A powerful AI system tasked with ensuring your safety might imprison you at home. If you asked for happiness, it might hook you up to life support and ceaselessly stimulate your brain’s pleasure centres. If you don’t provide the AI with a vast library of preferred behaviours or an ironclad means for it to deduce what action you prefer, you’ll be stuck with whatever it comes up with. And since it’s a highly complex system, you may never understand it well enough to make sure you’ve got it right.”
― James Barrat, Our Final Invention: Artificial Intelligence and the End of the Human Era
“Compassionate Artificial Intelligence can transform and heal the world from a much deeper sense. They can add values and transform our world, our families, our workplaces, and our communities.”
― Amit Ray, Compassionate Artificial Intelligence
It is 2020, and the world has awakened to something bigger and more significant than true human values. We are growing towards automation and less human intervention. Sit back and letting machines do what you have been doing since past centuries. To feed machines with social values to lighten the baggage, is it possible? Is it easy? These question might not have been answered before a few decades, but today it has, and the answer is Artificial intelligence.
Artificial Intelligence is not individualistic but an ensemble. An ensemble of concepts like machine learning and deep learning. You might have come across these terms before, Machine learning and deep learning. How different are these concepts from each other. How similar these concepts are from each other. Let’s dig into the three one by one.
Artificial Intelligence came into existence in 1956 by John Mc Carthy to carry out mimicking machine to human behaviour. Machine learning and deep learning are the two radical subset of Artificial intelligence. Arranging the hierarchy, it can be seen as
Artificial Intelligence- Machine learning- Deep learning. Artificial intelligence is the superset of Machine learning, and deep learning is the subset of Machine learning.
Machine learning came into existence in 1959 by Arther Samuel.
In machine learning, an application is not explicitly programmed, but it learns the patterns with their own training and testing methods. Basically, it learns from its own mistake and build habits upon its personal experience. The aim of machine learning is to let the machine learn by their own with least human intervention.
Machine learning methods can be categorised into:
1.Supervised machine learning algorithm:
This learning algorithm has access to previous data set or labelled data, which is being processed. The machine learns the habits of previous data to produce output from the new instantaneous input.
Every output has two edges. One is “supposed to be” output, and the other one is the “actual” output. This algorithm carries out fetching error, i.e. (Error= Correct output- Actual output) which helps in modifying data towards more relevant accuracy.
2.Unsupervised machine learning algorithm:
This learning algorithm has got no past inferences or labelled data. It infers output based on behavioural patterns for an instantaneous input. It uses a clustering technique where outputs are based on similarity and grouping of data.
3.Semi-supervised machine learning algorithm:
It falls in between the previous two. Where some parts are labelled, and some are not. This method of learning considerably helps in enhancing the learning accuracy.
Deep learning is a machine learning technique in deeper levels which teaches machines to think like humans. The concept of artificial neural network is the key which drives the idea of many voices controlling apps such as phone tablets tv and speakers.
Deep learning is the major curator in solving classification task which teaches machine to distinguish between things by text, images or sound being the input which passes through several stages of neural networks. Deep learning is the evolved version of machine learning due to the fact that it reaches a higher level of accuracy. In invading accuracy, it may sometimes exceed human intelligence too.
So yes, the USP of deep learning is its accuracy. Deep learning requires sufficient computational power. Its major component relies on
- Lots of labelled data
- High-performance GPU’s
- Sophisticated algorithm
We can differentiate the two based on four parameters.
Machine language can work on low-level machines, whereas Deep learning tends to work on high-end devices.
The approach with which two algorithms proceed to solve a problem is different.
How you can go with the machine learning approach;
- Break the problem into a sub-category
- solve them individually
- combine the result
Suppose you have to recognise an image from the picture. You have to lean it through
- Object detection
- Object recognition
Using bounding box algorithm to see through the image and find all possible patterns. Using SVM with hog, recognise relevant objects and name it.
If you are dealing with solving the problem with the deep learning approach, the problem has to get solved from end to end.
Euronet a DL based algorithm would pass on the image, and it would tell you the name along with location hence identifying the object.
Machine learning is able to handle small data, and Deep learning can process extensive data, be it structured or unstructured.
Training time is required for Ml is than Dl or model training period.
A testing time for Dl is more than ML.
Machine learning < Deep learning
Deep learning> Machine learning
Choosing between Machine learning and deep learning depends on the size of Data and the complexity of the problem required to be solved. If you have a high processing GPU and lots of labelled data, it is convenient to go for Deep learning algorithms. If you have neither of these two, Machine learning can do the work. For a deep learning approach, you will need more than a thousand images to get the relatability and a high-end GPU to process through it.