As someone who often explains machine learning to non-professionals, I have compiled the following ten points as some explanations of machine learning.
Machine learning isn't like the hype of smallpox: you can solve a myriad of problems by providing the right training data to the right learning algorithm. Call it AI if it helps sell your AI system. But know that AI is just a buzzword, it only represents what people expect of it.
There is a lot of excitement about advances in machine learning algorithms, especially in deep learning. But data is a key factor that makes machine learning possible. Machine learning can be done without complex algorithms, but not without good data.
Machine learning trains models based on patterns in the data, exploring the space of possible models defined by parameters. If the parameter space is too large, it will over-fit the training data and train a model that cannot generalize itself. If you want to explain this in detail, you need to do more math, and you should take this as a guideline to keep your model as simple as possible.
The saying goes that if you input a bunch of junk into a computer, the output is a bunch of junk data. Although the phrase predates machine learning, this is precisely the key limitation of machine learning. Machine learning can only find patterns in training data.
oi As the fund's prospectus warns, past performance does not guarantee future results. Machine learning should also issue a similar warning statement: it can only work on the basis of data that is distributed in the same way as the training data. Therefore, it is necessary to be vigilant about the deviation between the training data and the production data and to repeat the training model frequently to ensure that it is not outdated.
Under the hype of machine learning technology, you might think that machine learning is primarily about choosing and adjusting algorithms. But the reality is simple: most of your time and energy will be spent on data cleaning and characterization engineering, which is converting raw characteristics into characteristics that better represent data signals.
Deep learning has also been hyped up for its application and development in many fields. In addition, deep learning has led to the automation of some of the tasks traditionally performed through feature engineering, especially for image and video data. But deep learning is not a panacea.
Apologies to the NRA, it's the people who are the killers, not the machine learning algorithms. When a machine learning system fails, it's rarely because of a problem with the machine learning algorithm. It's more likely that an artificial error has been introduced into the training data, resulting in biases or other system errors.
In many applications of machine learning, the decisions you make today affect the training data collected tomorrow. Once a machine learning system incorporates bias into its model, it can continue to generate new training data that is bias-enhanced. And, some biases can ruin people's lives.
A considerable number of people seem to have gotten the concept of artificial intelligence from science fiction films. We should be inspired by science fiction, but we can't be so stupid as to mistake fiction for reality. From consciously evil humans to unconsciously biased machine learning models, there are too many realities and dangers to worry about.
The content of machine learning goes far beyond the 10 points I mentioned above. Hopefully these introductory content will be useful for non-professionals.
Translated from the Global AI Big Data Plateau