How I Would Learn Machine Learning (If I Had To Start Over)
HTML-код
- Опубликовано: 7 сен 2024
- The first 200 people to sign up for Brilliant for free at brilliant.org/j... get 20% off the annual premium subscription!
If I had to learn machine learning from scratch, here’s how I’d do it. Check out @KenJee_ds 's channel for the inspiration for this video!
Nebula Link: nebula.app/vid...
Me on Other Parts of the Internet:
Twitter - / jordanbharrod
Instagram - / jordanbharrod
Tiktok - / jordanbharrod
RUclips and PhD Gear: www.amazon.com...
My Favorite AI Books: bookshop.org/l...
Anywhere Else You Might Find Me: linktr.ee/jord...
For business inquiries, contact jordanharrod@standard.tv
Resources:
MIT OpenCourseWare (Check out Course 18, which is the Math Dept): ocw.mit.edu/in...
Fast.AI: www.fast.ai/
Fast.AI Practical Deep Learning: course.fast.ai/
Fast.AI Practical Data Ethics: ethics.fast.ai/
Kaggle: www.kaggle.com/
Google Datasets Search: datasetsearch....
Papers With Code: paperswithcode...
Google Colab: colab.research...
Statistics Fundamentals (Brilliant): brilliant.org/...
Intro to Probability and Statistics (MIT OCW): ocw.mit.edu/co...
6.434 (MIT) Statistics for Scientists and Engineers: web.mit.edu/fmk...
6.436 (MIT) Fundamentals of Probability: ocw.mit.edu/co...
Real Analysis (MIT OCW): ocw.mit.edu/co... 
Наука
A few people correctly pointed out that I somehow managed to not include resources for learning statistics! I've linked a few classes that I've taken here and have updated the description.
Statistics Fundamentals (Brilliant): brilliant.org/courses/statistics/
Intro to Probability and Statistics (MIT OCW): ocw.mit.edu/courses/mathematics/18-05-introduction-to-probability-and-statistics-spring-2014/
6.434 (MIT) Statistics for Scientists and Engineers: web.mit.edu/fmkashif/spring_06_stat/6.434J-16.391J.htm
6.436 (MIT) Fundamentals of Probability: ocw.mit.edu/courses/electrical-engineering-and-computer-science/6-436j-fundamentals-of-probability-fall-2018/
Real Analysis (MIT OCW): ocw.mit.edu/courses/mathematics/18-100c-real-analysis-fall-2012/
Jordan had a very clear mind and great explanation skills. Your inspired me that human is an Machine at all levels including classical and quantum. Spirits who choose to have finite identities rely on tools such as AI to be balanced.
That last point is important. When your only tool is a hammer, everything starts to look like a nail.
Thanks for the shout out! Love how you described the "Why" behind each of your learning steps!
Here's a comment for the engagement algorithm.
Your video is basically telling me to not even start, there's sooo much to learn. Think I'll stick with the data collection and annotation management pathway along with humanoid robotics QA testing. Much easier route right?
This was great! I think one thing you missed though is learning the basics of scientific computing. If you’re doing any ML at scale, understanding compute clusters and distributed computing is critical.
I believe that term refers to the use of computation, to analyse, model & further other sciences. Cluster, parallel computing, etc are on the hardware side, and falls under computer science, so she didn't miss a beat. Scaling is important but on the software coding & api side. She's not a hardware dev, a programmer and a hardware dev or technicians are different disciplines under computer science. While we made breakthrough in AI/ML, the underlining technologies isn't new, servers, raid, etc, are clusters, we host websites, files, render content or just brute computing, while parallel computing has been used as graphic, sound, simulation etc. You just gave me an idea though, I would like to run ML benchmarks on a ASIC miner 🤔
@@Tr3xShad Anyone using ML at scale to solve problems in other fields (like Jordan, who specializes in medical applications) is doing scientific computing. Even if you’re using ML for purely computer science problems (like vision or NLP), it’s still considered applied and therefore scientific computing by most researchers (because computer science is a science after all). Finally, even if your application is completely theoretical or somehow outside of a “scientific” scope, the basic principles of cluster and distributed computing that were developed for scientific computing still get used for any ML at scale. Without understanding these concepts, a person new to ML might think they can fine tune a language model like BERT or GPT2 on their laptop (which is a common “let’s play around with a language model” intro tutorial approach to ML), when the reality is you need a dedicated compute cluster to use such models for anything more than a toy example.
@@qwerty_and_azerty Again you're just stating what I clarify without understanding the difference stated. You spoke about clusters, ML etc which are computer science. The use of computation both hardware & software in the study of natural science is what is referred to as scientific computing. Clusters or distributed computing were need because classical computer did not have the process power. Like I agreed with you that scaling is important, not on the hardware but software side. Now I understand you drag Jordan's personal in this but I kept it in context of the video, in that light, Jordan is an AI/ML content provider. Again she didn't miss anything, except you are assuming that her coding & AI lectures skipped all that, and No she wasn't listing every module or topic studied. You don't need hardware, clusters or networking knowledge to be an AI developer or use computation in science work. Point is, someone can be an expert in clusters & parallel computing and their scalability without having any ML or AI knowledge. Oh yh, because we mimicking neural networks in AI isn't a direct study of neural but a framework for processing info like the brain. If we had fast enough stand alone computers or quantum computers, we may not need a clusters 😉 but you still need to scale your code for difference platform and processing budget.
