Since the term “Machine learning” was developed in 1959 at IBM, it has become a hotter and hotter topic. Especially recent years, tech companies can actually utilize machine learning technique to bring innovations to their products. Just a few days ago, Google showed the world their ability and virtuosity on machine learning (the audience who watched the keynote might probably hear hundreds of times of “machine learning”). For example, their new phone Pixel 2 can create the shallow-depth effect using only one lens through machine learning (even the newest iPhone X needs two lenses to do that).

 

So, what is machine learning? “Is it that machines can learn by themselves autonomously?” “Sounds so scary and creepy!” Well, machine learning is a field of computer science that deals with the computer ability to undertake tasks without the explicit programming (it doesn’t mean no programming needed). Engineers realized that rather than teaching computers and machines how to do everything, it would be far more efficient to code them to think, and then plug them into the internet to give them access to all of the information in the world. Machine learning is originated from the study of pattern recognition. In fact, it right now is still mostly used for classification problems and regression problems.

 

So, machine learning is not “self-regulated” learning. Rather, it is more like the task performing under human’s supervision. Computers or machines “learn” through training from humans in order to achieve the prediction, which is their “learning goal” (or the goal that humans set for machine to meet). The general workflow of machine learning is shown in Figure 1.

 

The concept of machine learning and is rooted in the understanding of human learning, which is the development and modification of the neural networks. It has been the perfect analogy to understand the information processing in the computer. The critical difference of these two “neural networks” is that computer neural network does not have working memory and therefore there is no cognitive load to limit the storage of information. So, when it comes to the evaluation and optimization process (Figure 1) during machine learning training, the error only come from the inaccurate prediction, instead of dealing with forgetting information.

 

An example of the supervised learning could be classifying merchandises, say, on Amazon. Humans set up the criteria for every category, such as books, home equipment, and clothing, and tell the computers the characteristics of each category through algorithms. Then the computers will be able to put merchandises into corresponding category based on the matching of the item’s characteristics and the category’s. The “learning” part during the process is that how computers recognize the merchandise’s characteristics and pair them with the category, since the computer developers do not and cannot program every trait of every item.

 

 

A more complex machine learning is the so-called “unsupervised learning”. The type of machine learning takes place when humans have no idea about how we should differentiate or classify the data, for instance, Amazon wants to identify the different patterns of customers’ purchasing behaviors in order to provide targeted advertisement and promotions. In this learning process, computer developers still will program the algorithm, but the purpose is more about seeing structures in data and clustering data.

 

Basic Machine Learning Workflow

Credit: University of Michigan “applied to machine learning in python” course

 

 

Machine learning is always talked about with another term, artificial intelligence (AI). It always makes people confused and think they are interchangeable terms. They are not. We might say that machine learning is geared towards problems, for which we have (lots of) data (experience), from which a program can learn and can get better at a task.

Artificial intelligence has many more aspects, where machines do not get better at tasks by learning from data, but may exhibit intelligence through rules, logic or algorithms.

 

AI is a broader concept and is applied when a machine mimics cognitive functions that humans associate with other human minds. Therefore, the machines that can be considered as AI should be able to carry out tasks in a way that we would consider “smart”. If machine learning is like humans answering multiple-choice questions, then AI is like answering open-ended questions. Machines may surprise you by giving you an unexpected answer. When the answer is too good that you as a teacher feels it does better than you do, you start to panic. Some people, like Elon Musk, even predict a doomsday when someday AI outcompetes humans’ intelligence.

 

Do you think it is going to happen?

 

REFERENCES

https://www.coursera.org/learn/python-machine-learning/lecture/hrHXm/key-concepts-in-machine-learning

 

https://www.forbes.com/sites/bernardmarr/2016/12/06/what-is-the-difference-between-artificial-intelligence-and-machine-learning/2/#51ee38d7483d

 

https://www.forbes.com/sites/sap/2017/10/08/artificial-intelligence-and-machine-learning-dont-have-to-be-creepy/#7d3137ed58e0