Hello all, till now, we have learned about various machine learning concepts like regression and classification. We applied some regressors and classifiers on test datasets and predicted values for them. However, we did not have much control over accuracy of those results and we could not do much to increase their accuracy. Accuracy of these predictions can be increased by solving the same problem using deep learning approach. So, in this article, we will have some introduction to deep learning.
Introduction to Deep Learning:
Deep learning can be achieved using neural networks and they have been around for quite a long time i.e. around 40-50 years. However, they were not used much in practical applications and did not gain much popularity in early days because training a neural network requires huge datasets and high computing power and they were not available in early days.
What has changed now?
Since the inception of internet, and numerous tasks are being performed on computers, we have huge datasets available to train neural networks.
The processing power of computers has increased substantially along with storage capacity and the hardware has become cheaper. We can buy hardware easily to implement deep learning applications.
A human brain is the best learning mechanism. Engineers and scientists are trying to mimic human brain behavior to teach computers to learn and to give them the ability to think themselves. Introduction to deep learning involves a lot of similar concepts. Human brain learns when millions and billions of neurons work together. Similarly, deep learning involves a lot of neurons working together.
Deep Learning is used in
- Artificial neural networks for classification and regression
- Convolutional neural networks for computer vision
- Recurrent neural networks for time series analysis
- Boltzman machines for recommendation systems
This article just gives an introduction to deep learning. From next article onwards, we will dive deeper into the Neuron, artificial neural networks and convolutional neural networks.