The rise of AI has been largely driven by one tool in AI called machine learning. The most commonly used type of machine learning is a type of AI that learns A-to-B or input-to-output mappings.
This is called Supervised Learning.
Let's discuss some examples of supervised learning.
1) Spam Filter: Spam filters are designed to identify emails that attackers or marketers use to send unwanted or dangerous content. It needs some knowledge of Natural Language Processing(NLP).
Read the article: "How To Design A Spam Filtering System with Machine Learning Algorithm"
Input: Email
Output: If it is spam then output is 1 or else output is 0.
2) Speech Recognition: Speech recognition is a machine's ability to listen to spoken words and identify them. E.g. Google Home, Amazon Echo, Apple Siri. It also needs some knowledge of NLP.
Read the article: "Audio Deep Learning Made Simple: Automatic Speech Recognition (ASR), How it Works"
Input: Audio Clip
Output: Text Transcript
3) Machine Translation: Machine translation is the process of using artificial intelligence to automatically translate text from one language to another without human involvement.
Read the article: "Language Translation with RNNs"
Input: Text in English Language
Output: Text in any different language like Spanish, Chinese etc
4) Online Advertising: It is the most lucrative form of supervised learning. It helps target ads to specific audiences based on their behavior, interests or demographics. AI system predicts whether users will click on the ad or not.
Read the article: "Machine Learning in Advertising – Predictive Targeting and Moderation"
Input: Some information about the user and some information about the ad
Output: If the user clicks then the output is 1 or else the output is 0.
This type of application has a huge impact on the economy.
5) Self-Driving Car: Driverless cars can identify objects, interpret situations, and make decisions based on object detection and object classification algorithms.
Read the article: "How Machine Learning in Automotive Makes Self-Driving Cars a Reality"
Input: Images captured or some information from the radar or the sensors
Output: Position of the cars and obstacles
This can help to avoid other cars and obstacles while driving.
6) Visual Inspection: Visual inspection using machine learning is a process that uses AI to analyze image and video data to automatically verify product quality. It can help manufacturers detect product defects, improve quality control.
Read the article: "What is AI Visual Inspection for Defect Detection?: A Deep Dive"
Input: Image of a product e.g. cell phone
Output: If the device has some scratch or dent then the output is 1 or 0.
The idea of supervised learning has been through many decades but it has taken off recently. Consider the below graph:
On the horizontal axis, we plot the amount of data we have for the task. So for the speech recognition system, it can be the audio data. Over the last few decades, the amount of data we have access to has grown due to the rise of the computers and rise of the internet.
On the vertical axis, we plot the performance of the AI system. If we use a traditional AI system, then the performance of that system is given by the red curve. Initially, the performance of the system increases, but as we feed more and more data the performance remains constant and does not improve further. So the speech recognition system won't be accurate in giving the exact transcripts or the online advertisement system won't be very accurate in showing the most relevant ads though we feed on more data.
However, AI has recently taken off due to the rise of neural networks and deep learning. If we train a small neural network then the performance is given in blue curve. The performance is better than the traditional AI system. If we increase the size of the neural network then the performance of the AI system becomes better and better with an increase in the amount of the data.
At any particular point performance is given as follows:
For complicated applications like self-driving cars or speech recognition products where having a highly accurate performance is important, we need a much larger neural network to make speech recognition products much more acceptable to users and much more valuable for companies.
Consider the following graph:
For the best possible levels of performance, we need 2 things:
Large neural network
Huge amount of data
Due to the requirement of huge amounts of data for high performance, we hear the term Big Data.
Big Data: The definition of big data is data that contains greater variety, arriving in increasing volumes and with more velocity.
Another thing is that we must be able to train large neural networks. So the rise of fast computers and specialized processors including graphical processing units (GPUs) has enabled many companies to train large neural nets on a large enough amount of data to get very good performance and drive business value.
Hence, the most important idea in AI is machine learning(ML) and for any AI system to have high performance needs a large neural network and a huge amount of data.
Credit for the above blog: I have gained the knowledge for the above blog from a course 'AI for everyone' by Andrew Ng. Thank you Deeplearning.ai for such a wonderful course.