Embedded Files
Agnij Moitra
Agnij Moitra
Class 9 - 2021-22
Elements Of AI
Agnij Moitra
Agnij Moitra
In the 2nd decade of the 21st century where an app made using AI was searched 42,094,748 times on Google and ranked as the 12th most searched thing on Google[1] won't you want to know about AI?
Reinforcement Learning:
Reinforcement Learning:
So let's start with the most natural way we learn. Remember how you learned to walk. You just practiced, stumbled, until you finally were able to walk properly.
Similarly, imagine we have AI trying to explore a place and get to the star
in the below image. It randomly makes choices and finally gets to the star. Where every time it makes the right decision we give it a candy for example and if it makes a wrong decision we take away a candy[2].
But as you can see, there are many other faster ways to get to the same place. To handle this issue we let the AI "Explore And Exploit"[3]
where it can use the previous data and with some randomness
take some other decision to find the best result. As here finally, after many random choices, the AI found the optimal path.
The above way of training the AI is called Reinforcement Learning.
Supervised Learning:
Supervised Learning:
The internet is full of data i.e. more than 1,200 Petabytes of data[4]. Can it be used to do some AI stuff? Yes, it can!
For example, we want to differentiate between cold drinks and lemonades. Where we first feed in many images from a dataset and make features of cold drinks and lemonades. Like: Cold drink is blackish and lemonade is yellowish.
Further, once we have given the AI to learn from, we show a new image that it has never seen before and ask. Cold Drink? Or Lemonade?
As this one is more yellowish, our AI with a certain probability, say 80 - 95%, will predict it to be lemonade.
This method where we first give a large amount of data and then ask the AI to do something based on the data that we have given in Supervised Learning
Unsupervised Learning
Unsupervised Learning
Imagine you went to a garden and took some pictures of sunflowers and roses. And you want the AI to find out if it's sunflower or rose, and you don’t already have a dataset containing many images of sunflowers and roses. We can make the AI find it as follows. Based on features the petal radius and the redness of the flower.
As in the above image, we can see many dots, each representing a flower, and a very special thing in this graph is that there are two distinct groups of dots. The one with a greater radius and less is sunflower (top left), and the one with greater intensity of red color and less radius is rose (bottom right).
Recall that as we didn’t classify any data given, the AI found out the features in the image, by which it was able to classify the images without any supervision, this is known as Unsupervised Learning.
References and Footnotes:
References and Footnotes:
1) Soulo, Tim, and Ahref. Top Google Searches (2021). Ahrof, 2021, https://ahrefs.com/blog/top-google-searches/.
2) Garivier Aurélien, et al. Explore First, Exploit Next: The True Shape of Regret in Bandit Problems. 3rd Edition ed., arXiv.org, 2018. arXiv:1602.07182v3 [math.ST]
3) Just for the sake of simplicity, I said candy, in action it can vary from a binary signal to a multidimensional vector.
4) Starry. “How big is the internet?” How Big Is The Internet? Hint: Probably A Lot Bigger Than You Think, Starry, 29th July 2019, https://starry.com/blog/inside-the-internet/how-big-is-the-internet/. Accessed 14th June 2021.
Page updated
Google Sites
Report abuse