Machinelearning

Content:
This Monday, Dr. Ji introduced us the concept of Machinelearning and several algorithm of classification, which are K-Nearest-Neighbor classifiers, decision tree, and support vector machine. Machinelearning is to state problem of "how to make machine to learn". In my opinion, the process of machinelearning can be devided into two steps: inference and decision. In the step of inference, we need to make machine to build a model to explain or classify the given data, while the step of decision is comparatively easy, which is to give a response to the given data based to the model.

Dr. Ji put great emphysis on the introduction of three classification algorithms. Classification is the process of recognizing different datus.

K-Nearest-Neighbor classifier might be the simplest one. Maybe that's why it is called " lazy learning". And this algorithm is also very easy to understand. It is to classify an object based on its closest training examples. According to my understanding, the core idea of this classifier is that: We imagine that each feature of both traing and testing objects are different "dimension", our objects are in "N-dimensional space". If an object is close to several(N) other training examples in this "N-dimensional space", then this object should be regarded as of the same class of those training examples.

Decision tree is even easier than KNN classifier to understand, especially with a graph given by Dr. Ji. Actually, I use this algorithm in my daily life. In some cases, when I clean up our olding things, I would use this algorithm to decide which things can be threw away while which should be kept. But I think it is much more complex to apply decision tree in machinelearning than to understand, as there would be a lot of specific problems such as at which point should split the tree and which attribute should be the root.

I don't know why support vector machine is called this name. It seems that its algorithm has nothing to do with support vector. The core idea of this algorithm is to maximize the margine. Though the figure given by Dr. Ji is two-dimensional, this "margine" should be the margine in N-dimensional space, and N is the number of features.
Posted by
Jing Lu