Ask HN: In 2023 which is the best path to get good at machine and deep learning?

[newsoul]

Which is the best resource (book, public course, blogs, etc) to get started in machine and deep learning and then get good at it both as a practitioner and from theoretical understanding?

The ultimate goal is to become a good at implementing models and come up with new ones.

Is there something like teachyourselfCS but for Data Science, ML and DL?

[GoldenMonkey]

Caltech machine learning intro course: https://www.youtube.com/watch?v=mbyG85GZ0PI

karpathy's Zero to Hero series (https://www.youtube.com/playlist?list=PLAqhIrjkxbuWI23v9cThs...)

meta llama 2 - is open source https://github.com/facebookresearch/llama/tree/main

Tools:

ai - hosting Banana - Machine Learning Model Deployment on Serverless GPUs https://www.banana.dev

pinecone - vector database: https://www.pinecone.io

how to run AI language models on a single cpu pc - https://news.ycombinator.com/item?id=34869960

[zackproser → GoldenMonkey]

Thanks for the mention: we've got tons of free content at https://www.pinecone.io/learn

and a lot of open-source Notebooks at github.com/pinecone-io/examples which demonstrate techniques from semantic search to question and answering to retrieval augmented generation (RAG)

FWIW, our free tier is sufficient to use any of these notebooks or our open source vercel templates - so whether you're a student or just a solo dev looking to learn, you can create a free account and get started and never pay a dime - the one limitation is you can only have one index live at a time - so delete it once you're finished experimenting with a given notebook.

[max_]

For the basics read Micheal's Neural Nets & Deep learning - http://neuralnetworksanddeeplearning.com/

The Watch the Caltech telecourse - https://work.caltech.edu/telecourse

Read tutorials on Pytorch, Tensorflow & Keras.

Read, source codes on hugging face and deploys, test, train toy models.

Test your skills by participating in Data scientist competitions like Kaggle or Numerai.

It will give you a great way of guaging your competence with other data scientists.

[brudgers]

Starting is binary not continuous.

Starting is the best way to get started.

Stasis cannot be motion optimized. Motivation is the hardest part. Everything else is about equally difficult because all the rest is experience. Good luckz.

[bequanna → brudgers]

It can feel very overwhelming when you recognize how much you don’t know about a topic.

For some people, having some suggested learning tracks can help them have more confidence that they are moving in the right direction and not just chasing their tails.

[brudgers → bequanna]

Learning anything encompasses the possibility of failing.

The possibility can’t be optimized away except by not starting.

If you aren’t doing something badly and inefficiently, you aren’t learning.

The first problem isn’t finding the best path (per the question). The first problem is to stop standing still.

Searching for the best path is only pretending to learn.

[bequanna → brudgers]

I don’t disagree with you. Perfect is the enemy of good and action > inaction.

I’m simply saying if you come across someone who is lost and without a map, it is helpful to at least give some basic direction.

[brudgers → bequanna]

I understand what you are saying.

I think compasses have more utility than opinions drawn from different terrain.

[tikkun]

I collected some resources on this. See: https://news.ycombinator.com/item?id=36195527

[pizza]

the light way: fast.ai

the heavy way: kevin murphy's a probabilistic approach to machine learning. you could make use of this book basically every day.

[ggr2342]

Start with fast.ai courses for learning Deep Learning at the practitioner level.

[YossarianFrPrez]

Once you've learned what you can from online resources and textbooks, doing projects -- from Kaggle, etc. -- is a good way to practice applying what you've learned.