00:03 Focus on project-based learning to reinforce ML concepts. 00:53 Build a sentiment analysis model for mastering NLP basics. 01:45 Choose the right model for sentiment analysis based on your experience. 02:35 Train a mini LLM to enhance NLP skills. 03:27 Enhance your ML skills with practical projects and fine-tuning techniques. 04:19 Understand foundational concepts to fine-tune Transformer models effectively. 05:15 Leverage retrieval-augmented generation (RAG) for company-specific data access. 06:04 Implementing a CNN from scratch is vital for your ML resume. Crafted by Merlin AI.
Thank you for this video. I’ve noticed that when it comes to concepts like software development, there are thousands of videos on potential projects you can make, but change the subject to machine learning and there are probably less than 500. I appreciate this video as someone who just started their masters in ML and wanted to create projects to learn but didn’t know where to start.
What I like about ML more than SWE is that I don't bother thinking about a new project, I just go through some big deep learning paper and I try to reimplement it from scratch in pytorch. The list can be very long varying from detection or segmentation to text translation or sentiment analysis. It's litteraly an unlimited source of projects.
Just implemented a CNN using Numpy. Really been feeling the absence of impressive projects in my portfolio. Little reassuring to see that it was probably as impressive as I thought it was
This is something apart from the topic of the video, but since you mentioned that you will be graduating with an MS in ML, i wanted to ask what do you wanna pursue after this? And also as someone who is interested in pursuing Research in AI/ML, what tips would you give? Im currently doing my Bachelor's in Comp Sci, and i am very much interested in going into the Research aspects of ML
True. Become a youtuber, make your content with these thumbnails and titles and get ahead of 99% students who are spending time to learn DS while you are earning money through TH-cam
@@AbcTawte I agree that you are motivating him, which is the truest case. But I think the OP wants to understand the nuances of NNs and hence he is going vanilla.
Bruh. This is so irrelevant. Doing all this will put you on the average at most. You just throw llm, transformer at people while, practically, people implement less complex algorithms like seq2seq model since transformer is too expensive. Just get the math down, understand business problem and what you optimize, learn basic architectures, and you should be way ahead of the curve.
I would say that implementing a transformer and CNN from scratch (while really understanding all the details) are definitely above-average projects. But it's definitely a valid point that in some cases, transformers are too expensive and are overkill. Though, there are cases where they're incredibly valuable (fine tuning & RAG use cases) and more and more techniques are being developed to make inference less expensive. Two of the projects on the list (fine tuning and RAG) also don't require diving into advanced algorithms or architectures and are highly in-demand skills for AI/ML engineers these days. I appreciate your comment and am curious to hear more about your take on this.
I completely agree with you. These complex architectures are diverting focus from what the basics of ml are. Build a best predictive model with least cost.
hi, im new into this data world or ML. would you please recommended sites to get datasets (i only know Kaggle)? whether it is clean or not, I also want to practice data cleaning or EDA
Welcome! When you're getting started, Kaggle is a great place to find static datasets. Eventually, you can move onto dynamic datasets and grab data using an API!
It's an ML programming platform that I developed with Navdeep (NeetCode)! I created the problems & test cases and he was kind enough to host them on his website's code sandbox. Check the problems out under the "Core Skills" section !
I wanted to be a Data analyst first then I saw the job market and realised, that's too much competition and thought ok, lets goal for a ML engineer, more requirement then Data Analyst therefore less competiton, OHHH how stupid was I, now I am watching videos like "Get ahead of 99% of Machine Learning students" with 30% understanding of ML concepts and 5% coding knowledge, only god can help me 🥲
More resources!
Intro to NLP: th-cam.com/video/jwmpxuMn7p4/w-d-xo.html
Intro to LLMs: th-cam.com/video/cW7IGFdAfrM/w-d-xo.html
LoRA (the Fine Tuning trick): th-cam.com/video/PwmB97JDptI/w-d-xo.html
RAG: th-cam.com/video/DYpSk8-38LE/w-d-xo.html
First-Principles Framework (Learn Fundamentals): bit.ly/3O0ZMab
Beginner's Blueprint (Build Projects): bit.ly/48LWuBb
00:03 Focus on project-based learning to reinforce ML concepts.
00:53 Build a sentiment analysis model for mastering NLP basics.
01:45 Choose the right model for sentiment analysis based on your experience.
02:35 Train a mini LLM to enhance NLP skills.
03:27 Enhance your ML skills with practical projects and fine-tuning techniques.
04:19 Understand foundational concepts to fine-tune Transformer models effectively.
05:15 Leverage retrieval-augmented generation (RAG) for company-specific data access.
