ML techniques have been mature by them. What could we do as just a college student? Thank you !
That is very true. The college courses can't really keep up because everything is moving so fast. It is very hard for them to update the whole curriculum of a course each semester. Nevertheless, there should exist seminars at college where you discuss recent research in smaller groups. That could be one option. Otherwise, it is up to yourself to keep up :/
College teaches the fundamentals of ML and DL, which you definitely need. But beyond that, you need to keep up on your own, which college will hopefully have prepared you for by providing a good fundamental understanding.
That said, it's just moving so fast, you can't feasibly keep up with everything. Try to find a domain that interests you where you can dive deeper and also work on respective projects.
I hope this somewhat helps :)
Thanks for your reply. Selecting a specific field seems to be a good choice. However, most of the domains have been prosperous. Many online courses cover diverse topics and I do not know how I should start.I want to know how you can find your way. Thank you again!
@@chiahungchiang9506 Oh my, not sure if I can help you there haha. I agree there are a lot of interesting fields. Explore as much as you can, see what you can see yourself actively working on, and perhaps consider the what the labs at your college offer. It doesn't really make sense to go hard on e.g. RL if none of you college labs has expertise in RL. I mean, you can, but you won't be able to get the best support.
I think it is somewhat a blessing if your college doesn't keep up with the latest hype. It is hard to know if the hype is useful after 5 years. I studied a lot likelihood functions (following George Casella's "Statistical Inference")though nowadays there isn't much hype around them but understanding MLE gave me a solid foundation
thanks for your advice, even though i'm still not knowing how can i improved my knowledge in the coding during my 5 year studying in computer science major but here are all the advice i need to find my own path in the world coding. you are right to get mastering in certain area you need to sacrifice your hard work and time to get through over it! starting now again i will try more harder
So so so grateful for this video. I have an AI Application I want to design and was worried about the math peice until I watched the Harvard Deep learning video and now you reaffirmed what I understood there.
Do you need a data science intern for your AI Application?
I am glad to help with the project to gain hands on experience
You just gotta think in the language of 'Moth'. like a 'Mothamatician'
@9:46. I agree that memorizing or coding opimtization is overkill. I wanted to learn everything by scratch and learnt that's a fool's errand. I learnt that the hard way. In deep learning, learning back propagation for FNNs, CNNs, RNNs are good ideas. But it is better to concentrate on the matrix algebra aspects of the forward prop than focus on the optimization part since most of the back prop is efficiently done in Tensorflow or PyTorch or for any ML algorithm in Scikit-learn. The mathematical derivation of any algorithm is more important than the cumbersome calculations. I took the Google's Tensorflow certification exam last year. What I learnt from the exam is the core idea of fitting the inputs and outputs in the correct matrix or numbers or any data augmentation or manipulation that better fits the model than raw data. To sum up, the practical aspects of the ML is all about how we frame the problem and how efficiently we solve the problem for real life use and debug our model when it goes crazy or throws up errors.
I'm enjoying your video thus far. Some constructive criticism: The summation code shown at 02:25 is wrong. It modifies the loop index variable, "i" and the "range(n)" function stops before 10, giving only 0-9. Your code has a value of "10" in "i" after the loop completes. However, if you change the variable "i" to "x" on lines 3 and 4 and increment the value passed to "range()", "i" ends up with the value "10". An equivalent one-liner: "print(sum(1 for i in range(n+1) if i != 4))".
Although my advanced math is VERY rusty, I'm certain that answer is incorrect. You must add the value of the loop index variable with each iteration, NOT 1. That is: "print(sum(i for i in range(n+1) if i != 4))".
You speak some straight up truth in this video, these are super helpful insights into how to learn. One thing I would say is for programming you kind of go against the super good advice you give for learning maths. When trying to understand a big codebase, the debugger will not help -- the debugger is there as a magnifying glass into all the details. But just like your example in maths, code is just human ideas expressed a certain way, so understand the higher level ideas of functions or modules, and go deeper as your understanding improves. The debugger is great for when you are debugging the behaviour of your code, or if you are very familiar with the code someone else wrote.
Thank you!
I definitely see your point! Perhaps I should have better described the scenario where the debugger is very useful to understand a code base or model.
When I e.g. wanted to better understand a certain Transformer model, I understood the high level idea, but then wanted to understand the details. I then imported the huggingface implementation, set my breakpoint, and went into the code. I got to see the data preprocessing, data flow through the model, and so on. That‘s where I fell in love with using the debugger to understand code/ large code bases.
But yes, you will most likely not want to have a certain high level idea of what you are looking at, but I still believe the debugger is one of the best ways of learning to understand code :)
I hope this makes sense ☺️
Nice video! Machine Learning, Deep Learning, NLP are all challenging subfields of AI. I am in the process of taking them right now in graduate school.
