By now we’ve collected a fair bit of knowledge of what AI is and how AI works, but this is not enough for understanding what the world is doing out there in the field of AI. In this chapter, we’ll learn about the different terms that you probably will encounter once you start to read more about AI.
AI is a diverse concept that can’t be explained using just 2 words that are “Artificial Intelligence”. Many umbrella terms come under Artificial Intelligence which have specific features different from each other. Let us see what those terms are.
You’ll often see people throwing around words like “Machine Learning”, “AI”, and “Deep Learning” as if it’s all the same. The aim of this chapter is to make you understand the subtle differences present between the different umbrella terms of Artificial Intelligence.
Machine Learning (ML) is a branch of Artificial Intelligence (AI) wherein the machine learns from the training data given to it and increases its algorithm accuracy with the increase in its experience (looking multiple times at the training data).
To lay it in even simpler terms, machine learning is a type of artificial intelligence wherein the model becomes more accurate at predicting the output without explicitly being programmed to do so.
There is a chance that this might have gone past your head. Just hang on with me and keep on reading further. You’ll get the hang of it soon enough.
According to a blog post by UC Berkeley (one of the world’s leading institutions in the field of Artificial Intelligence), the whole ML training process is divided into 3 parts.
Image the following example before we see those processes and how they come into play in the following scenario.
Suppose you have been assigned a task to create an ML model that can predict the price of a house based on its different features (number of bathrooms, number of bedrooms, size of the kitchen, location in the city, etc.). You provide your model with some training data (in this case, a database or spreadsheet that contains all the features of different houses along with their selling prices) to train on and improve its accuracy. Because the model knows the input (features of the house) and output (selling price of the house) during the training process, therefore it is going through supervised learning.
Decision Process
Based on the training data, which in this case is labeled, the model will try to find unique patterns between the different features of the houses and the selling prices of those houses. This process of finding patterns and relations within the training data is known as the decision process.
Error Function Process
Based on the patterns found by the model, it will predict an output (selling price of the house). This predicted output is compared to the actual output, also known as ground truth, to evaluate the accuracy of the model. This process of comparing the predicted output of the model to the actual output and assessing its accuracy is known as the error function process. The closer the model predicts the output to the ground truth, the less the error comes out to be, and the farther the model predicts the output from the ground truth, the higher the error comes out to be. For example, if the actual output, or ground truth, is 10 and the model predicts the output to be 9.87 then the error is low. On the other hand, if the model predicts the output to be 18.9, then the error is high. The model always aims for the least amount of error possible.
Model Optimization Process
Now, to increase the accuracy (or decrease the error) with which the model is predicting the selling price, different settings or parameters of the model are tweaked by the model automatically to attain better accuracy than the last time. This process where the model automatically changes its parameters and settings to increase its accuracy is known as the model optimization process. There is no human intervention needed in this whole process. All the settings and parameters that are modified by the model are done on the basis of the different mathematical functions.
This 3-step cycle is repeated by the model as many times as specified by the developer and is responsible for the working of the whole field of machine learning.
Phew, that was a lot! Kudos to you if you were able to make sense of the above example. If not, don’t worry, it’s totally expected. Just go through the example once again. You can even go through this online resource that I found: https://ischoolonline.berkeley.edu/blog/what-is-machine-learning/
#Note: we’ll be using the terms “Artificial Neural Networks” and “Neural Networks” interchangeably, so don’t get confused!
Artificial Neural Networks are a subset of machine learning and are an essential part of deep learning algorithms (no need to worry, you’ll study these algorithms later in this chapter). The name —Artificial Neural Networks— is inspired by the fact that the way these networks work resembles the way neurons present in the human brain function.
Essentially, Artificial Neural Networks are computer systems that use a process similar to the functioning of the brain to develop algorithms that can be used to identify complex patterns or predict output using the training data.