Some thoghuts about the AI related approach.

### Doing AI and Using AI is different

I try to divide the research into AMD previously, new appraoch, new method and new data. For the people that do not work on AI approach itsself, the AI or deep learning approach belongs to the dimension of the new method. When I start my undergraduate study, the AI things are popular, the statistical learning theory is popular at that time. and there are all kinds of papers start with the title, … based on SVM approach, these years, there are all kinds of papers such as … based on Deeplearning or neural network approach.

Beforeing diving into the AI approach, I think the important thing is to ask yourself what domain are you working on and what are real challenge here. AI method is not the bullet obviously, only using the AI approach in the right place for the right questions can provides some good benifits.

### When to use AI approach

There is a phrase in Chinese called 锦上添花. I think it is proper for using the AI method, for most of the cases, there are similar solutions without using the AI approach, and the AI approach may bring some benifits here.

Using AI means you want the system to be more smart and use more math model or computation operations here, that is a endless process, maybe that is the one reason that the AI related approach always have a high position in the research field.

Only when you get to the level that your existing approach does not solve your problem in an efficient way, then you may start to use the AI, otherwise, you may not figure out the history of how this problem are solved in history and what are research path for this method, it is dangerous to jump into the AI approach direactly at this time. It means you do not have enough understanding about the existing approach and the abstracted AI method might not be a critical one.

### Abstraction

If we do not work on AI and just use the AI, we usually just call the ai library and fit our data into it instead of developing the ai library.

In that case, figuring out what data is critical in our own domain is more important. More specificly, we need try to abstract out an math problem from the domain specific question. What are the form of the math problem? This video provide a really good explanation. Especially for the section about what are known things are what are unknown things. Highly recommended.

### Reference

A really good reference that can explain clear the thoughts about some AI abstraction, highly recommended, some ideas about the perception, creativity etc.

https://www.ted.com/talks/blaise_aguera_y_arcas_how_computers_are_learning_to_be_creative?language=en&subtitle=en