Amazing! Best of luck!
that last part of the video , is just so calming
💛☺️
Hey Boris, Can you help me with my little question..The question I had was, what topics of ML should I know to get my first job next year?
And BTW, I'm way too fascinated by your knowledge, and believe me when I say you have helped my love for ML grow by 100x. Nothing's able to demotivate me anymore, and I'm maintaining my pace. I'm following your roadmap too.
Thinking of different degrees. I saw and experienced statistics, computer science, physics, and cognitive science and contemplated the curricula at my university (Helsinki). Many times I felt these fields to go deep in their own direction, if one wants to go primarily towards ML or AI. Then the internet is full of learn X quickly types of content.
However, what I get from your experience is, that perhaps your/some ML curricula is streamlined in a way to get optimal amount of depth and breadth specifically for ML. For me the big question is, how deep into math one should go, because prioritization gives you other skills too...
great job mate really I hope they teach us this stuff at the uni. Keep it man
Bro you are making me get there.Everyday I spend 3+ hours coding and learning something new of machine learning.So far I can develop Linear Regression models,Logistic Regression and Decision Trees. I am a final year student for Bsc in Mathematical Sciences and multi-tasking.Soon going for an honors degree to advance.I do work on projects too,I hope that in the next 6+ months I will be here to tell you I got a job in machine learning,now its 02:00 (South Africa) and here I am on Yt for ml ,passion!!!
talk when you will get a job , thats what matters
I'm too pursuing bsc mathematical science, India
amazing! Keep it up! See you in 6 months 😤😊
Amazing video. This video comes at the right time because as a college student, who is about to make an ML course in the next semester, I was a bit anxious as well. Thank you for the guide!
Really happy to hear that ☺️💛
Good luck and try to have fun while learning! :)
Thanks for this video. Many of my doubts are clear because of this video.
What is Mac specs to start learning ML? Currently I have an old MacBook Air which is I’m not sure to start with! Memory size, speed processing…etc
You are honest man, thank you🎉
My pleasure! Doing my best 😬😊
Нереально мощная связка ❤❤ спасибо тебе за контент
¿Podrías hacer un video sobre cómo gestionar el riesgo en trading de opciones?
Hello I would like to enter the realm of Machine learning, but as you know AI is becoming better and better everyday, what should I do to be a Machine Learning Engineer...what's would I study?
Wonderful video 🙏🏻🙏🏻🙏🏻🙏🏻 i am a phd student and i am struggling a lot. This video gave me some hope.
Glad it was helpful! All the best! 😊
Good valid and applicable points: 1) Debugger as a self-teaching tool 2) work systematically 3) create a list of "math rules" to reflect your learning
Exactly 😊
Awesome!!! I loved your Math explanation! Thank you very much!
Your lessons help me improve my trading strategy. Thank you for your expertise and experience!⚠
Great video man, I learnt alot, I am just starting as an ML engineer myself
Exciting!!
Really glad I could help a bit with your journey ☺️
Happy learning 💛
Thanks! this video has really saved me from the upcoming traps and forged a path through.
Thank you Man. You made my day and my Future.💌💌💌
Hey boris I really like your advice on paper implementation but I am really confused on which paper to start implementing first. Could u please recommend papers which a beginner should focus on implementing
Watch the video "Let's build GPT: from scratch, in code, spelled out." Karpathy walks through implementing the transformer architecture from the paper Attention is All You Need. This is probably the best paper to try to implement from scratch given the ubiquity of transformers
Thanks for keeping it real! :) I have a renewed perspective and I think I'll be able to enjoy the process more!
really really good. I think you have explained this very well. I think I knew a lot of this from my own experiences of learning different things over my life but you explained this well and I found your words truly encouraging. Thanks.
The code at 2:24 is very wrong and bad at the same time
A better version would be :
-----------------------
n = 10
for i in range(n):
if i == 4:
continue
Yeah hahaha
I looked at the edit so many times, but never realized there was a bug in the example code until I rewatched it when it was already live 🥲
Correct (and better) would be
n = 10
j = 0
for i in range(n):
if i == 4:
continue
j += i
Please make a vid on how to train some model to generate 'text to image' over some template image (not a character). Both online / local machine.
great video!! some one on yt tells 6month enough to master something
What is your opinion on fully remote ML engineering positions?, it is realistic for someone that lives for example, in Latam to look for a job at this field?
There definitely are companies that are remote work heavy (Weights and biases, Huggingface, AirBnB) and there I don‘t see much reason why it shouldn‘t work (given you have the skills they require). :)
proofs are there so you understand the thinking behind the end result. It isn't there so you memorize it, its there so you understand it and can make the concept yours to be adapted for your use. Most concepts already are as generalized as can be but having it as a tool set rather than as a rule is the main idea.
