Please accept this root directory on ARTIFICIAL intelligence – Blog Collation series (part 1)

Those Things about Data Science — Blog Collation Series (Part 2)

Machine Learning Handbook — Blog Organizing Series (3)

Expand the horizon here — Blog Finishing series (four)

Deep Learning Essentials (PART 1) — Blog Organizing Series (Part 5)

Deep Learning Essentials (Part 2) — Blog Organizing Series (Part 6)



Machine learning is a sub-field of computer science. In the field of artificial intelligence, machine learning has gradually developed into the study of pattern recognition and computational science theory. Since 2016, machine learning has reached its unreasonably hot peak. However, effective machine learning is difficult, because machine learning itself is an interdisciplinary subject, without scientific methods and certain accumulation is difficult to get started.

If you want to learn machine learning or are currently learning machine learning, this manual will definitely help you get to your “peak of life”. The manual includes how to get started with machine learning, popular algorithms for machine learning, actual machine learning and so on.

one Introduction to Machine Learning:

1. Save your Tracks: Here’s a short guide to starting machine learning

Abstract: This article shares a simple experience about the development of machine learning, the purpose is to provide beginners with basic guidance, mainly explained the establishment of the system, choose the appropriate evaluation indicators, data processing, system optimization and other content, to help beginners to take some detours.

2. Introduction to Machine learning “Secrets”

Abstract: Machine learning has become one of the most popular technologies, for beginners, how to quickly start machine learning is very important. This article belongs to the entry level treasure Canon, master please detour!

3. Knowing how to play Super Mario Bros., how hard can machine learning be?

Abstract: Can xiaobai understand machine learning? This article uses The principles of Super Mario to teach you what machine learning is, and to make cutting-edge technology less difficult to understand.

What can machine learning do for your business? Some things you would never guess! (Introduction to Machine Learning part I)

Abstract: machine learning is an incredible technology, you need to understand the basic knowledge of many, many, to make a business function as much as possible is not influenced by the complex algorithm, so you can ask the right questions, understand the development process of machine learning model and set up a team to promote the continued cooperation between disciplines, rather than the data science is seen as a miracle of the black boxes.

5. What you need to know about Machine Learning Algorithms (Introduction to Machine Learning 2)

Abstract: The classification of learning algorithms is based on the data needed to build the model: whether the data needs to include inputs and outputs or only inputs, how many data points are needed, and when to collect the data. According to the above classification principles, it can be divided into four main categories: supervised learning, unsupervised learning, semi-supervised learning and reinforcement learning.

6. How to develop machine learning models? (Introduction to Machine Learning 3)

Abstract: Building a good machine learning model is the same as building any other product: start from the idea, and consider the problem to be solved together with some potential solutions. Once you have a clear direction, you can prototype the solution and then test it to see if it meets the requirements, as this article does step by step.

7. How to Run machine Learning Teams efficiently (Introduction to Machine Learning part 4)

Abstract: A “traditional” product team consists of designers, engineers, and product managers. Data analysts are sometimes included, but most often multiple teams share this scarce resource. This article will reveal the roles and organizational structure of the machine learning team.

8. What user experience problems will machine learning produce? (Introduction to Machine Learning chapter 5)

Abstract: Many machine learning algorithms are black boxes: take a lot of data and get a model that works in some mysterious way. This makes it difficult to explain the results of machine learning to users. In many algorithms, there are also interaction effects that make the model more difficult to interpret. You can think of this as a compound effect of features, which are combined in many strange, complex and incomprehensible ways, and the whole is greater than the parts.

9. Simple Self-learning Machine Learning Theory — Introduction (Part I)

Abstract: this article is “machine learning theory” trilogy in the first part, mainly introduces the basic theory of machine learning motivation and learning, the detail of machine learning problem, the generalization error, and learning problems could you solution formulation, said as the preliminary research of machine learning personnel introduces the basic process of machine learning.

