Preface

Why machine learning

I have learnt programming for about two years. The most common used language is Python. I have also tried other languages like HTML or C, but Python is easy for beginner to access and approach to natural language. There are plenty of friends around me learning programming. Unlike them who usually start from designing video games or little application, I’d like to try something more interesting and challenging.

I took the python course on Codeacdemy for a month. I approached most of the basic sentence and rules. I got a chance to go to Harvard taking the class Artificial Intelligence and neuroscience I learnt the application of machine learning in the area of medicine. In next several articles, I will briefly conclude what I have learnt and some basic application.

What is machine learning

The concept of machine learning has existed for decades. Since last century, engineers try to teach the machine to learn things as human. The neuroscience happened to develop rapidly in the same period which bring essential inspiration and support. In order to build a machine which can imitate the way brain functions, scientists try to mimic the way neurons calculate and connect with each other. Various algorithms are invented in a short period.

Supervised learning

Supervised learning is the most popular and easiest one in machine learning. When train a model, people need to feed the machine by tons of data which has already been labeled. In this way, the machine can remember the pattern of each data and generate a model by adjusting parameters. Simple supervised learning models are regression and classification.

Unsupervised learning

Unlike supervised learning, unsupervised learning is more like human learning in my opinion. When using the supervised learning, people need to feed the model by labeled data which could limit it in some level. However, when training the unsupervised learning model, people are not able to tell the machine what the input is. In contrast, the machine should learn the feature of inputs by themselves just like how children do.

Therefore, unsupervised learning is usually used to do classification work. Like one new-published paper has introduced, current machine learning needs more rethinking. Do we teach the machine to recognize the same images as one certain classification or we need to tell it what this image really is. For instance, when people label the cows as dogs, does the machine can still recognize the picture as cow, or it can tell this is exactly a cow?

What does machine learning need?

• a considerable big data which are already labeled (the energy of machine learning)
• a brilliant performed computer (better with GPU)