Python program @2:28 is wrong, the 1 should be an i
Thank you so much for this Boris as has helped me so much as I am learning ML
Really glad it helped!!! 😊
@Boris Meinardus : What is the tablet you are using at 4:47 ?
Thank you for the video, it was very helpful. Any way you can share some of your notes such as the rules/definitions at 4:40?
after learning python, where do i start learning for ML. WHATS THE FIRST STEP?. HOW DO I START CODING FOR IT. BORIS CAN YOU HELP ME OUT
Thanks Boris, wonderfully clear video and really appreciate you sharing with us!
Hi 👋, great video!
I’m a bachelor college student and I wanna start with research (as long term goal) in the field of ml. I learned the basics and did a few projects. As I said, my long-term goal is to provide truly new insights in this field. Do you have any advice to get there and maybe how to find a person that could help on my way? Are there any contact persons or institutions in German universities?
I would say, your best bet is to look around your college (or other ones if you want to relocate) and find departments that do research in a domain that interests you. Get to know other PhD students or even professors that can be your advisors.
Your bachelor thesis can be the first step of doing research. In your masters, you can then work as a part-time student researcher and/ or on cool projects as part of your program (that give credit points).
Either way, it always comes down to how much effort you put into your work, but I can highly recommend to find people who are ahead of you who can mentor/ advise you. At college, you should be bale to find such people :)
Good luck and have fun learning! 💛
@@borismeinardus thanks for the advice👍 and have a nice week.
@@dassystem1837 Thanks, you too 😊
Boris i really like u and connect with ur style. U simply state ur experience and how u r improving and i somehow keep that as a benchmark and learn from that
What is the source for that math for AI and programming
Thanks a lot. This will be helpful for me
I am just in high school, trying to learn ml by myself , got stuck in linear regression, i get how it works , but cant progress
Can you give source of math that supports your learning like book or website that you see it easy to understand and how you can find it please
How can Impliment Hybrid model on ResNet5o and any other classify(e.g RandomForest, or SVM etc), any Resources will be Helpful, thank you in advance
not quite sure what you mean or how that would work tbh. They are fundamentally different models. Perhaps you could use the features your resnet produces as input to one of the classical models (like a SVM) but not sure if that would work.
Okay, what I mean is I want to use a ResNet50 to extract features of Images,then use those features to train a classifier like random forest or SVM,so when am requesting,is how do I go about it,or even point me to some resources that I can use,thank you
I don't know any particular resource for this approach :/
But I think what you just described is all I can think of. You have a (pretrained) ResNet feature extractor and feed those features as a feature vector into let's say a SVM. There are libraries with which you can very easily try out this idea and perform some experiments! Sounds interesting! Good luck 😊
@@borismeinarduslet me try I will get back with the initial results, thanks
Ask this exact thing to chat gpt.ask gpt to break it down into approachable and practical step.basically prompt chat gpt until it gives you step by step description.
First understand the problem yourselve.....u will be better able to explain it to gpt.good luck.
thanks brother, your videos are incredibly useful!
Glad you like them! 😊
So relatable ...I just relived all the moments you mentioned in detail ... Though I had given up on ML ....will start again now ❤
Nice video, agree with the points. I think your last point is particularly important. Patience! It takes time to learn complex things. We live in a world of (learn/do) xyz quickly, so expectations need to be managed.
Absolutely!
Regarding the Rule nr 1: as much as I liked and respected my math teacher in high school, that was exactly problem. She would write some formula on the blackboard and start from there. Then I realized that function in programming languages is exactly the same thing as a function in math, or that many of the formulas were used to solve some real life problems, or were probably inspired by real world problems. At least that's what I ignorantly assume: the problem in real world was the first, then the "solution", the formula was invented, it didn't happen in vacuum.
Needed this so much. I have been sitting and solving issues for the entire day. I am working at a startup.Was just so much frustrated. Can anyone suggest a good community where we can post our queries and get a bit sonner replies?
Would you recommend doing leetcode-type questions if the end goal is to become a machine learning engineer?
Yes, sadly those type of questions are still going to be used for ML engineering interviews. Perhaps no system design in the end but rather „ML design“
You just told the honest journey of ML. Thanks a lot
Very nice explanation by breaking things down and addressing the right points. This is very helpful and encouraging. thank you for this video.
Thanks man.. 🤗
Enjoyment = win
No need to reconfirm thx 4 vid
Like how many hours or days did it take you to build on large code base
Uhm... I'm not sure if I can give you a good answer. It varies a ton.
It depends on you skill level and how much time you have (if you need to get it done in 1 week or 1 year).
