Chestnuts come from aofei Temple
No. | public QbitAI qubit reports
What is the path for a cute new NLP researcher who wants to cultivate himself into a mature NLP researcher?
A friend from South Korea named Tae-hwan Jung made a complete mind map, covering the knowledge comprehensively from basic concepts to common methods and famous algorithms of NLP.
From zero to one, so to speak, everything you need is here:
This exquisite resource went viral on Reddit over 400 times in less than a day, receiving a flood of praise and thanks:
“Thank you.” “That’s all I need! “Wow, how nice!
So, what does this rich mind map contain?
The four section
Even if you don’t know anything before, you can start with the first section:
1 Probability & statistics
From the gray square in the middle, 5 aspects emanate:
Basic, Sampling, Information Theory, Model, and Bayesian.
In every aspect, there are many knowledge points and methods that you need to master.
After all, with a foundation of probability and statistics, one can enter the second plate with one’s head held high.
2 Machine learning
This section has seven branches:
Linear Regression, Logistic Regression, Regularization, non-probabilistic, Clustering, dimensionality reduction Dimensionality Reduction and Training.
Master the basic knowledge and common methods of machine learning, and then formally to NLP.
3 Text Mining
Text mining is a method to obtain high quality information from text.
There are 6 branches in the diagram:
Basic Procedure, Graph, Document, Word Embedding, Sequential Labeling, And THE NLP Basic Hypothesis.
NLP brings together all kinds of necessary tools on the road.
Natural language processing
Once equipped, it’s time to practice. This is the central idea of the last image:
There are only four branches, but they are rich in content.
The first is Basic, which combs several kinds of networks commonly used in NLP in detail: cyclic model, convolution model and recursive model.
The second is the Language Model, which includes the encoder — decoder Model and Word Representation to Contextual Representation. Many famous models, such as BERT and XLNet, have been fully disassembled here and are something you should work hard to learn.
The third is Distributed Representation, where many commonly used word embedding methods, including GloVe and Word2Vec, become your good friends one by one.
Fourth, Task, machine translation, question answering, reading comprehension, emotion analysis… You are already a qualified NLP researcher, teach the AI to do what you need.
After seeing the brain map, someone asked: is it necessary to implement all kinds of technologies?
The Korean boy said:
No, no, you don’t have to do it all. Find something that feels fun and implement a wave.
One More Thing
On Reddit, many people have expressed their admiration for the brain map and want to know what it’s made of.
Balsamiq Mockups, said the South Korean teenager.
GitHub portal:
graykode/nlp-roadmap
Reddit Portal:
www.reddit.com/r/MachineLe…
– the –
Qubit · QbitAI
վ’ᴗ’ ի Tracks new developments in AI technology and products
Click the “+ attention” in the upper right corner to get the latest information
If you like it, please share or like it