Andrew Ng came to Beijing to give two lectures this Tuesday, one in Baidu and another in Tsinghua U.. And I was very lucky to go to his lecture in Tsinghua to be acknowledged with his ideas in Machine Learning.
I’ve been focusing on Big Data and Data Mining for several months. From where I stand, I believe that the fundamental elements of Data Mining and Machine Learning is not actually the same. The research on Data Mining is mainly about the design of the algorithm which would best adapt to the real situation, while the Machine Learning ‘trains’ the algorithm better and easy to use. However, they still have many aspects in common and Andrew’s lecture is really worthwhile to attend.
Andrew’s report contains a lots of examples of research on graphics, audio and document. He shows how is computer perception done.
He explains an example of motor cycle recognizing in detail. It shows that to confirm the information of an object by graphical parse we need to extract many Characteristic quantities such as handlebars and wheels here. And through these characteristic quantities we can be acknowledged with some positive and negative relationship which can help us to determine if it is the object we find.
An important concept in machine learning is that the whole process can be divided into different layers, just like the TCP/IP module of the Internet. The lowest deals with the basic objects such as pixels while the higher layer is used to extract the more complicated characteristics.
Another interesting topic is on the natural language processing. It tells how the computer can make up phrases with words and make up sentences with phrases so that the information contained in the sentences can be revealed.
Andrew said that he found it cool to do research in AI as he was a middle school student while he still feels he has not make some breakthrough that. Maybe the knowledgeable professors are always the modest ones.