Why do you need an ML strategy?
Hello everyone, welcome to listen to this lesson, how to build your machine learning project, that is, the machine learning strategy. I hope that through this course you will learn how to optimize your machine learning system more quickly and efficiently. So, what is a machine learning strategy?
Let's start with an inspiring example. Suppose you are debugging your cat classifier. After a period of adjustment, your system has reached 90% accuracy, but it is not good enough for your application.
You may have many ideas to improve your system. For example, you may want us to collect more training data. Or you might say that the diversity of your training set is not enough, you should collect more pictures of cats in different poses, or a more diverse set of counterexamples. Or you want to use the gradient descent training algorithm again, training longer. Or you want to try a completely different optimization algorithm, such as the Adam optimization algorithm. Or try to use a larger or smaller neural network. Or you want to try dropout or regularization. Or you want to modify the network architecture, such as modifying the activation function and changing the number of hidden units.
When you try to optimize a deep learning system, you can usually have a lot of ideas to try. The problem is that if you make the wrong choice, you are likely to waste 6 months in the wrong direction. It was only after 6 months that I realized that this method didn't work at all. For example, I have seen some teams spend 6 months collecting more data, but after 6 months, they found that these data hardly improve the performance of their systems. So, assuming that your project does not have 6 months of time to waste, if there is a quick and effective way to judge which ideas are reliable, or even propose new ideas, judge which ones are worth trying, and which ones can be assured Abandoned.
I hope that in this course, you can teach you some strategies and some methods of analyzing machine learning problems that can guide you in the most promising direction. In this course, I will share with you the experience and lessons I learned when building and deploying a large number of deep learning products. I think these contents are unique to this course. For example, many university deep learning courses rarely mention these strategies. In fact, machine learning strategies are also changing in the era of deep learning, because what can be done for deep learning algorithms is now quite different from the previous generation of machine learning algorithms. I hope these strategies can help you improve your efficiency and make your deep learning system more practical.
Course PPT
