How to Deeply Understand the Concept of Machine Learning

This article is mainly for non professional readers. It introduces the concept, connotation and related problems of machine learning. For professionals, you can also use this article to have a deeper understanding of the concept of machine learning and see how to explain your work to your friends.
1. Machine learning means learning from data, while AI is a cool and fashionable word.
Machine learning is based on the assumption that we can train and solve a series of complex problems by putting the right data into the right algorithm. When you need to raise money or release a product, you can call it artificial intelligence (AI) without hesitation, but you need to understand that AI is a buzzword for almost everything.
2. Machine learning includes data and algorithms, but the most important part is data.
Machine learning algorithms, especially deep learning, have achieved great success in recent years, but what you need to understand is that data is the key factor that makes machine learning possible. You can use simple algorithms to implement machine learning, but without good data, you can't do anything.
3. If you don't have a lot of data, you'd better use a simple model.
The task of machine learning is to train a pattern from data and explore the model space defined by parameters. If your parameter space is too large, the model will over fit the training data and make the model lose generalization. There's a lot of math to say about over fitting, but you need to remember that the simpler the model, the better.
4. The ability of machine learning can only reach the level that training data can provide.
"Useless input, useless output" reflects the limitations of machine learning. Machine learning can only find patterns in the training data provided, and can not learn new patterns out of thin air. For supervised learning tasks like classification, you need to be robust in collecting well labeled feature rich data as training data.
5. As long as the training data is representative, machine learning will be effective.
Just as the lesson book once taught us, "past performance is not the guarantee of future results". Machine learning can only be effective on training data and distributed data. You need to be alert to the statistical asymmetry between the training data and the actual data, and you need to keep the model constantly trained to keep it up to date.
6. The most complex task in machine learning comes from data transformation.
When you read the literature, you will see a lot of dazzling algorithms. You may think that the most important work of machine learning is to select algorithms and adjust parameters. But the real situation is: the most important work in machine learning is data cleaning and feature engineering. You need to transform the original features of data into new features that can better represent the information in them.
7. Deep learning is a revolutionary technology, but it is not a panacea for all diseases.
In recent years, deep learning has been hailed as a shrine, far more than other machine learning algorithms. One of the reasons is that deep learning can automatically complete the tasks that need Feature Engineering in traditional machine learning algorithms, especially in image and sound data processing. But we need to understand that deep learning is not a panacea, you can only use this technology in a certain range, at the same time, you also need to spend a lot of energy on data cleaning and transformation.
8. Machine learning is easily affected by misoperation.
"Machine learning algorithms don't kill people, but humans may dig their own graves.". When the machine learning algorithm fails, it is seldom due to the error of the algorithm itself, but in most cases it is caused by human error. In many cases, you accidentally introduce errors in your training data, or introduce bias and other systematic errors. You need to be skeptical at all times to use machine learning algorithms, and strictly check the application process.
9. Machine learning can realize self prediction unconsciously.
In many machine learning applications, today's decision will affect the training data collected in the future. Once the machine algorithm model introduces a certain model deviation, it will continue to collect new data to strengthen the deviation. In fact, some of these deviations can really take people's precious lives. Every machine learning practitioner should bear in mind: do not create self fulfilling prophecies!
10. AI will not have self-consciousness, nor will it rise to destroy human beings.
It is surprising that in today's machine learning is so common, many people still use science fiction and film plots to define and understand AI. It is true that science fiction can inspire creativity, but it should not be so credulous that we misunderstand the real world. Today's world has a lot of dangers that we need to pay attention to, from evil people with ulterior motives to innocent and abused machines. Therefore, please do not worry about the emergence of Skynet and super artificial intelligence. Instead, we should treat machine learning with a prudent attitude, so that it can develop more healthily and serve human beings.