So sorry for the unsatisfying answer 🥲
Pretty good points. A couple comments from a 16+ year software engineer with a Master's in CS:
- Understanding that a sum corresponds to a for loop is great, but in practice, it should still almost always be thought of as a function. You should think in self-documenting ways, and this will also lead to easier optimization later. For example, libraries that can vectorize functions like sum can perform operations on large datasets much faster than a for loop would.
- Good advice for examining ML codebases, but more generally _follow the data flow._ Knowing how data flows through a system makes understanding the pieces along the way much easier. This is also a good way to estimate code quality. If the data flow of the system is easy to follow, it might be good code. If it's hard to follow, it's almost certainly poorly structured code, and changing or adding anything will be a struggle without refactoring first.
Things i wish i was told about math... "see in pictures." I know that this serves geometry more (i heard it from dr takashi lecture in south africa). And little by little, i am understanding the things i learned years ago as an undergrad of physics.
I think the fact that it is a language is rarely spoken about.
Sir if you can give books and resources exactly from where to learn maths and maths behind code , I am ready to give full attention . Thanks a lot !!
Best advise is actually the last one. I hate how the first few are not so useful because it's already obvious that the key to understanding machine learning is understanding the math and the implementation. could use some discussion about the best beginning topics such as understanding of some functions that are widely used
The last one was also really important to me.
It's often difficult to intuitively asses what is obvious to others when one falls into the "curse of knowledge" where those things are obvious and trivial to oneself.
The other 4 tips are there to help people be less scared of math and code (which many are) and realize that it still requires practice, but is less "impossible" than you might initially think (before knowing these tips).
I hope this makes sense :)
Great video! Can you link us to a good tutorial that explains how to take advantage of the "breakpoint/debug" method you described to better understand large codebases? I currently just insert print statements everywhere to better understand large codebases, but I'm eager to use the debugger if it is a more efficient method. Perhaps you can make a video of you explaining how you use the debugger with a simple Python script?
Also, there are different ways of inserting break points: (1) in VS Code, or (2) in any general IDE, you can import the 'pdb' module and insert a breakpoint manually?
Very nice explanation. Very genuine in terms of setting expectation and explaining the reality. I am no AI guy, no interest in doing so, but really captivated by his explanation and understanding, even though he is very younger than me. :D
How to find a unique project as already there are so many same projects on internet
There indeed are a lot of similar projects on the internet. One tip I like to give is to not necessarily reinvent the wheel, but try to develop a "standard" model that works on your own custom data. E.g. when looking at LLMs, you could want to fine-tune a model on your native language where you might need to collect your own data!
As long as you are building things yourself, show how you are solving real problems and learn a lot, you have a decent project!
I hope this somewhat helps :)
I’m so happy to watch your video
🥰
I am middle schooler with IGCSE level knowledge. Is it possible for me to learn machhine learning online just by myself??
hey your suggestions are really useful unlike the roadmap of other creators thanks a lot for this. I am really confused that should i start learning about deep learning and neural networks or I should focus on the supervised learning algos like regression, classificatin and xgboost stuff that is used in data science as my goal is pursue carrier in data science ?
Thank you!
So, in any case you should learn the fundamental ML techniques (ideally before delving into deep learning). Now, when it comes to data science, then you would probably want to indeed focus a bit more on the classical algorithms and xgboost, since when it comes to DS, your main job will often be to work with tabular data and analyse that so that you can then potentially apply a rather simple ML algorithm (e.g. xgboost).
P.S. I will soon publish a video on the different ML jobs. Perhaps that might give you some more insights :)
@@borismeinardus your response is highly appreciated! so what i got from your response is that i should focus majorly on all the classical algorithms and statstics stuff which is majorly used by data scientists and should avoid jumping on neural nets and deep learning. waiting for the video.......
@@kishantripathi4521 It‘s good to also learn about the recent DL developments, but you will very likely be working with techniques like xgboost, yes :)
Hey! Your videos are really awesome to watch and helpful. Also, How to start reading research paper? Should I start from reading basic papers and increase the difficulty? Can you give more insights on this?
Thank you!
So, yeah, in general, I would suggest to work your way up in complexity. But papers are inherently a bit more difficult to get used to.
I would recommend to first start watching UA-cam videos on paper explanations and then read the same paper yourself. That should be a much nicer start!
From there on, it just comes down to practice. After a few paper, you get used to the lingo and learn to read them like a normal report or news article :)
hey @Boris Meinardus given your expertise in ML what papers should i read if i wanna build a product / a tool like say sora or any generative ai any advise would be much appreciated!
There are a lot of details in those state-of-the-art models but they all still build on the diffusion model. I would recommend you start by studying (Variational) Autoencoders -> GANs -> diffusion models (and possibly autoregressive image generation models like Google's Parti model)
@@borismeinardus Thank for the insight boris Cheers🥂
@@oii0712 🍻