10. Simple Self-learning Machine Learning Theory — Generalization Boundaries (Part II)

Abstract: This paper is the second part of the “machine learning theory” trilogy. It mainly introduces the basic mathematical knowledge such as independent isodistribution, law of large numbers and Hoeffding inequality, and deduces the generalization bounds and its decomposition in detail.

11. Simple Self-learning Machine Learning Theory — Trade-offs between Regularization and Bias Variance (Part III)

Abstract: This article is the third part of the “machine learning theory” trilogy. It mainly introduces the variance decomposition and regularization of the objective function. It can be seen from the simulation that the introduction of regularization term limits the scope of solutions to learning problems.

12. Entry-level Walkthroughs: Machine learning vs. deep learning

Abstract: This paper introduces the definition and application of machine learning and deep learning in simple and easy to understand language, as well as the differences between source data requirements, hardware support, feature engineering, problem solving, execution time and interpretability, etc., which has great enlightenment significance for beginners.

13. Enhance learning? This article introduces you to reinforcement learning

Abstract: The entry of a new thing will always be some do not know how to start, read this article hope to give you some help and understanding.

14. Become an expert in machine learning with these 10 great books from the beginning

Abstract: Machine learning is an interdisciplinary subject, and has a great role in practical applications, but no book can make you an expert in machine learning. For this article, I’ve selected 10 books that are different in style, subject matter, and date of publication. Therefore, whether you are a beginner or a domain expert, you will find something suitable for you.

15. Want to know how machine learning is getting on? Here is a self-test (with answers and explanations)

Abstract: The pursuit of automation and intelligence has been promoting the progress of technology, and machine learning technology has played a huge role in various fields. Over time we’ll see machine learning everywhere from mobile personal assistants to recommendation systems on e-commerce sites. Even as a layman you can’t ignore the impact of machine learning on your life. This test is for those who have some knowledge of machine learning.

16. Machine Learning Ebook — (TensorFlow)RNN Introduction

Abstract: The author of this article is writing his new book Machine Learning with TensorFlow. This blog is only a small part of his new book. The author introduces RNN in simple language, and introduces how to use the built-in RNN model in TensorFlow for prediction without a small example.

8 fun machine learning projects for getting started

Abstract: Are you still feeling helpless when you can’t find the machine learning entry training program? In this guide, we will bring you eight fun and easy-to-learn machine learning projects for beginners, which will increase your confidence in learning machine learning.

18. Machine Learning Quickstart: Three algorithms you Need to know

Abstract: Machine learning dominates the headlines every day. If you want to get started, you need to understand these three algorithms first.

two Machine Learning Algorithms

1. Quickly choose the right machine learning algorithm

Abstract: Machine learning beginners can learn how to quickly find the appropriate machine learning algorithm through this article.

2. Multiple Perspectives: How does Bayesian reasoning work

Abstract: This paper first introduces the origin of Bayes, and uses simple examples to vividly explain how Bayes’ theorem works, explains its basic principle and the physical meaning of the formula.

3. Simple and Easy to Understand: A simple example perfectly explains Naive Bayes classifier

Abstract: There are many explanations for Naive Bayes classifiers, but not many are based on a simple example. This article explains Naive Bayes classifiers based on a simple and easy to understand example.

4. “Learning” and “Learning” in one: Supervised Learning — Introduction to Support Vector Machine (SVM)

Abstract: SVM is a kind of supervised learning in machine learning, which is usually used for pattern recognition, classification, and regression analysis. This article uses a small example to introduce SVM, concise and comprehensive, easy to understand.

5. Machine learning tools — Decision trees and random forests

Abstract: Machine learning is the most popular field at present. This paper introduces its core algorithms: decision tree and random forest through a small example.

6. Graph Based Machine Algorithm (I)

Abstract: Graph-based machine algorithm learning is a powerful tool. Combined with module characteristics, it can play a greater role in collection detection.

