Deep Learning is a subset of Artificial Intelligence that falls under the umbrella of machine learning. It is inspired by the structure and function of the human brain. Deep Learning is a method of enabling computers to carry out specific tasks that require human intelligence without any human intervention. The human brain consists of a complex web of neurons connected together, this network of neurons (also called neural network) is responsible for complex functionalities of the human brain. Deep Learning was built on the core idea of mimicking the complex process of human brain to enable machines to think on their own i.e. machines with artificial intelligence.
Artificial Neural Networks
Artificial neural networks were at the center of the deep learning revolution, these algorithms are loosely based on the structure of biological brains, which consist of networks of neurons interconnected by signal-carrying synapses. An artificial neural network consists of an Input Layer (as the human brain takes input through the different senses), the number of hidden layers (where all the processing takes place) and an output layer (as the decision is conveyed through some actions). In Artificial Neural Network as data travels through this artificial mesh, each layer processes an aspect of the data, filters outliers, spots familiar entities, and produces the final output.
Roots of Deep Learning
Deep learning has its roots back to the 1950s when the first neural network was invented. In the 1960s and 1970s, there has been a lot of excitement in this field resulting in various novel researches. In the decades of 1980s and 1990’s some breakthrough research in deep learning like the development of Convolutional Neural Network (For Image Processing and Computer Vision) and Recurrent Neural Network (For analyzing time series data) led to the further bifurcation of the field. This was the time when every year there is some new breakthrough research which changes the course of deep learning, the field was all set to explode and grow exponentially. But when deep learning was applied in real-life scenarios, soon people realized that there were two main challenges —
- The first and most prominent challenge was the lack of computing power.
- The other challenge was the limited availability of data.
This led to an AI winter, where research in the field has been dampened because of the above limitations, and people thought of it as an idea that is way ahead in the future.
Again, after two decades, in 2010 with the launch of Image Net competition (A very large dataset of around 14 million labeled images were made available opensource with an aim towards developing state-of-the-art Image Classifier), the field of deep learning picked up the speed and this time with ample computational power and plethora of digital data available, it was all set to explode. In 2011, Alex-net was developed that kicked off a convolutional neural network renaissance in the deep learning community. In 2014, Generative Adversarial Networks (GAN’s) was developed which was the biggest breakthrough in the era of modern deep learning.
Currently, deep learning is all around us. It’s used to determine which online ads to display in real-time, identify and tag friends in photos, translate your voice to text, translate text into different languages on a Web page, and drive autonomous vehicles. It’s a key technology that is achieving results that were not possible before. Recent advances in deep learning have improved to the point where deep learning outperforms humans in some tasks like classifying objects in images.
Deep learning is also found in less visible places like credit card companies use deep learning for fraud detection, businesses use it to predict whether you will cancel a subscription and provide personalized customer recommendations, banks use it to predict bankruptcy and loan default risk, hospitals use it for detection, diagnosis, and treatment of diseases. It is being widely used to automate processes, improve performance, detect patterns, and solve problems and the range of its application is almost limitless.
Feasibility of Deep Learning
The performance of deep learning depends on two key factors- first is the availability of intensive computation power and the other is a huge amount of data. In terms of computing power, we have moved much faster than what is described by Moore’s law (Computational capability will double every two years) and with the wide adoption of digital technologies, we are generating more data than ever before, making applied deep learning feasible.
“The question of whether a computer can think is no more interesting than the question of whether a submarine can swim.” — Edsger W. Dijkstra
Deep Learning: The Way Ahead...
In the current state of deep learning, it can mimic an infant’s brain. An infant’s brain is like a sponge, it will take some years for the web of neural networks in it to mature and infer or deduce like grown-up humans.
But with the development of new neural network architecture’s like Generative Adversarial Networks (which is capable of generating art, music like humans), Siamese Network (Classification of images through one-shot learning), OpenAI’s GPT-2 Model (capable of generating coherent paragraphs of texts, reading comprehension and text summarization with human-level accuracy), we are getting closer towards building deep learning-based systems which can adequately mimic the complex functionalities of a mature human brain.
In future deep learning, systems can far surpass human intelligence, resulting in advance cognitive systems that can intelligently and fluently interact with humans.