Ask HN: How to get good as a self taught ML engineer?

[y3sar]

I am a self taught programmer and ML engineer. At work I find myself usually downloading models from huggingaface and sometimes training them to get the job done. That is enough to impress people in my country. But I know I'm not a good ML engineer. I want to get good, real good, Andrej Karpathy good dare I say, How do I do it? I have done the Andrew Ng course and the Fast.ai course. Those got me into the field, but what do I do to thrive in this field. I really want to do some great work and help people. Great people from HN please tell me how can I level up as an ML engineer.

Thank you in advance

[tsingy]

Start with this https://mml-book.github.io/

Then this https://probml.github.io/pml-book/book1.html

[y3sar]

Thank you for these my friend

[brucethemoose2]

> I really want to do some great work and help people.

Have you looked into ML compilation?

https://github.com/merrymercy/awesome-tensor-compilers

IMO there is low hanging fruit in the space between high performance ML compilers/runtimes and the actual projects people use. If you practice porting projects you use to these frameworks, that would give you a massive performance edge.

[y3sar]

So porting architectures for these compilation libraries. I was already doing this for tinygrad. Was trying to port a vq vae model. But I guess tiny grad is not ML compilation. Thank you for this.

[tikkun]

What does thriving in this field look like to you? Getting a job at a certain company, getting a certain type of job, making a side project that reaches a certain level of success, making a startup that reaches a certain level, just having fun and making things you love, making things that a handful of users love, something else

[y3sar]

By thriving what I meant was be a really skillful person in the field. But to answer your question definitely building things a handful of people love is my goal. Can you tell me how to improve ML skills?

[tikkun]

> But to answer your question definitely building things a handful of people love is my goal.

If that's your goal, then you might be better served by finding a group of users that you care about and can reach, understanding them deeply, and then only learning what you need to better solve the problems that really matter to them. You might end up needing to learn ML skills, you might end up needing to learn design, sales, copywriting, other domain-specific skills, and so on.