@@Tr3xShad During my CS undergrad, learning about clusters and distributed or parallel computing was NOT a requirement (there was just one elective that some chose to take). Now during my PhD in ML, I’ve had to learn about this stuff in order to run my experiments. There is no way I could complete my PhD experiments in a reasonable amount of time without it. So my point is that any person doing anything state of the art in ML, not just playing around with toy examples, needs to learn about clusters and distributed computing.
Now, most scientific computing courses happen to teach about clusters and distributed computing because that’s the bread and butter of how they run their experiments. And most people who use ML have a scientific application in mind, and are therefore doing scientific computing. So learning about scientific computing is a great way for almost all people to learn about both at once.
That’s my point. Basic courses tend to skip these lessons on scaling up experiments, but understanding how to scale them is very important.
I don’t really get why we are talking past each other on this issue.
@@qwerty_and_azerty That is true. The CS courses and the modules in some university focuses on the wrong things but hopefully with ML & AI being mainstream and becoming key skills within IT, 5G bringing a new era to distributed computing, I hope it becomes a fundamental topic.
This is something I really wanted to know 😄
Having just started my machine learning journey six months ago, this video is pretty great. It's always really nice to hear the thoughts of someone with more real-world experience!
What do you recommend for learning about optimal ways to deal with missing data in machine learning?
The last point gives me “tech solutionism” vibes. If you only have a tech hammer, everything looks like a tech nail.
Wonderful. Really useful.
Very helpful video, as someone who is interested in pursuing a career in ML
This is really cool, I just started my first summer research opportunity this summer last week and have been working on learning how to implement image recognition in Python. Thanks for the tips as I start my journey into machine learning!
These are really great suggestions! (Cool title animations by the way. I just noticed them)
Thanks! 😄
You inspired me to start a research apprenticeship in ML at my school and I’m so happy to be dipping my toes into the field! :,)
So much wisdom here. Rooting for you.
Thank you for the video. What are your thoughts on GNU Octave as an alternative to Matlab?
Since you mentioned differential equation, did you also looked at some of the theory from control theory, or maybe more in the direction of how dynamic models can be derived such as the Euler-Lagrange equation?
Amazing!!!! Im currently getting into ml, thank you!
thanks for the advice @jordan I am looking at MIT Pro AI course. Where is a good place to learn with people. The hard part is learning on your own
Loved it!
Which library do u recommend? Idk if I should dive into pytorch or tensorflow
Does anybody know who one should follow in economics that is doing applied ML?
I disagree with the approach suggested in this video of learning all the background before you start playing around with actual machine learning projects. I can see how these subjects can be relevant as a PhD student specializing in machine learning, but advising someone interested in machine learning to learn 3 programming languages and differential equations first feels unnecessarily intimidating to me. Many practical problems can be solved with a much less broad understanding, and by playing around with actual machine learning code you can figure out which background knowledge would be worthwhile for you to learn from the kind of problems you run into. Personally I am much more motivated to learn something when I have a clear application that I'm learning something for.
I had a terrible machine learning prof; I also had so many problems myself
Jordan why don’t you try AI to beat RUclips algorithm to promote your content more and get more views?
Please can you provide me with resources where I can learn optimization theory.
Commenting for engagement
Love
some how the volume in your videos is lower than the rest of the ones I am watching. I hope I am not the only one..
Super video! I applauded for $50.00 👏👏👏👏
No mention of statistics (especially during the math section)? i'm just starting to really get into the field, but the two courses i took so far focused heavily on statistics (especially the baesian view of statistics) and got the feeling that this is very much the foundation many algorithms are build on.
Welp I’m 5 months In and thx u
In 2015, I set out to try to use statistical analyses to de-anonymize a twitter sock account that was part of a targeted harassment campaign against a guy in the UK
I kinda wish I'd had access to ML resources, because I was kinda stuck using a set of hand selected metrics for matching sample text to potential posters...
Ultimately the greatest tool, I found, was NOT text but day of the week and time of day (in seconds) .... but if I'd thrown all data I had at a ML algorithm, perhaps I wouldn't have had to do that much thinking :)
I tried, later, to take a job with ML but they didn't want to hire me unless I had a pre-existing experience :( I suppose I'm stuck doing selenium automation til I die lol j/k
you're so cool, man. keep going, do what you love!!
Rip my math is horrible
Ditto 💀
Haha same, I'm going to need to start doing a lot of practice 😅
in 2021 no one needs c for speed