Neural network and deep learning Chapter 1: introduction reading questions

1. Is neural network equivalent to deep learning ？

Not equivalent , Deep learning can use neural network model , Other models can also be used （ Such as deep belief network ）

2. What is the main difference between shallow learning and deep learning ？

An important feature of shallow learning is that it does not involve feature learning , Its features are mainly extracted by manual experience or feature transformation methods , The deep learning model can automatically extract features

3. What is learning ？ What is a good expression ？ What are the representations ？

In order to improve the accuracy of machine learning system , We need to transform the input information into effective features （ such as word embedding）, Or more generally, it's called expressing （Representation）． If there is an algorithm that can automatically learn effective features , And improve the performance of the final machine learning model , Then this kind of learning can be called representation learning

The key to representation learning is to solve the semantic gap （Semantic Gap） problem ． Semantic gap refers to the inconsistency and difference between the underlying characteristics of input data and high-level semantic information . for instance , Computers cannot understand texts directly , The representation of text is only a low-level code in the computer （ Such as ASCII code ）, So we need some conversion methods to make the computer “ understand ” semantics .

（1） A good presentation should have strong presentation ability , That is, vectors of the same size can represent more information ．
（2） A good presentation should make the follow-up learning task simple , That is, it needs to contain higher-level semantic information ．
（3） A good representation should be general , It's task or domain independent ． Although the current large

Part of it means that the learning method is still based on a certain task , But we expect that the representations they learned can be easily migrated to other tasks ． In machine learning , We often use two ways to express features ： Part means （Local Representation） And distributed presentation （Distributed Representation）．
Part means ： Such as one-hot code
Distributed representation ： Such as RGB Show color ,word embedding

4. What is deep learning ？

“ depth ” It refers to the number of nonlinear transformations of the original data （ The layer number ）, Deep learning is a sub problem of machine learning , Its main purpose is to learn effective feature representation from data automatically , The key problem to be solved in deep learning is ： Contribution distribution problem , That is, the contribution or influence of different components or their parameters in a system to the final system output