Back in the early 2000s, when I was a young kid whose main hobby was reading the newspaper, I read a series of the documentary about the battles between Garry Kasparov, who was the strongest chess grandmaster from 1986 until his retirement in 2005, and IBM supercomputers. At first, the supercomputer loss all the matches but after each time the engineer at IBM improve and update new versions, the supercomputer had the ability to tie and then win over Kasparov. I was so surprised and fascinating about those matches and I was asked myself how a machine could play chess, learn, predict all the moves and win chess matches over a human being. After ten years, the term “machine learning” comes to me naturally and help me have a deeper understanding of how machine learning works. Therefore, in this article, I am going to give a brief introduction to machine learning, a subset of artificial intelligent (AI), to see if it can become a killer robot just like in the movie “I, Robot” or become a robot in love in “Bicentennial Man”.
Machine Learning: A brief introduction
Machine learning is a scientific method that software engineers and scientists try to make computer system works and finishes tasks without receiving explicit instruction or assistance from the human being. The term “machine learning” was coined by Arthur Samuel in 1959, who is an artificial intelligence pioneer.
In general, software engineers and scientist create and apply mathematic algorithms to the machine so that it can learn and act automatically based on the algorithms and the action in the past. So in general, machine learning is about learning to do better in the future based on what was experienced in the past.
For example, when you buy or find a table in Amazon, the Amazon computer system will save the result. The next time you go to Amazon, all Amazon recommendation product will be related to that product. For this case, Amazon will recommend many types of the table or just simply table cloth.
Another example is the text feature of Apple IOS. When you type “How are you?” for three or four times, the system will store that result and put in the prediction text. The next time when you type “How”, the prediction will show “are you” on the keyboard. That can help you save time when texting.
What is the difference between Artificial Intelligence (AI), Machine Learning, and Deep Learning?
Artificial Intelligence (AI) and machine learning are related to each other. AI is the creation of intelligent machines that can simulate, act, react and have the problem-solving skill just like a human. There are two types of AI: narrow (weak) and general (strong).
Narrow AI is the AI that is designed for a specific task or aspect and can implement the task very well. The supercomputer DeapBlue from IBM is the example of narrow AI, which designed to learn, analyze and predict chess moves. Another example of narrow AI is Apple assistant Siri which was designed to recognize the voice and execute the action based on the voice command.
General AI is the super intelligence AI that can perform multitasks as well as or better than human. For example, Jarvis in Iron Man is a general AI. However, in real life, general AI does not exist. Most of the AI systems, that we have, are narrow because we still don't have enough ability to create such a powerful AI. We are still in the narrow AI period, but with the rapid development of the technology, I believe that we will have general AI in the near future.
Machine learning, as we mentioned above, is the subset of AI. This is the core step before creating an AI. The data scientists and software engineer have to train the machine learning by applying many algorithms and data so that the machine can receive the information, learn from it, and improve its performance every time it runs.
Deep learning is a subset of machine learning. Just like humans learn from their experience, deep learning is when a machine receives a large amount of data and learn from it by using algorithms and neural networks. Because of its high accuracy, deep learning allows human to have the ability to reach the thing that we cannot do before.
- Neural network: is an information processing model that learns from the data to observe and recognize patterns by using hidden layers of network structure connected by nodes. The neural network takes data from the input and sends it to hidden layers to classify and grouping unlabeled data based on the similarities among the data.
- For example, the neural network is applied to the ATM. When you deposit money, the bank computer system will transfer your money to your bank account based on the account number.
- Deep learning is also called a deep neural network because the deep learning has a large number of hidden layers in the neural network. The normal neural network has only 2 or 3 hidden layers but for deep learning, the number of hidden layers can be over 100 and more.
How does Machine Learning work?
The way machine learning works is based on the data, the algorithms, and the aspect that the developers want to approach.
For example, a machine is created to find which fish human can eat and the machine has a ton of fish image data. If somebody uploads a salmon picture, the machine can recognize the fish is salmon based on multiples salmon image in its data sources. This is called supervised machine learning.
Another example is that somebody uploads an image of a mushroom to check if that is a poison mushroom. The machine doesn’t have any information about that mushroom but it has an algorithm that said if the mushroom is colorful, it will be poisonous. This is called unsupervised machine learning.
