ML, DL, AI, Sounds confusing…?

Original article was published on Deep Learning on Medium

ML, DL, AI, Sounds confusing…?

Every Data Science or AI enthusiast at the beginning of his career would have been bewildered about the dissimilarity between the terms Artificial Intelligence, Machine Learning, and Deep Learning. Do they mean the same? No, but they agree on the same revolution, evolving computers into human brains.

Artificial Intelligence is the branch of Computer Science aiming at making machines performing actions/tasks exactly the same way a human is supposed to. On a boarder perspective, given a scenario which involves a human to think, considering various factors and make a decision is bought about by machines using the genre Artificial Intelligence. Making it simpler with an example, imagine yourself driving a car at on the main streets at a higher speed, you come across the traffic signal gleaming red, you stop the vehicle and wait for the signal to turn green. Likewise, you tend to stop the car upon seeing a dog or any animal including the human race across the road, these are more of human thoughts than rational ones. It becomes necessary for the machines to think and act both rationally and humanly. This feature of machines is brought about by AI.

Yeah, so make machines behave like humans all good, but how?

Machine learning and deep learning to the rescue. To make it simpler, we use machine learning and deep learning to attain AI. They are considered to be subsets of Artificial intelligence. Machine learning is defined as the ability of a machine to learn and train itself automatically based on a pattern or experience derived from data without being explicitly programmed. Making it understandable via layman terms would be

You’re planning to go to watch a movie with your friends. Which movie to choose would be of great confusion out of the 5 blockbusters running. How to choose a movie by humans –> asking other friends who have watched the movie already and discussing, IMDB rating, your favorite actor, director in the movie and so on. If you ask a machine to decide which is the recommended movie, it would judge the movie based on the same factors IMDB rating, responses from people, twitter and arrive at a conclusion. Training the machine with a pre-dataset available and allowing it to predict the viewer’s choice is machine learning.

The difference between the human race and machines would be humans making decisions with their brains while machines use the data available to make a good decision.

Folks, we might be clear with AI and Machine Learning. But why do we have deep learning popping in while we have ML?

Machine learning only works in such cases where the test data is similar to that of the data it has been trained with. Confusing? sure it is! Deep learning uses artificial neural networks which imitates the neural networks of the human, hence the name. Machine learning requires human intervention, we edify the algorithm, preside over its implementation. The place of origin of any learning algorithm is its data. Though, data can never be identical. With a larger dataset and on not having a “label” to build a classifier, DL uses neural networks that make decisions on the features and classifies without it being aware of the labels.

Do we humans make any decision with a pre-programmed mind? Humans are for a reason with the sixth sense that we can decide for ourselves as to what we introspect from a situation. ANN programs itself with the data rather than humans providing it with a program to follow.

Hey Alexa, “Lights”, “ON” — Using the mentioned terms, Alexa will “turn on” the smart bulb.

However, if we were to ask “Hey Alexa, it’s dim..“, “It’s dark here inside the room“, “The light is very pale“, “It is gloomy in here“, “My room is just pitch black“, can it make a decision to turn on the lights if we were to implement machine learning. If we were to ask the same to a human, he would immediately turn on the lights realizing that it’s dark. Neural networks are provided with these data of ‘dim, dark, pitch’ and it profounds to learn from that without having a counterpart output. For the same, DL requires a large set of data to learn the variances of how humans speak just for turning on a light which brings me to unveiling the biggest requirement of DL, a massive dataset. Like just how a man is responsible for his own decisions, a neural network’s decisiveness is contingent on itself.

Hey, I have a dataset of features, with its label, would like to build a classifier to differentiate between cats and dogs.

AIBot: Sure man, use supervised learning and classify them. It’s as simple as that because we have machine learning in the world doing wonders!

Hello, I have a dataset of features only, I do not have the result or label for the same, I still would like to build a classifier. Do we an alternative method?

AIbot: Yep, do you need a label to differentiate between cats and dogs? You can do it with your brain. Just by ANN scrutinizing the dataset, it extracts features and classifies it. Be careful, it trains itself, might be wrong at unavoidable circumstances.

Though neural networks perform better than machine learning, it’s important to pick the best one! With a larger dataset and diverse features, DL is the pre-eminent choice. While with a smaller and labeled dataset, ML would be the choice of AIBot.