As I < 2019 summary says mentioned>, I will begin a series of natural language processing (NLP) notes.
Many people say, AI is not said than done, transfer existing library of API and cloud can be friends.
But in fact not the case.
First, AI this field is very large, and from 1950, Turing proposed the Turing test, 1956 Dartmouth conference began, AI has developed fifty years, some academics believe there six times , there It is considered three cases of two down .
So Ai development today, already has a considerable scale, and can not have a person familiar with all areas of AI, which is most familiar with several areas associated, such as NLP and OCR as well as knowledge maps associated with this has been the top day.
So I can not say I'll AI, I can only say that I did Natural Language Processing (NLP) project.
Put another way we know the expression, AI is equivalent to .NET, .NET has a Dragon Ball, across web, desktop, mobile terminal, mobile games end, and so, a person can not be familiar with all areas, to become familiar with web and desktop two the top three areas has been very days.
When used to make a web and desktop programmers do move suddenly end and the upstream end of the hand, because of lack of basic knowledge in the field of correspondence, we can not get started right away, but also pass the interview, which is why so few .NET programmers can be as from the web and the desktop client make a transition in mobile games end. the most typical is a problem, so do a web of .NET programmers do hand travel customers automatically find its way function, is not easy to end, not tone API can be solved .
So although I did NLP projects, let me do it now suddenly Alpha Dog, I am also not acceptable.
I met a lot of people have done NLP project, because only tune existing library and cloud API, the final results are poor, had to fall back to the way a regular expression processing, even almost similar to the " worth a million AI core code . "+ or take artificial intelligent way.
I do not mean this in a variety of methods more ridicule, after all, the first level is not how I do, and the second regular expression + artificial intelligent way indeed temporarily solve part of the problem.
But people always have to pursue, as Eric said, even if do have to do a salted fish salted fish ideals, but in the long run, but still rely on mathematics to solve fundamental problems.
Mathematics and AI as diverse types of complex, the same study mathematics, study calculated two people conformal geometry theory and research areas is very difficult to communicate.
Natural Language Processing (NLP) corresponding to the branch of mathematics, probability theory, while the probability theory will be used in differential and integral calculus, collectively known as calculus.
However, there are many sub-categories of probability theory, there is simply out about probability theory used in NLP knowledge of it:
- Probability (probability)
- Maximum likelihood estimation (maximum likelihood estimation)
- Conditional probability (conditional probability)
- Full probability formula (full probability)
- Bayesian decision theory (Bayesian decision theory)
- Bayes rule (Bayes' theorem)
- Binomial distribution (binomial distribution)
- Expectation (expectation)
- Variance (variance)
However, to understand 9:00 above, we also need to understand the following concepts:
- Permutations
- Frequency and probability
- Classical probability and geometric probability model
- Conditional Probability
- Full probability formula
- One-dimensional and two-dimensional discrete random variable
- One-dimensional and two-dimensional continuous random variables
- Covariance and correlation coefficient
- Law of Large Numbers and the Central Limit Theorem
- Sample and sampling distribution
- Point estimates
I have a specific concept behind each writing experience.