Hi readers!
In this post, we will see the history of DL. This will help us to know about the effort that has gone into developing AI to the level it is today.
AI Today
Almost every company is talking about AI. Google has transformed itself into an “AI first” company. Zuckerberg said that their AI engines had created their “own language”. Elon Musk said “AI is a fundamental risk to the existence of human civilization”. Almost all our smartphones have atleast one AI powered product running on them.
Twilight’s Kristen Stewart has co-authored a paper on Deep Learning.
That’s how popular AI is today. Everyone talks about it. Everyone uses it!!!
But it has taken a lot of research and work to get it to where it is today.
McCulloch, Pitts and Rosenblatt
One of the first milestones in the field of AI was the development of the McCulloch and Pitts Neuron (MPN). It was a simulation of the neurons in the human brain. The MPN has two parts. A function g that aggregates all the inputs. A function f that outputs 1 if the output of g is greater than a threshold and 0 otherwise. The MPN was designed in 1943.
Rosenblatt developed the Perceptron in 1957 which was quite similar to the MPN. But it takes in an additional bias term and can have negative weights.
Image Credits: researchgate.net
Turing Test
Alan Turing devised a test in 1950 to determine whether a computer can think like a human being. It requires three players. One interrogator, one human respondent and one computer respondent. The three of them are isolated and so the interrogator will not know who the computer is and who the human is. His task is to identify the computer. If he cannot identify the computer, then the computer can think like a human.
AI Summer Camp
The Dartmouth Summer Research Project on Artificial Intelligence was held in 1956. This is considered to be the founding event of AI. It was organized by John McCarthy who coined the word AI. The Camp manifesto believed that if a selected group of researchers got together and brainstormed, they will be able to make significant breakthroughs in the field of AI.
AI Winter
This was the period when the interest in AI started to fade away. The first AI Winter was in 1969. This was when Marvin Minsky (known as Father of AI) said there were limitations to what the Perceptron could do. He and Seymor Papert wrote the book titled “Perceptrons” in which they managed to prove theoretically that the perceptron was incapable of learning non-linear functions. For example, it can’t even learn the XOR function.
AI was in trouble again in 1984. This was due to the of the pessimism in the AI community and pessimism in the press which led to a cutback in the funding for AI research.
The Backpropagation Algorithm which was first developed in the 1960’s, was brought back to life in 1989 in a paper titled “Learning by Backpropagating Errors“. But due to the lack of data and computing power, neural networks weren’t able to scale then. The Neural Nets were in for a deep freeze in the 90’s.
AI is back and now it’s DL!!!
The term Deep Learning was coined in 2006. The “neural net” was rebranded as “deep learning“. One of the major breakthroughs was in 2010 when the Imagenet dataset was created by Fei-Fei-Li’s group at Stanford. This led to the Large Scale Visual Recognition Challenge.
Since then, the growth of AI has been exponential. One of the main points to note is that the current growth in the field of AI is mainly due to the increased computer power and data which was not available 10-15 years ago.
So that’s how AI has grown into the superpower it is today.
Before we can get started with Neural Networks, we need to know about the activation functions and error functions that are fundamental for Machine Learning. So we will be covering these topics in the next post.
Tech Wizard 