06:04 Implementing a CNN from scratch is vital for your ML resume.
Crafted by Merlin AI.
3:21 Implement paper Attention is all you need
Thank you for this video. I’ve noticed that when it comes to concepts like software development, there are thousands of videos on potential projects you can make, but change the subject to machine learning and there are probably less than 500. I appreciate this video as someone who just started their masters in ML and wanted to create projects to learn but didn’t know where to start.
Glad you found this useful, and best of luck :)
I love the way you explain things. That's fast, concise, and clear. Keep it up!❤
What I like about ML more than SWE is that I don't bother thinking about a new project, I just go through some big deep learning paper and I try to reimplement it from scratch in pytorch. The list can be very long varying from detection or segmentation to text translation or sentiment analysis. It's litteraly an unlimited source of projects.
Exactly what i was looking for, love the project ideas.
Just implemented a CNN using Numpy. Really been feeling the absence of impressive projects in my portfolio. Little reassuring to see that it was probably as impressive as I thought it was
That's incredibly impressive! The forward and backward passes using only NumPy are no joke.
This is something apart from the topic of the video, but since you mentioned that you will be graduating with an MS in ML, i wanted to ask what do you wanna pursue after this? And also as someone who is interested in pursuing Research in AI/ML, what tips would you give? Im currently doing my Bachelor's in Comp Sci, and i am very much interested in going into the Research aspects of ML
You da goat fr fr
@@RoboticsandProgramming-y1x That’s you 🙏
Imagine those 99 percent students watcing this video💀.u are no more ahead only
True. Become a youtuber, make your content with these thumbnails and titles and get ahead of 99% students who are spending time to learn DS while you are earning money through TH-cam
I have built a very basic RAG work flow, but I'm unable to build a NN that's even barely good from scratch 😢
Well no worries knowing pretrained models and finetunning them is the new trend. Nobody expects you to beat a pretrained model.
Look for pytorch tutorial. It is very easy to implement in it
@@AbcTawte I agree that you are motivating him, which is the truest case. But I think the OP wants to understand the nuances of NNs and hence he is going vanilla.
Don't beat yourself up! Often, the tiniest bug can throw off your NN. Let me know if you figure it out. And congrats on building a RAG workflow!
Hey thanks for these ideas! Do you have any ideas for ML Engineering projects, using Spark, preparing an end to end ML Engineering solution?
Will add this to the queue of video ideas!
Bruh. This is so irrelevant. Doing all this will put you on the average at most. You just throw llm, transformer at people while, practically, people implement less complex algorithms like seq2seq model since transformer is too expensive. Just get the math down, understand business problem and what you optimize, learn basic architectures, and you should be way ahead of the curve.
I would say that implementing a transformer and CNN from scratch (while really understanding all the details) are definitely above-average projects. But it's definitely a valid point that in some cases, transformers are too expensive and are overkill. Though, there are cases where they're incredibly valuable (fine tuning & RAG use cases) and more and more techniques are being developed to make inference less expensive.
Two of the projects on the list (fine tuning and RAG) also don't require diving into advanced algorithms or architectures and are highly in-demand skills for AI/ML engineers these days.
I appreciate your comment and am curious to hear more about your take on this.
I completely agree with you. These complex architectures are diverting focus from what the basics of ml are. Build a best predictive model with least cost.
good video recommend more
Thanks for the support!
Can you divide your video into named segments for easier note taking?
Definitely - will start to add chapters back into the videos :)
hi, im new into this data world or ML. would you please recommended sites to get datasets (i only know Kaggle)? whether it is clean or not, I also want to practice data cleaning or EDA
Welcome! When you're getting started, Kaggle is a great place to find static datasets. Eventually, you can move onto dynamic datasets and grab data using an API!
As a non-machine learning student myself, any simple neural network isn't 'simple' xD
Ha ha...
hey anyone know what site this is at 0:52 ?
It's an ML programming platform that I developed with Navdeep (NeetCode)! I created the problems & test cases and he was kind enough to host them on his website's code sandbox. Check the problems out under the "Core Skills" section !
but how about the Computer vision approach? this is clearly for NLP engineer
I wanted to be a Data analyst first then I saw the job market and realised, that's too much competition and thought ok, lets goal for a ML engineer, more requirement then Data Analyst therefore less competiton, OHHH how stupid was I, now I am watching videos like "Get ahead of 99% of Machine Learning students" with 30% understanding of ML concepts and 5% coding knowledge, only god can help me 🥲
Don't beat yourself up - just focus on getting better every week :)
same percentage... hopefully hands-on ML book would help you!