MACHINE LEARNING ENGINEER
How do you get hired as ML engineer? What are the things to learn. How do you sell yourself by just learning all new frameworks and all on ML but no actual experience on ML.
Get a Master's in ML or a PhD in Science.
In my view, the single most important thing for ML Eng is knowing how to optimize the inference phase of ML. Selling yourself in interviews should highlight some knowledge of ML architectures (where the bottlenecks are for different scenarios), and a lot of knowledge on runtime optimization (how would you alleviate the bottlenecks in a specific language). As a researcher/staff scientist, I expect ML engineers to take the inference portion of my sloppy Python code and reduce the latency of queries from 1s to below 100ms (we serve models to millions of users per second, so usually the target is even lower than that). This might entail refactoring my code, or writing some new stuff in a different language. Ideal candidate would have in depth knowledge of a low level language (C++ or Java), Python, and runtime optimization. Knowing some ML architectures (at the very least; know how the data flows from input to output for conv/recurrent/feed forward neural networks, and approximate nearest neighbors), and an ML framework like tensorflow is a huge bonus.
Side note, if you just want to implement ML algos on new data in Python; apply for data scientist roles. For DS; Avoid larger firms, as the work there is too partitioned (you will only get to do a subset of normal DS role) and the pay is meh compared to startups/medium size companies (you can easily break 200k TC in SF just porting research papers into tensorflow code at medium sized companies under DS title with 0 YOE. I was literally offered 350k TC to do this for a "stealth mode startup" just last week).
Did you took it?
So ML is not about the frameworks. It is very different from other types of software engineering.
I have a few friends who have started out with the Andrew Ng Stanford ML course online. ML is not about frameworks but more about foundational prob and stats skills.
NPL.
For real ML you need strong calculus, strong linear algebra, and strong statistics. The level is likely way beyond what you’re used to. Then Andrew Ng and then a couple of higher level courses. Definitely deep learning and unsupervised learning. This will tell you if you have the math skills or you need to study https://davidrosenberg.github.io/mlcourse/Notes/prereq-questions/math-questions.pdf
Ignore some of these answers - classic gate keeping. You really don’t need to know that much to get started, although it depends on what you want ‘ML engineer’ to mean. Learn the basic ‘standard’ state of the art models at a high level to demonstrate curiosity and the ability to learn (reading papers is a necessity in this field, but they tend to be pretty light on hard math). In CV you should know convolutions, resnet and maybe some object detection models like faster rcnn. NLP is maybe language models, transformer and perhaps some RNN I’m not familiar with. Learn one of the major frameworks and actually use it. Optional because it’s time consuming, but implementing a real model (i.e. not a regression) will probably help give you something to talk about in interviews.
If you're on the job hunt and eyeing ML roles at TikTok, drop me a DM. There's no need to send me any resume, but if you've faced rejections and would like feedback on your resume, I'm here to help with that too. Just upload your resume on Google Drive and share the link. #severance #layoff #hiring #resume #ml
By MA do you mean ML?
My bad. Updated. I meant ML.