# Deep Learning : Introduction

Nowadays, we keep on hearing “Deep Learning” every now and then, it has become a popular buzzword in machine learning world. Frankly speaking, it is just a term to describe certain types of neural networks and related algorithms that work on raw input data. These neural networks process input data through many layers of nonlinear transformations and then finally compute the desired output.

Not only in classification and clustering, but also in feature extraction deep learning is a pretty good option. Feature extraction is a data preparation technique applied on the input data when an algorithm is able to automatically derive meaningful features or attributes of the data to be used for further learning, generalization, and understanding. In conventional approaches, this burden is on the data scientist to carry out the feature extraction process to reduce dimensionality of the data, carry forward important data towards next processing steps. This leads to better results in learning process. In deep learning technique, not only data scientist’s burdon is reduced but also it’s performance is increased in tersms of simplification, reduction of computation and utilization of memeory. This process always gives you meaningful outcomes resulting in less chances of garbage out disaster.

Deep learning is one of the most popular cutting edge machine learning and artificial intelligence techniques of this time. The algorithms in deep learning are usually supervised, semsupervised and unsupervised types. Nowadays, tradional neural networks are called as shallow learning algorithms. Shallow algorithms are comparatively simpler and need more up-front knowledge of optimal features to apply, which typically involves feature selection. In deep learning technique, the algorithms rely on model tuning through selection of attributes. Hence, they are suited in problems where advance information of selected features is not necessary, and whereever labelled data is not available or not needed for the major application. Deep learning requires statistical techniques, neural networks and concepts and techniques from signal processing including non linear transformations. Non linear transformation fucntion is the one that cannot be characterized by a single staright line in feature space. Hence it needs many parameters apart from slopw to model the relationship between the input (independant variable) and output (dependant variable). Non linear fucntions contains polynomial, logarithmic, and exponential terms, and any other transormations that are not lonear. In actual physical universe, are modeled with non linear transformations.

In other words, the deep learning techniques, are designed for complex problems and find optimal solutions. Of course, on the cost of complexity they give best results.

## One thought on “Deep Learning : Introduction”

please write more about CNN, MLP, RNN etc