Mobile-first used to be the war slogan of the tech giant until a few years ago. Now, AI-first has replaced the mobile strategy of businesses. Although we do not realize it, we are living in a world surrounded by AI-enabled personal assistants that do everything from fetching information from the Internet to playing music over voice commands or even doing menial tasks like setting intelligent reminders. They have become the defacto go-getters for the digital generation.
These virtual assistants are powered by the intelligence supplied by two major technologies – Machine Learning and Deep Learning. Machine Learning and Deep Learning have taken over the task of teaching computing systems how to think, understand and react in a given scenario.
In a way, they are software programs that can see, hear & speak just like human beings. Their progressive learning based on feedback and data models help them get better at what they do.
But, there are finer details that distinguish Machine Learning and Deep Learning from each other.
In today’s blog, we delve into these finer details to understand how they are responsible for making computers work the smarter way they work today.
Machine Learning – Self-learning Workers Who Learn From Experience
Tom Mitchell, the Author of Machine Learning and the E. Fredkin University Professor Machine Learning Department at Carnegie Mellon University says about machine learning as a,
“A computer program is said to learn from experience E with respect to some class of tasks T and performance measure P if its performance at tasks in T, as measured by P, improves with experience E.”
Let’s break that definition into simpler terms.
For instance, let us assume there is an algorithm to perform task ‘T’, like to drive a driverless car.
The performance of the algorithm can be measured using ‘P’
The P of the algorithm improves with Experience, which is measured using ‘E’.
Thus, P = T + E.
The efficiency of a machine learning algorithm improves with learning and experience. The learning is imparted by domain experts with sample datasets. With input from real-time data pipelines, the ML system gets better are delivering accurate predictions.
Google’s search algorithm, Netflix recommendation system, Amazon product recommendations, Tesla auto drive, are all examples of Machine Learning at play.
Broadly, there are 4 types of Machine Learning:
- Supervised machine learning
- Semi-supervised machine learning
- Unsupervised machine learning
- Reinforcement machine learning
Supervised Machine Learning
In supervised machine learning, the system is provided with labeled data called ‘training dataset’. The system learns to identify positives and negatives from the labeled data. The training dataset remains in place until the system is able to predict positives with reasonable accuracy.
Semi-supervised Machine Learning
In semi-supervised machine learning, the system is fed with partially labeled data. The system learns to predict positives and negatives through reiteration.
Unsupervised Machine Learning
In Unsupervised Machine Learning, the system is provided with no labeled data at all. It learns to make predictions using ‘clustering’ and ‘association’ of the data characteristics.
Reinforcement Machine Learning
In reinforcement machine learning, the system follows a rules-based procedure to complete a single goal. For example, playing a game of chess. The system knows how to move each coin and keeps making predictive moves until it wins the game.
Deep Learning – A Cognitive Technology That Mimics The Human Brain
Deep Learning works similar to the human brain. It develops patterns from the data to create a neural network that works like the convolutions of our human brain. To put in simple terms, it connects the data from various patterns observed from data.
The same definition that Arthur Samuel gave for machine Learning fits for Deep Learning too – “a field of study that gives computers the ability to learn without being explicitly programmed”
A deep learning system scans data and discovers patterns that can be used to distinguish one object from another. For example, a beer bottle and a glass of wine. Both have distinct edges with varying dimensions.
Deep learning goes further to identify what the object truly is based on complex calculations. It sees through the color, density, and other parameters of the object to come up with a prediction.
For example, deep learning can predict whether a liquid is a beer or wine based on the color and alcohol composition. Unlike machine learning, which needs to be fed with data, deep learning will learn to prepare data models from the given dataset. This is perhaps the point of difference between the two technologies. This self-recognition ability is what makes deep learning a success in image recognition, voice recognition, recommendation systems and so on.
Deep learning enables the system to adjust itself to the data that is fed to it. It combines both historical data and real-time data to make predictions. Google’s AlphaGo algorithm which beat the former world champion Lee Sedol at Go in early 2016 is made up of deep learning systems.
Imagine Artificial Intelligence to be sports. Machine Learning is soccer and Deep Learning is the Champions League of Soccer. Each cognitive technology is an inner layer of the bigger technology that is building the next generation of self-learning systems.
In fact, Machine Learning and Deep learning will take away our dependencies on devices. They can enable surfaces like Google Home, Alexa, etc. to respond to our voice commands and perform actions that were previously considered science fiction. Sundar Pichai in his Google blog says, “Looking to the future, the next big step will be for the very concept of the “device” to fade away.”