7. Graph Based Machine Algorithm (II)

Abstract: Graph-based machine algorithm learning is a powerful tool. Combined with module characteristics, it can play a greater role in collection detection. This article is the second in a series on graph-based machine algorithms.

8. Easy to learn! Machine Learning Algorithms — Markov Chain Monte Carlo (MCMC)

Abstract: For simple distribution, many programming languages can achieve. But for a complicated distribution, it’s not easy to sample directly. Markov chain Monte Carlo algorithm solves the problem that simple sampling algorithm cannot be used for sampling, and it is a very practical sampling algorithm. This article will briefly and clearly explain the Markov chain Monte Carlo algorithm, take you to understand it.

9. Advanced implicit matrix factorization — How to achieve a faster algorithm

Abstract: This paper focuses on the Conjugate Gradient (Conjugate Gradient) method to discuss a better matrix decomposition algorithm.

10. Pure dry | the classification of the gradient descent method in machine learning and contrast analysis (with source)

Abstract: This paper introduces different types of gradient descent methods based on the amount of data used, time complexity and algorithm accuracy, and explains the comparison of three gradient descent methods in detail.

Dropout for Deep learning Networks (I) — Deep Analytical Dropout

Abstract: This paper introduces the idea of dropout skills in deep learning in detail, and analyzes the two versions of Dropout and Inverted dropout. Moreover, it is refreshing to associate single neurons with Bernoulli random variables.

Dropout For Deep Learning Networks (II) : Discards learning as Integrated learning

Abstract: This paper analyzes that discarding learning can be regarded as integrated learning. In ensemble learning, a network can be divided into several sub-networks and each sub-network can be trained separately. After training and learning, the output of each sub-network is averaged to obtain the integrated output. In addition, it shows that discarding learning can be regarded as the integrated learning performance of different hidden node sets in each iteration, and also shows that discarding learning has the same effect as L2 regularization.

13. Sigmoid VS Sofmax activation functions

Abstract: This paper introduces two kinds of activation functions in neural network, softmax and Sigmoid function, briefly introduces their basic principle, properties and use, and uses Python for example demonstration, at the end of the paper summarizes the differences between the two activation functions.

14. Novel training method: Train neural network with iterative projection algorithm

Abstract: This paper introduces a training method of neural network using iterative projection algorithm. Firstly, the basic knowledge of alternating projection is introduced, which shows that projection method is an effective method to find solutions to non-convex optimization problems. Then it introduces the basic knowledge of difference graph and combines difference graph with some other algorithms to make difference mapping algorithm converge to a good solution. Finally, the iterative projection algorithm is applied to the neural network training, and the experimental results show that the results are good.

15. Vehicle tracking algorithm big PK: SVM+HOG vs. YOLO

Abstract: In this paper, SVM+HOG algorithm and YOLO algorithm are used to compare the accuracy and speed of vehicle detection and tracking, and the conclusion that YOLO algorithm is more advantageous is drawn.

16. What is video vectorization? This article takes you through video recommendation based on DeepWalk

Abstract: In this paper, video vectorization is briefly described, and the DeepWalk algorithm is simply explained.

17. More Advanced Dimensionality Reduction than PCA — (R/Python) T-SNE Clustering Algorithm practical Guide

Abstract: This paper introduces t-SNE clustering algorithm and analyzes its basic principle. The accuracy of t-SNE algorithm is compared with PCA and other dimensionality reduction algorithms, and the results show that t-SNE algorithm is superior. Finally, examples of R and Python implementation and common problems are given. T-sne algorithm has great research prospects in natural speech processing, image processing and other fields.

Random forest VS gradient elevator — My view on model fusion

Abstract: This article is adapted from the Quora community “When Would one use Random Forests over Gradient Winmachines (GBMs)?” In response, several bloggers discussed the appropriate scenarios for Random Forests and Gradient Elevator Machines (GBMs), as well as their advantages and disadvantages.

3. Machine learning library:

LightGBM VS XgBoost: Who is the strongest gradient library?