There is a situation that if the machine gives the result of unpoisonous mushroom is poisonous. The machine fails, but it will take this result as an experience and learn it, so it can give the right output in the next time. This is called reinforced learning.
Type of Machine Learning.
There are three main types of machine learning: supervised, unsupervised and reinforcement.
Supervised
Basically, supervised learning is learning with the teachers. The teachers, in this case, are the data scientists. They teach the machine by making it compare the output with the data and the algorithms that they give to the machine. By using the given data and algorithms, the machine can predict the result. The more data the machine has, the more accurate the result is. Here are some types of supervised learning.
Classification: The machine classifies the inputs into different categories.
- The categories can be whether a car is “blue” or “red”.
- The categories can classify if a person is “tall” or “short”.
Regression: The machine classifies the inputs with real value such as “price”, “weight”, or “height”.
- The machine can predict the price of a stock or bond.
- The price prediction of the lastest car.
Unsupervised learning
Unsupervised learning is basically learning without teachers. In this type of machine learning, the machine has to predict the output of unlabeled data based on the algorithms not on the data source. It learns from the observation and finds the structure of data. The algorithms will look for the patterns and the logic of the inputs to categorize them into different groups.
Unsupervised learning is usually used for analyzing big data such as transaction data at the bank. Here are some types of unsupervised learning:
Clustering: is when you want to split the input into different groups based on the similarity among them.
- Use clustering when you want to group consumers that have the same purchasing behavior.
Association: is the type that when you want to discover those inputs which appear together in the dataset.
- Use association to see what people tend to buy a product and another product together at the same time.
Reinforcement Learning
Reinforcement learning is also known as semi-supervised machine learning. Most of the machine learning system is in this type. Reinforcement learning is the act of training the machine with unlabeled data and learned from that data with the grading output. The machine learns from its experience by creating a feedback loop to improve it every time it performs.
The most basic example of this type of machine learning is solving a maze. Reinforcement learning will analyze many choices of a maze based on its failure on each choice to find the ultimate result.
The Implement of algorithms in Machine Learning
We train machine by teaching them the most suitable algorithms so the machine can perform the task perfectly. Therefore, choosing algorithms is very important in machine learning. There are so many algorithms, but in this article, I’m gonna introduce three basic algorithms: linear regression, K-Means Clustering, and decision trees.
Linear Regression
Linear regression is supervised learning in machine learning. It finds the relationship between two continuous data and to analyze if those data have a deterministic relationship. For example, we can relate to the relationship between height and weight. Based on that relationship, we can forecast the next results. We can use this algorithm to predict future sales or revenue or the stock price on the market.
K-Means Clustering
This algorithm is unsupervised learning. K-Means Clustering uses the similarities among the input data to put them into may discrete groups. Because unsupervised learning does not have any labeled data so after each execution, the groups’ border will be clearly visible and easy to distinguish.
Decision Trees
This is the most popular algorithms and is widely used in machine learning. Decision trees are supervised learnings algorithms. It basically can use for classification and regression. We use it by choosing a variable attribute on the top, ask a question to see the true statement to choose the next attribute below. In the example above, we have a situation to decide whether I am hungry or not so I can choose the next move.
Human Bias
The benefit of Machine Learning
There is a lot of benefit of machine learning in real life. Here is some typical example.
- Apply machine learning in selling helps companies such as Amazon, BestBuy, or any selling company optimize the cost and increase marketing opportunity to consumers.
- Using machine learning in the financial sector can help the industry increase the control of trading stock, predicting the flow of money or decrease fraud detection.
- Help increase the efficiency in the manufacturing industry and predict the management plan to prevent failure.
- It can also help doctors enhance the possibility to find and give the right treatment to patients.
- Increasing the efficiency of traffic to reduce traffic jam.
Future of machine learning and Conclusion
We are still at the early age of machine learning, so there are a lot of things that we need to research and develop. If we have the right vision and proper technology, machine learning will become the most important tool in our life.
Thank you for spending time reading my article. If you are interested in machine learning you can try Python, JavaScript or C++ to learn deeper about it.
Sources
https://www.cs.princeton.edu/courses/archive/spr08/cos511/scribe_notes/0204.pdf
