本文源自YouTube网站《Deep Learning SIMPLIFIED》系列视频，文章内容为视频的字幕，供深度学习初学者参考学习。
If you want your computer to recognize very complex patterns - then trust me on this - you really need to start using neural networks.
When the patterns get really complex, neural nets start to outperform all of their competition. Plus, GPUs can train them faster than ever before! Let’s take a look. Neural nets truly have the potential to revolutionize the field of Artificial Intelligence. We all know that computers are very good with repetitive calculations and detailed instructions, but they’re historically been bad at recognizing patterns. Thanks to deep learning, this is all about to change.
If you only need to analyze simple patterns, a basic classification tool like an SVM or Logistic Regression is typically good enough. But when your data has 10s of different inputs or more, neural nets start to win out over the other methods. Still, as the patterns get even more complex, neural networks with a small number of layers can become unusable. The reason is that the number of nodes required in each layer grows exponentially with the number of possible patterns in the data. Eventually training becomes way too expensive and the accuracy starts to suffer. So for an intricate pattern - like an image of a human face, for example - basic classification engines and shallow neural nets simply aren’t good enough - the only practical choice is a deep net. Have you ever run into a wall when trying to work with highly complex data? Please comment and let me know your thoughts.
But what enables a deep net to recognize these complex patterns? The key is that deep nets are able to break the complex patterns down into a series of simpler patterns.
For example, let’s say that a net had to decide whether or not an image contained a human face. A deep net would first use edges to detect different parts of the face - the lips, nose, eyes, ears, and so on - and would then combine the results together to form the whole face. This important feature - using simpler patterns as building blocks to detect complex patterns - is what gives you deep nets their strength. The accuracy of these nets has become very impressive - in fact, a deep net from google recently beat a human at a pattern recognition challenge.
It’s not surprising that deep nets were inspired by the structure of our own human brains. Even in the early days of neural networks, researchers wanted to link a large number of perceptrons together in a layered web - an idea which helped improve their accuracy. It is believed that our brains have a very deep architecture and that we decipher patterns just like a deep net - we detect complex patterns by first detecting, and combining the simple one.
There is one downside to all of this - deep nets take much longer to train. The good news is that recent advances in computing have really reduced the amount of time it takes to properly train a net. High performance GPUs can finish training a complex net in under a week, when fast CPUs may have taken weeks or even months.
Before we talk more about the various Deep Learning models, we’re going to briefly discuss which types of deep nets are suitable for different machine learning tasks. That’s coming in the next video.