Deep Learning 101 — Role of Deep Learning in Artificial Intelligence

Since the last decade or so, the developments in information technology have been propelled by advancements in areas of Artificial intelligence and Machine learning. Recently, there is a healthy debate going on regarding potential advantages and disadvantages of same between two powerhouses — Elon Musk of Tesla and Mark Zuckerberg. While the media is jumping on the bandwagon, it is important to understand some basic concepts of AI, ML and Deep Learning to get a better sense of What they do and How they can be useful.

  • Artificial intelligence (AI) is an area of computer science that emphasizes the creation of intelligent machines that work and react like human.
  • Machine learning (ML) is an approach to achieve Artificial Intelligence. ML approach provides computers with the ability to learn without being explicitly programmed. At its most basic is the practice of using algorithms to parse data, learn from it, and then make prediction about something in the world
  • Deep learning (DL) is a technique for implementing Machine Learning. DL is the application of artificial neural networks to learning tasks that contain more than one hidden layer.
  • Artificial neural networks are computing systems inspired by the biological neural networks that constitute animal brains. An artificial neural network is an interconnected group of nodes, akin to the vast network of neurons in a brain. Here, each circular node represents an artificial neuron and an arrow represents a connection from the output of one neuron to the input of another.

Refer to the picture below to get a better sense of co-relation between AI, ML and Deep Learning and how do Artificial Neural Networks work.

How does Deep Learning work?

While deep learning itself is a concept, neural networks are a model for deep learning. The architecture of a neural network is inspired by the way biological neurons interact with each other. But, unlike a biological brain where any neuron can connect to any other neuron within a certain physical distance, these artificial neural networks have discrete layers, connections, and directions of data propagation.

As a simple example, suppose you want to predict how much your house will sell for. Some of the steps you may follow are

Data Collection & Attribution — You will start by gathering data of how much the houses in your area have sold for. Then you will label the data by interesting attributes, for example square footage, number of bedrooms, age of the house, size of lot, month of sale.

Data Analysis & Prediction — Once you have the labelled data, you may plot graphs on the different attributes, different combination of attributes and pass them to different algorithms to determine which model provides the best “curve fit,” — the model wherein the predicted output matches the desired output. That model is then used and you will get a ballpark estimation of value of your home.

Imagine doing this for millions and millions of homes globally. In Big Data world where Volume, Variety and Velocity are critical, techniques like Deep Learning comes in. The power of Deep Learning lies in its ability to determine the features of a model from the data, rather than having features defined upfront by an expert.

  • In the example above, if the labelled data is passed to a neural network, then it will automatically learn the influencing features, if there is a large enough set of labelled data.
  • Once the neural network is supplied with a large enough labelled data, it can then establish a suitable mathematical relation between the different factors (or variables) by comparing it with the desired output.

One of the most famous Deep Learning example is through recognition of a “Cat” through analysis of 10 million You Tube videos, exhibited by Andrew Ng in 2012 at Google. To learn to recognize pictures of cats, a DL model will determine the features that comprise a cat, such as ears or whiskers, by looking at many example pictures fed to it. In a supervised DL model, the model would be provided labelled pictures of both cats and non-cats, explicitly telling the model what is and is not a cat. The system then determines which features are needed to build its cat model and can then recognize pictures of cats in the wild. Ng put the “deep” in deep learning, which describes all the layers in these neural networks.

Types of Machine Learning Models

Machine learning approaches can be classified under three headers — supervised, unsupervised and reinforcement learning.

  • Supervised: The computer is guided manually by a human to learn. The computer receives feedback from the human after every task. Herein linear regression is used to classify the data. In linear regression, we plot the given data on a graph and then use the plotted data points to obtain a relationship (maybe a mathematical equation) between the dependent and independent variables. Once we have obtained the relation between the two variables, we can then use this relation to predict values of unknown variables.
  • Unsupervised: As the name suggests, herein the computer isn’t guided. Feedback is not given at all, irrespective whether the machine wins (achieves the desired result) or loses (doesn’t achieve the desired result).
  • Reinforcement: Feedback is only given when the machine achieves the desired result

Practical Applications of Deep Learning

From cybersecurity, fraud detection, targeted marketing, healthcare, transportation, housing, space exploration, image recognition, finance, games — the practical applications of Deep Learning are many. Driverless cars, better preventive healthcare, even better movie recommendations, are all here today or on the horizon. On the other hand, concerns & debates around jobs reduction and overall impact on future societies when machines are smarter than humans also continue. The real story of Deep Learning is yet to be written, stay tuned for more….