Unsupervised Learning… Deep Learning !

I like to listen to songs of Ed Sheeran.. I like his voice, his writings, his voice modulation and his throw of voice. Being a fan, I love to collect his songs, via downloads from digital media and also from online shops, I tend to buy few stuffs related to his albums, and advertisements. As against that, my daughter likes to listen to Shawn Mendes. She is a huge fan of him and therefore, she tries to prove the point by sharing his songs. And I do that too.

Now welcome to Digital World of Artificial Intelligence (AI). This world is watching both me and my daughter. The data that we collect, share and access, gives AI the understanding and it will start streaming and promoting all the updates of both Ed Sheeran and Shawn Mendes. We will get it on our social media websites as well as on our phones, as advertisements and so on. Now comes the next stage.. I am listening to Shawn, just because my daughter is sharing his songs… not because I like his songs, specifically. And same is true for my daughter. Is the machine going to know that I am actually a fan of Ed Sheeran and Not of Shawn Mendes? A conventional AI machine will not know. But, if the machine has learnt the Deep-learning way, YES, it WILL KNOW..! This is one level of learning more, which is done by an AI machine, called as Deep-learning.

As mentioned in the online literature, Deep learning is a subset of machine learning in AI, that has networks capable of learning unsupervised, from data that is unstructured or unlabeled. This is also known as deep neural learning or deep neural network. Deep learning is an artificial intelligence function that imitates the workings of the human brain in processing data and creating patterns for use in decision making.

So how does this work? Well, as we saw in previous Blog (Blog 4: How machines learn?), the huge data collected from different sources of social media, of a person, is analysed in quick time to judge his/her priorities. Now there are some algorithms which work on the data and lead to decision making. This is exactly the way in which they learn that I love music and in that I like Ed Sheeran. But normal AI will also see that I listen to Shawn. However, a Deep-learnt computer system will easily understand that I am not visiting social media sites of Shawn, nor am I downloading anything from the authentic sites of the singer. This means that I am listening to a derived source, indicating my lesser interest..! Thereby, judgment of liking is derived.

The structured algorithms are NOT given to the computer which is deep-learning. They are unsupervised. This is just like a small child, not given any lessons of school education (which is highly structured). But the child is given everything that he is supposed to know. The child figures out the way to acquire the information and create knowledge from it.. Sounds familiar…? In the same way, computers are fed with the Big Data. But processing is not “supervised”. They find out their own methods using “Neural networks” which they build themselves, using the volumes of data they get. If this data shows that I am processing more information of Ed Sheeran, it will learn that he is my favourite. Neural networks are constantly built and broken and re-formed. Just like humans evolve, these Artificially intelligent machine (Robots?) also evolve. They also become “wiser”, “smarter” and their judgement improves…. this is Deep Learning.

Technically, I do not understand much… but the machines are configured to handle data (big data) in layers (Hierarchical neural networks). The lowermost layer takes all the data. The upper layers keep adding information about the addresses, locations, personal information (which is authentically online) and so on. Hence the data is given some weighted points. This forms quadratic equations, mathematically (this is one of the models). This equation keeps modifying as the machine learns with more and more data. Again the same example… If the machine finds that today I heard 2 songs of Shawn, and only one song of Ed, it will give more weight to Shawn in the quadratic equation..!! and slowly it will do the reverse, as it collects more information about me.. This way, Mathematics keeps evolving and the machine keeps learning. I will definitely see the advertisements and new releases of Ed Sheeran and not of Shawn, in this new Digital era…. unless, ofcourse, my loyalty changes…. But trust me, the computer will know that too… and it will flip too..!!!

Now, if this is the way in which a computer learns us in such a “minutest” of details, it can become a better buddy? It can create wit and humour of my taste? It will recommend me the food of my liking (and mood of that time)? Not only that, it will order it for me? And play the music and lights of my desire? Well, the answer is yes, again..! Well, then will I not enjoy the company of such devices? An intense thought indeed… we should be careful. Where are we in this technological upheaval? Not Far…. Heard of Sophia or Zeno??