Deep Learning

Have you ever wondered how the Face Recognition system works or how google translator works? These all come under Deep Learning. So, what is Deep Learning?

What is Deep Learning?

Deep learning is a subset of machine learning and which in turn is a subset of artificial intelligence. Artificial intelligence is a process of teaching a computer to work as a human, machine learning is a technique to apply AI through algorithm trained using data and Deep learning is a type of machine learning inspired by the structure of the human brain. In Deep Learning, this structure of the human brain is called an artificial neural network.

Deep learning model learns to predict from past examples. image, text or sound can be given as an input. Deep learning algorithm will perform a task repeatedly, tweaking its output little bit every time to reach the desired goal. We call it deep learning because the neural network has various layers for learning and training an algorithm.

Why Deep learning is used more?

In terms of accuracy deep learning model gives pretty good accuracy compared to other traditional methods. Deep learning can achieve accuracy, sometimes exceeding human-level performance. Due to good accuracy deep learning is used these days in almost every AI project and even with safety-critical applications like a self-driving car. To train a model using deep learning algorithm tons of data is required. This is a reason why deep learning is advancing and being used these days. We have ample data on the internet and many companies store lots of user data.

Deep learning allows a machine to solve complex problems. Even if the dataset is diverse, unstructured or interconnected.

Why Deep learning is used more?

The algorithm used for prediction in deep learning is called a neural network. Generally, a large amount of data is required to train the neural network. In the training process Image text or sound is given as input to the network. It converts this information in numbers, these numbers are called neurons. They are passed to hidden layers connected to every node in the layer. The hidden layer consists of many layers. In each layer mathematical computation are performed and passed forward. During this process, important features are extracted from input data which helps a network to recognize an object or a thing. After this process network gives output in terms of percentage i.e. according to below example, if there are two products in the system to train banana and orange. If we show orange to the system than it will say how much percentage it is orange in comparison with banana.

Difficulties faced while implementing the Deep Learning algorithm

1. Big Dataset: Deep learning is good with unstructured data but it needs a massive amount of data to train a model with good prediction capacity.
2. Computational Power: As deep learning model needs a good amount of data, we need to process this large data on a computer. But this processing is not handled by all the CPUs for this purpose user need GPUs which have more cores compared to CPUs and GPUs are more expensive.
3. Time of training: deep neural network takes hours or even months to train this data. Training time increases with the amount of training data and the number of layers used in a neural network.

Mostly these difficulties don’t matter much in comparison with the output and accuracy deep learning gives.

Applications of deep learning

Deep learning applications are used from driverless cars to medical devices.

Driverless cars: Deep learning is used to detect objects, person, signboards and lane on road. This detection helps the car to take turns, maintain speed and avoid hitting any object on road.

Medical: Detecting tumour in the brain is possible by giving Magnetic Resonance imaging (MRI) to the neural network. It can extract features of having a tumour from an image and recognize it.

Music: deep learning can even be used in music for automatic music generation.

Security: deep learning is used in recognizing face whether the person appeared is an authorized person or an intruder. This way security of place increases.

There are lot more applications where deep learning as played an important role.