[Machine Learning] 머신러닝에서 키포인트

Key Lessons for ML [Domingos, 2012]
● Learning = Representation + Evaluation + Optimization
● It’s generalization that counts: generalize beyond training examples
● Data alone is not enough: “no free lunch” theorem--No learner can beat random guessing over all possible functions to be learned
● Intuition fails in high dimensions: “curse of dimensionality”
● More data beats a cleverer algorithm: Google showed that after providing 300M images for DL image recognition, no flattening of the learning curve was observed.

 

 

 

차례대로 해석하자면

1. 학습 = 평가, 최적화, 표현

Data visualization is the graphical representation of information and data. 를 근거로 해석했다.

 

2. 일반화가 중요하다. (학습을 넘어서 패턴을 찾는 것)

 

3.데이터 자체만으로는 충분치 않다. 

 

4. 직감은 고차원에서 실패한다.

(앞서 머신러닝의 개념에서 고차원일수록 실패한다고 언급했다)

 

++ 아래 글 읽으면 도움 될 듯 

Intuition in data science is not about using your gut feel. In data science, intuition is the intuitive understanding of concepts, in other words, how to apply the concepts.

 

A lot of people make the mistake thinking that to be a successful data scientist I have to learn all the math. Yet the only thing that is important to be successful is knowing how to use the concepts.

 

Many concepts in data science are very technical, driven by very complex mathematics and statistics. Remember, you only need to understand how the concept is used, without necessarily going into the mathematics.

https://www.superdatascience.com/podcast/sds-102-intuition-vs-mathematics-data-science

 

 

5. 데이터는 다다익선(多多益善) 

(많은 학습으로 더욱 정답에 가까운 패턴을 만들 수 있다.)

 

 

++ 데이터 융합과 머신러닝의 시너지를 생각해본다