Abstract: Many people compare XGBoost to dragon slaying sword, LightGBM to heaven sword, so when heaven meets dragon slaying, who is stronger?

2. Learning and Using: Pandas Introduction and Time Series analysis

This post is a presentation by Alexander Hendorf at PyData Florence 2017. The first half of the report introduces the basic functions of Pandas for beginners, such as data input/output, visualization, aggregation, selection and access. The second half of the report describes how to use Pandas for time series analysis.

3. Yandex, Russia’s largest search engine, has opened source CatBoost, a gradient enhancement machine learning library

Abstract: Russian search giant Yandex has announced that it will submit CatBoost, a gradient-elevating machine learning library, to the open source community. It can “teach” machine learning with sparse data. CatBoost can also operate on transactional or historical data, especially in the absence of sensory data such as video, text, and images.

4.Net Flix open-source Vectorflow, a lightweight neural network library for sparse data optimization

At Netflix, our machine learning scientists tackle a variety of problems in a number of different areas: from customizing your TV and movie recommendations to optimizing coding algorithms. A small part of our problem involves dealing with extremely sparse data; The total number of dimensions of the problem at hand can easily run into tens of millions of features, even though there may be only a few non-zero items to look at at a time.

5.Python High performance computing library — Numba

Abstract: In the era of computing power is king, libraries with high performance computing are widely used to deal with big data. For example, Numpy. This article introduces a new Python library, Numba, that performs better than Numpy in terms of computational performance.

6. Second most popular language: From beginner to Master, Python data Analysis Library complete

This article introduces some common Python libraries for data analysis tasks, such as Numpy, Pandas, Matplotlib, Scikit-learn, and BeautifulSoup. These libraries are powerful and easy to use. With this help, data analysis becomes much simpler.

7. New tool — TensorLayer: Managing the complexity of deep learning projects

Abstract: This article introduces a new Python library based on TensorFlow, TensorLayer, which can effectively help developers manage their deep learning networks. It also provides a lot of powerful apis to help developers get things done.

8.Pandas are not perfect

We use Pandas all the time, but we don’t know the details about Pandas. Pandas is an in-depth review of the top 10 key issues to address by explaining how to use Apache Arrow to solve them.

9. Seven things you probably didn’t know about Numba

Abstract: Numba is becoming a popular way to speed up Python programs. This article explains seven things you may not know about it.

Four. Machine Learning:

1.57 lines of license plate recognition code worth $80 million

To prevent stolen vehicles from being sold on the black market, police use a web-based service called VicRoads, which checks the registration status of vehicles. The department has also invested in a stationary license plate scanner: a fixed tripod camera that scans passing cars and automatically identifies stolen vehicles.

2. How to use machine learning to predict housing prices?

Abstract: The author uses what he has learned in the past three months to predict house prices in his city. Techniques used include network crawling, text natural language processing, deep learning models on images, and gradient enhancement.

3. Technical debt in machine learning

Abstract: Many people frown when they encounter technical debt, but generally speaking, technical debt is not a bad thing. For example, technical debt is a reasonable tool to use when we need to meet a release deadline. But technical debt has the same problem as financial debt: when it comes time to pay it back, we pay more than we did in the first place. This is because technical debt has a compounding effect.

4.DIY image compression — K-means clustering image compression: Color quantization

Abstract: Taking image compression as an example, this paper introduces one of the practical applications of machine learning.

5. How can machine learning be used for rule-based validation

Abstract: This article introduces some advanced questions, such as: How much of the verification of intelligent autonomous systems can be achieved by machine learning? Are most requirements still rule-based, and if so, how do they integrate with machine learning? How do unstable interfaces between machine learning and rules affect machine learning-based systems?

6.Certigrad — Random computing graph optimization system

Abstract: Certigrad is a proof-of-concept, which is a new approach for developing machine learning systems.

