The origin of the plan
We are a group of machine learning interest of small partners, for the magical machine learning often have "to explore what the impulse, but because alone to study the lonely, or busy examination little procrastination without sustaining this thirst for knowledge and enthusiasm.
Because feel similar to the case of small partners, we hope to establish a "ml credit program" -- learning machine learning and sharing plan -- to help us with more efficient learning, more concentrated arrangement to share our knowledge and experience. Because we are also convinced that "the best way to prove that you really have a thorough understanding of the knowledge,...
In front of a pick a lecture, see old monk himself face ignorant force, but you think you do a quiet beautiful man (feeling always have the courage to do deep learning of girls is a Chinese paper) can in schools such as Stanford graduate? Break the pattern Tucson, get rid of high learning content gradient, the top universities of assignments and exams will let you (will) fly (BU) MEng (Yu) (Sheng). ...
This is the Stanford CS224d deep learning and natural language processing of the third class, this class first will introduce the monolayer and multilayer neural networks and their in machine learning classification task in the application, then introduced how to use back propagation algorithm to train the neural network model (in the method, we will use partial derivative of the chain rule to layers of update neuron parameters). After giving the neural network and the rigorous mathematical definition of these algorithms, this paper introduces some practical skills and tricks of training neural network. ...
In this class, we will discuss the internal task evaluation and the evaluation method of the external task. The main content is the word word (analogies) technology, we will put it as an internal task evaluation techniques and demonstrate its related examples, it will play an important role in the word vector tuning (tune). We also discuss how to train the weights and parameters of the model, and focus on the word vector used to evaluate the external task. Finally, we will briefly introduce the artificial neural network, which performed well in Natural Language Processing. ...
This article is the content of Standford CS229, is also the background of the CS224d Curriculum Mathematics knowledge. The majority of linear algebra knowledge used in machine learning and deep learning is summarized. In order to facilitate them to leak filled, here to sort out a Chinese version according to English version. ...
In this paper, the Chinese version of the Chinese version of the CS224d curriculum, Stanford University, has been authorized translation and publication of the Stanford University, Professor Socher @Richard...
There are such a class of neural networks, can play a huge role in the NLP, processing from the language model (model language) to bilingual translation, to text generation, and even to the problem of code style imitation. This is the recurrent neural network that we want to introduce today. ...
The purpose of writing this article is to hope that it can be dedicated to engage in data science and machine learning in the learning algorithm of the people on the way to go. I will be in the article for example some machine learning problems, you can also think about the process of solving these problems in the process of inspiration. I will also write some personal understanding of the various machine learning algorithms, and provide R and Python implementation code. After reading this article, readers can at least try to write a machine learning program. ...
The system is based on the CVPR2015 the "deep learning of binary hash codes for fast image retrieval, the massive data of content based image retrieval system, 250W pictures, for a given image, retrieve the top 1000 similar time is about 1s, and the basic background and ignore in mentioned below. ...
The blogger is white chess chess rules, can not remember, there is no design AI chess program. This article is mainly to read "Nature" papers and on the AlphaGo related articles of learning experience.
The main purpose of this paper is to enhance the sharing, exchange of learning, easy for beginners to understand the AlphaGo algorithm, as well as some of the common ideas in machine learning. The real project implementation process is much more complicated than the introduction of this paper.
This paper is more heuristic to be described and analyzed, including some authors combined with their own understanding of the simplified processing. ...
Maybe you and this called "machine learning" the guy is not familiar, but you raise the iPhone camera when has long been used to it helps you to frame the face; also natural open today's headlines to your news; also used Taobao shopping point to find similar goods than three, or loved Microsoft age recognition site results brush burst circle of friends. Well, the core algorithm of these functions is the content of the machine learning field.
Machine learning is the study of how the computer simulates human learning behavior in order to acquire new knowledge or skills, and to re organize the existing knowledge structure to improve itself. While...
To write this blog, my heart is fear. The reason is very simple, as a junior research student is not eligible for computational advertising such a great industry, fields and disciplines to remark upon the. The reason to do so, one is to sum up their own knowledge, the two is to reduce the students' learning costs. My ability is limited, but the lack of practical experience, the content for the impressions of after reading of books and papers, if there are improper or wrong place, also hope everybody that I carefully ask for advice. Here, to the preparation of the "computational advertising" Liu Peng and Wang Chao two teachers, thanks to the paper. ...
We before the NLP series (2) with the naive Bayes for text classification (on) "discussed, naive Bayes limitations stem in the conditional independence assumption, it will text as the bag of words model does not consider the order information between words, will the" Wu Song killing the tiger "and" tiger killed Wu3 song "considered is a meaning. So is there a way to improve the recognition of the word order? Yes, this section is to receive the N-gram language model. ...
Bias method is very powerful, has a solid theoretical foundation. Many senior Natural Language Processing models can also evolve from it. Therefore, learning Bayesian approach is a very good entry point for the research of Natural Language Processing problem. The Bias formula is one line: P (Y|X) =P (X|Y) P (Y) P () (X) P () (Y|X)...
The first two blog post introduced the naive Bayesian this name read "Adorable" but actually simple direct and efficient way, we also introduced some details of the Bayesian approach. According to the old rules, "hoe" to you, teach how to use and the matters needing attention in charge of, but also the way we removal of weed. Well, this section as closer to the actual application will be presented the advantages and disadvantages of Bayesian method, common application scenarios and optimization, and then find a real scene Luan examples practice on the hand, look at how to use tools. ...