We welcome the ascension of our artificial overlords!

Comparison is the eater of joy and salespeople of progression.

You know what data science is since the first Cupoy chapter, but what exactly is the study of artificial intelligence (AI)? Because it, as an academic topic, does not simply cover the construction process of machine learning models.

The scope of AI

AI is the broadest field of study. It includes both learnable models and unlearnable models that refer to programs written by ordinary programmers according to specifications.

Machine learning (ML) consists of models that can learn, usually based on a given input. Machine learning is actually one of the modern technologies to realize artificial intelligence, because in recent years, machine learning has far outperformed other traditional technologies, and it has become a trend.

Deep learning (DL) places more emphasis on this type of model built with neural networks. It is a branch of machine learning.

What is ML?

Machine learning is a subfield of artificial intelligence (AI) concerned with the development of algorithms and statistical models that can learn from data without being explicitly programmed. These algorithms improve their performance on a specific task over time through experience gained from the data.

A primary goal of this subfield is to allow machines to find patterns and trends from data without having to be given rules by humans, and these rules will learn to allow machines to find patterns from data.

Classification and application of machine learning

The methods in which machine learning can lean mainly derives from statistical learning theory, which can be divided into three main types of training methods, each with different applications:

• Supervised Learning (often used in this category): algorithm learns from a labeled dataset. Can be applied to prediction, image classification, fraud detection, and more.

• Unsupervised Learning: identifies inherent patterns or structures within unlabeled data without explicit guidance on the desired outcome. Can be applied to cohorts, exploratory data analysis, data compression, and more.

• Reinforcement Learning: rewards or penalizes agents (AI) when interacting with an environment to educate it on making decisions that maximize its long-term reward. Can be applied to board games, real-time strategy games, and more.

Identifying supervised and unsupervised learning

Supervised learning (SL) models have a pair of x and y data, and there is some correlation between x and y, such as image classification. Each graph (x) has a corresponding label (y). For example, given a picture of a duck and a duck label respectively, so that the model can learn the correspondence between x and y. In the present, most mainstream machine learning applications with high accuracy are mainly taught by SL, but a con of this method is that labeled data must be collected.

Unsupervised learning (UL) models have x and y with no labels, e.g. only image data but no labels. In the clustering model, you would want to divide the given data (x) into groups, which can be divided into smaller groups according to each data's different attributes. UL is generally less accurate than supervised learning, but can be a more efficient choice when data collection is difficult.

Further elaboration on reinforcement learning

You might have seen thumbnails or watched one or two videos of AI training videos on YouTube about teaching machine learning models to control cubes or blocky humanoid models to perform specific goals. Every time the AI fumbles or takes too long to complete each task, the instance resets and they are notified as punished for their undesired performance in the last instance.

This is reinforcement learning (RL). Instead of simply feeding AIs data, learning everything in one fell database, RL learns one piece at a time, starting out inaccurate, but after a period of training, learns how to give a prediction strategy and more correct prediction. RL defines the environment, agent, and reward so that the agent (AI) can learn how to obtain the highest reward through interaction with the environment.

If machine learning is so powerful, is it possible to accomplish any task?