Recent research in machine learning, the learning process will be readily recorded aspects of their own learning and review
1. What is machine learning?
Machine Learning (Machine Learning, ML) What is specializing in computer simulation or realization of human learning behavior to acquire new knowledge or skills, re-organize existing knowledge structures so as to continuously improve their own performance of a science and technology.
It uses computer technology, application of calculus, probability theory, statistics, approximation theory, many different theories and disciplines convex analysis, algorithms, etc., to establish goals for the analysis of targeted data model by past learning from historical data ( classification, regression, clustering, etc.), complete the basic algorithm model, and through continuous follow-up study (data entry), while the messy data conversion (output) into useful information, but also to continuously optimize the transformation itself.
2. Machine learning can solve the problem?
For the study, the human learning process is thinking (process) for the content observed by observing the (input) based on past experience, the process is concluded (output).
The machine learning computer use of existing data (usually massive data), come to some kind of model (algorithm models), and a way to use this model to predict the future.
Mainly to solve the problem of machine learning can be summarized as: optimization, forecasting, correlation. It can be said, as long as it relates to optimization, prediction, personalization related problems, machine learning and deep learning can handle.
For example, we buy passion fruit to eat, we did not buy experience passion fruit, do not know what those are sweet sour, and then we bought a lot back, there are large, small, light, heavy , red, purple, white, yellow, peel smooth, dry and wrinkled ... each one a taste of it, and finally found that there are two types of passion fruit, purple and yellow and white categories, like purple , the deeper peel (the more black purple) is more sweet, more sweet more yellow yellow class, with nothing to do with smoothness irrespective of the size of the peel, and the same size, the more the weight of the juice. Purple fruit sweetness than the class of yellow and white class higher. So the next time to buy, naturally know how to pick and choose.
The machine learning with a similar process to buy fruit, through modeling (judgment of the sweetness of the fruit), design strategy (for passion fruit in different shapes properties) and algorithms (passion fruit taste, sweet and sour draw conclusions), and then enter a lot data (and a lot of different kinds of passion fruit characteristics) for training, and finally learn to judge the process (know what type of passion fruit sweeter). After the model is built, it can test for more data and outputs the result (directly next to buy fruit picked after the fruit type of study can determine directly from the feature of sweet and sour), and these results will continue the correction model (bought and continue to taste, sweet and sour observation degree, continue to sum up experience, the adjustment determination method), to enhance the accuracy of the model, the better to help us forecast data.
At present, widely used in machine learning search engine, junk mail handling, recommended advertising, data mining, image recognition, natural language processing, biometrics, medical diagnostics, securities investment analysis, DNA sequencing, speech and handwriting recognition, such as the use of robots in many areas, and areas of application more widely.
3. Learn machine learning, need to have what capacity?
Started learning machine learning, these only need to have the following capabilities:
- To understand some basic knowledge of mathematics
- Master a programming language (preferably python)
For starters, it is not necessary to have all the mathematical theory foundation to start, we do not have this knowledge does not mean that the individual can not be flexible operating machine learning library, but understand that some algorithm will be more difficult. The basic theory in the subsequent need to slowly fill their relationship to the height of your ceiling in the AI field. And starting from practice, it would be easier to understand the algorithm, in-depth study follow-up algorithms are also a great help.
Advanced and want to become an expert, in addition to the need to strengthen the basis of theoretical mathematics outside (the mathematical basis of the high number of linear algebra, statistics, probability theory, information theory, etc.) to learn, but also need to learn knowledge and skills related to big data (such as: Hadoop, HBase, Spark, kafka, Flume, Sqoop, Storm, etc.). Then according to the direction of development, we have content targeted learning NLP, neural networks, to more specialized areas of in-depth study.
4. how to learn?
Learning machine learning, learning step by step in stages, as not all of a sudden in-depth derivation algorithm (unless you are a very solid mathematical theory), do not try to master all the math before you start to learn, it will be very easy from entry to the collapse, from the collapse to give up.
For starters, it is recommended to learn from several aspects:
- First, understand what is machine learning, machine learning theory and common sense, have a general understanding and knowledge of machine learning.
- Know what learning resources, which is suitable for beginners to understand these resources.
- General understanding of the technology stack machine learning to understand what knowledge involved, then compare with their own technology systems, learn to understand the direction, and make long-term plans with the mentality of learning.
- Then take the time to understand the technical terms commonly used machine learning, probably understand the meaning of these professional terms, easier to understand the learning content when it is convenient follow-up study (do not need to know all, just a rough idea of what is on the line, learning machine learning We will continue to deepen the process of learning and understanding).
- Then go directly to learn, you can learn "real machine learning," this book, you can also learn ApacheCN share out of the text or video tutorial, learn a variety of machine learning algorithms, the first practice and then to the theory.
- Entry after learning based on the use to the basic theory and machine specific algorithms and knowledge, then targeted learning, gradually increase.
References: