What is the difference between deep learning and ordinary machine learning

1, ordinary machine learning generally refers to the decision tree, logistic regression, support vector machines, xgboost, etc.

2, deep learning is mainly characterized by the use of deep neural networks: deep convolutional networks, deep recurrent networks, recursive networks, etc.

Difference:

1, algorithmic level there is no similarity, hard to say that the similarity may be It is that everyone's function is to fit a high-dimensional function.

2, ordinary machine learning is better at analyzing the lower-dimensional, highly interpretable tasks. For example, data mining, recommendation algorithms. They are characterized by the general situation of the collection of data dimensions are not high, in order to advertise the push task, for example, the general analysis of data dimensions will only contain gender, age, education, occupation and so on. Interpretability is very strong, and the direction of the tuning reference is more clear.

3, deep learning algorithms are good at analyzing high dimensional data. For example, images, speech, etc. In the case of images, for example, a picture pixel may be tens of millions, equivalent to the feature vector dimension reaches tens of millions, and the relationship between pixel points and pixel points is not particularly obvious. This time with convolutional neural networks can be very effective in dealing with this kind of problem, basically very accurate to capture the features of the image. However, the interpretability of the weights of each dimension is extremely weak, and the direction of the tuning parameter is very unclear (the number of neurons, the number of hidden layers, etc.)

In summary, the two are actually very different. Deep learning has only been developed in recent years. Traditional machine learning algorithms are mostly derived from probability theory, informatics. For programming, traditional machine learning models are basically integrated in the sklearn package, deep learning can use tensorflow as a framework

Want to understand in detail, traditional machine learning can look at Mr. Li Hang's Principles of Statistics or Mr. Zhou Zhihua's Machine Learning (also known as the watermelon book). Deep learning because it is only developed in the past two years, there are very few related books, you can go to check the last two years of deep learning paper

Of course, both need a relatively solid mathematical foundation, mainly these three books: Linear Algebra or Advanced Algebra, Advanced Mathematics or Mathematical Analysis, Probability Theory or Stochastic Processes

Thanks

.