Have your ever wondered how every time you shop for shoes on Amazon, all the different kinds, makes and sizes of shoes pop up in just a matter of seconds? Yes, all shoes look the same but these days, there’s so much variety and creativity that even furry indoor slip-ons also line up the search results page despite being a far cry from regular office or street footwear.
These precision-focussed searches are driven by algorithms that use deep learning. As a science, deep learning is a subset of machine learning and involves the ability of data to improve its performance by studying examples on its own. All programmers need to do is feed as many algorithms as possible and the computer learns on its own to distinguish images, recognize voice patterns, translate languages as well as classify products based on what users are looking for. Shoes, cars, flowers or even detection of spam mail.
The way deep learning works is very similar to the way neural networks function in the human brain. Neural networks are extensive networks of neurons or nerve cells that are chemically connected to one another. Biological neural networks help humans transmit and process information from the outside world and are the main reason behind an individual’s cognitive and behavioural makeup. In short, it’s what helps us ‘think’ and ‘learn from previous experience’.
Artificial neural networks, as used in AI, mimic biological neural circuits in an attempt to make computers think and reason like human beings. Like the human brain, these neural networks are a series of layers where information is passed from one layer to the next. In addition, one layer’s output serves as the input for the consecutive layer. Between the input and output layers, there are several stacks of hidden layers, each segment analysing the information at hand.
So, while searching for footwear online, the computer rakes through millions of images and categorises only those images that have the features of a shoe, viz., its unique shape, heel size, colours, design etc. This is termed Feature extraction and is fed into systems by data scientists for helping computers accurately arrive at results that describe the original data set. Since deep learning require infusion of large amounts of data sets, they are built on extensive and powerful hardware that takes longer computational time.
The earliest computer-modelled neural networks were created by neurophysiologist Warren Sturgis McCulloch and logician Walter Pitts in 1943. This was followed by development of deep learning algorithms by Alexey Grigoryevich Ivakhnenko (credited with developing the Group Method of Data Handling) and Valentin Grigorʹevich Lapa (author of Cybernetics and Forecasting Techniques) in 1965. The first convolutional neural networks, where filters were added to the input, were used by computer scientist Kunihiko Fukushima to enable image recognition and process pixel data.
Today, Deep Learning is synonymous with Big Data and AI because of its sheer capability to draw patterns out of humungous data within seconds. Teams at DeepMind (acquired by Google in 2014), Google Brain, FAIR (Facebook AI Research), OpenAI, Baidu, Microsoft Research and IBM are driving automation to such an extent that actually makes you feel machines can actually ‘think’ and ‘act’ according to personalized needs.
In a 2019 interview to Forbes, Andrew Ng, founder and CEO of Landing AI and the man behind Google Brain AI research team and Chinese search giant Baidu, equated AI to electricity when it was first discovered to be a powering mechanism back in the day. “Remember that a hundred years ago, companies were hiring vice presidents of electricity. They were brought in to support the whole company and dictate how to handle this newfangled thing called electricity. AI today is in that early stage of development. So, you need that chief AI officer working closely with the CEO or CIO to give them the ability to drive change across the company,” he said.
According to Grand View Research Inc., a US and India-based market research company, the
global deep learning market size is expected to reach USD 10.2 billion by 2025, expanding at a CAGR of 52.1%. In India, deep learning and AI have been extensively used in the airlines, retail and fin tech sector. A 2019 report by businesswire.com shows that nearly 0.18 million to 0.2 million new jobs were created in 2018 for professionals with skills and expertise in machine learning applications.
Research labs like IBM Research, AIDAR Laboratory at IIT Kanpur, IISc & Wipro GE Healthcare at CDS Bangalore, Microsoft Research Lab India, IBM AI Centre of Excellence & Government E-Marketplace, and IIT Kharagpur have set up an AI-ML tech innovation hub to further R&D in innovative technologies.
The rise in use of voice assistants and digital entertainment services have accelerated trends within the machine learning industry. According to a 2021 survey carried out by statista.com, “39 percent of the respondents across India stated that they spent between three to nine hours per week on Netflix”. A report by IANS revealed that shipments of smart speakers market in India would cross 7.5 lakh units by the end of 2020, affirming the dominance of Deep Learning applications in every day life.
