원문링크
https://www.datasource.ai/en/data-science-articles/don-t-learn-machine-learning

• 연구할 것 인지 제품을 구현할 것인지 방향을 잘 잡아야 한다는 글
• 아래 핵심구문 인용

How to learn ML by building software

If you’re the type of person who learns best by taking courses, the best resource, and probably the most vocal advocate for this top-down, learn-by-doing approach to ML, is fastai’s Practical Deep Learning For Coders course.

• learn-by-doing approach to ML : 문제해결을 의한 학습방법
  만들면서 막히면 이론을 공부하고, 전부다 갖추고 공부하는 것이 아니라 하면서 머신러닝에 대한 이해를 높여가는 방법을 의미한다.

• Practical Deep Learning For Coders course : 코더를 위한 현실적인 딥러닝
  https://course.fast.ai/

A good approach for getting familiar with what it’s like to build with ML, assuming you learn this way, would be to:

ⓐ Identify a goal, like building a text auto-completer or a license plate identifier.

목표를 잡자 : 번호판 식별같은 것들

ⓑ Find a pre-trained model that fits your project—GPT-2 or YOLOv3 would work for the previously mentioned projects, respectively.

누군가 학습시켜놓은 모델을 찾아서 갖다 써봐라

ⓒ If you’re feeling fancy, you can even use a library like gpt-2-simple to fine tune (customize to your own data) your model.

니 데이터 가지고 튜닝도 해봐라

ⓓ Finally, deploy your model as a microservice.

작은 서비스 형태로 배포해라

Once you can build with machine learning, then learn the theory

If you’re an engineer, you’ve probably implemented some form of authentication before, which means you’ve (hopefully) hashed passwords.

When you set up your password hashing, did you write a custom hashing algorithm? Did you spend weeks studying cryptography? Or did you just use bcrypt?

로그인 서비스를 배포하면서 암호학등 배경지식을 다 알고 니가 운영하는 것이냐?

마무리

Learn-By-Doing

배우면서 따라가봅시다!