7. Play Flappy Bird with neural networks and genetic algorithms

Abstract: This paper presents a machine learning algorithm designed for Flappy Bird. The goal of this experiment is to use neural network and genetic algorithm to write an artificial intelligence game controller, hit the highest score of the game, not to challenge!

8. Teaching Machines to Write Code: Enhanced Topological Evolutionary Network (NEAT)

Abstract: NEAT stands for “enhanced topological Evolutionary network”. It describes the algorithm concept of self-learning machines inspired by genetic modification during evolution. Let’s see how it teaches machines to write code.

9. In machine learning, use SciKit-learn to handle text data simply

Abstract: In machine learning, we always first process the source data into the form that conforms to the input of the model algorithm, such as the text, sound, image into a matrix. Text data is tokenized first, stop words are removed, words are converted into matrix form, and then input into machine learning model. This process is called feature extraction or vectorization.

Five. Machine Learning:

1. 10 Facts you need to know about Machine learning

Abstract: The author explains common misconceptions about artificial intelligence from the perspective of non-specialists.

2. Who wins? — Random search V.S. grid search

Abstract: Random method and grid method are both common and effective structural optimization methods. So which of the two is better? In this article, the author finds the answer through an interesting terrain search experiment.

3. Without any formula – intuitive understanding of variational automatic encoders VAE

Abstract: This paper briefly introduces the basic principles of VAE for variational automatic encoders, from bayesian calculation probability method of classical neural network to optimization problem for VAE in variational automatic encoder neural network, using KL divergence to measure error, provide a VAE basic framework. There is no formula, easy to understand.

4. Enhance the design of obstacle avoidance system: from the perspective of machine learning, analyze the design ideas of learning obstacle avoidance car

Abstract: FF91 was successfully launched in Las Vegas on January 4, 2017, which opened the prelude of Internet ecological electric vehicles. Automatic parking makes parking a pleasure, and novice drivers no longer have to worry about how to fit in. But will driverless cars work well in busy environments? This article will stand in the Angle of machine learning and share the design idea of learning obstacle avoidance car.

5. After AlphaGo becomes the best in the go world, how should we carry out machine learning?

Abstract: Machine learning is undoubtedly a hot topic of science and technology. Unmanned driving, machine chess, stock market prediction and other fields, we can find machine learning busy and tall figure. So for beginners, how do you start? How do you learn?

6. Share machine learning project presented by Andrew Ng at deep Learning Summer School

Abstract: Deep learning project process, take you out of confusion.

7. Comparison of distributed machine learning platforms

Abstract: Machine learning, especially deep learning (DL), has recently achieved success in speech recognition, image recognition, natural language processing, recommendation/search engines, etc. These technologies have very promising applications in autonomous vehicles, digital health systems, CRM, advertising, Internet of Things and so on. Of course, money is driving these technologies forward at breakneck speed, and recently we’ve seen a lot of machine learning platforms being built.

8. The love-hate relationship between machine learning and statistics can end

Abstract: Machine learning and statistics have been in love for years in the field of data science. Today, let’s team up with both ML practitioners and statisticians to untangle their decades-long love-hate relationship.

9. Summary of unsupervised feature learning research results in the first half of 2017

Abstract: Unsupervised learning is the core technology of the era of artificial intelligence. Today, we will take stock of the important research results of unsupervised learning in the first half of 2017.

Supervised similarity learning: symmetric relation learning based on similar problem data

Abstract: In this paper, symmetric relation learning based on similar problem data is briefly introduced. By applying twin convolutional neural networks on Quora dataset and StackExchange corpus, the results show that symmetric networks can greatly improve the detection accuracy.

11. Applied Machine Learning: A Guide to Evangelism

Abstract: The authors of this article have compiled a knowledge collection of concepts, definitions, resources, and tools that will be useful to anyone working in this complex field.

This series of blog content is recommended by beiyou @love Coco – Love life teacher, translated by @Ali Yunqi community organization, in charge of the translation team Yuan Hu. This series of long-term updates, more high-quality articles on machine learning, a lot of recommendations!