Neural Network vs New Born Baby
Having a 1 year old at home, I have never related to Neural Networks more than now. If you wonder how I can relate both, this blog will walk you through it. This is a fun read :)
Every day in the past 1 year, I have seen that my baby keeps learning something new. Initially he didn’t know how to drink milk, then he learnt to drink milk. Similarly he learnt to crawl with his tummy, sit by himself, crawl in 4 legs, stand up by holding something, walk a few steps, stand up by himself, pick food and eat, drink using a cup and the list goes on. Now he is even able to identify his favorite foods- banana (yummy!) and blueberry (Classification Problem). How has all this been happening? It is because his brain is learning new things everyday. Everything in this world was new to him when he was born. But today, he is used to so many things and so many people. He recognizes mom, dad, grand parents, close relatives and friends. All this is because, his brain is constantly reacting to anything and everything he sees and slowly learns them.
Being a data scientist by profession, I have constantly dealt with data, machine learning models, statistics, deep learning, neural networks and so on. But, after my baby was born I am totally able to relate to all the different methodologies we use in neural networks to train a model. For example, when there are different classes in a data set and we want to reduce the False Positives and False Negatives, we generally penalize the model. If a class A is mis-classified as class B and the vice versa, then we penalize the model heavily so that the model learns not to do the mistake again (Deep Reinforcement Learning). Similarly, we can also reward a model if it classifies the classes correctly.
This is exactly the same situation I face with my baby every day. When my baby boy eats vegetables, he keeps throwing the vegetables from his high chair. If I say a soft ‘no’ he stares at me and tries to repeat the same thing again (penalize less). But, if I say a strong ‘no’ then he understands it is really a mistake (penalize more) and tries not to do it at least for the next few minutes (haha!!). And if I keep telling him ‘no’ repeatedly, he tries to understand that he shouldn’t do it. At the same time when we saw our baby turn on to his tummy for the first time, my husband and me were so excited, that we unknowingly tried to encourage him to turn on to his tummy when he was trying to do so (rewarding). By this way he tried to turn faster on to his tummy. We did the same thing when he picked his 1st blueberry with his fingers and when he took his first few steps. This has made him to do the good things more and more. Now, he eats most of the food by himself and tries to drink mostly from a cup.
Similarly, my little one knows how a banana, beans, blueberry, mutter, milk, water and a few other things looks like. When I initially offered him coconut water, he thought I am giving him water, because of the same texture and color. So, he refused drinking it (sadly he doesn’t like water at all!). So I tried to put a small drop into his mouth and because of the taste, he learnt this is something different and loved the coconut water. This is nothing but re-training a model with new but similar data (Transfer Learning). You either retrain your neural network completely or train only the last few layers to make your model classify correctly.
For anyone who is so confused about how to deal with a neural network and how to make it learn, just think about how you teach your baby between good and bad, rewards and punishments. You can very easily relate between the two and this will help you understand better. After all, a neural network is nothing but training neurons like in the brain. A child’s brain is the best learner- relate the two and create wonders!
Thanks for reading through :) Hope you enjoyed it. Leave your comments below or contact me via LinkedIn
