Summary of deep learning

Deep learning is the subset of Machine learning. It skips the manual steps, extracting features. More specifically statistical machine learning techniques are used to learn feature hierarchies. The summary of deep learning is not a simple one. It is complex. Deep learning is the subset of machine learning. Some of the point-wise details about deep learning are

Limitations of machine learning

Deep learning is needed at the limitations of machine learning. machine learning is not useful for working with high-dimensional data. It doesn’t work where we have a high number of inputs and outputs. Machine learning isn’t able to solve NLP, image recognition etc. types of critical problems. Deep learning has been applied in this kind of scenario. It could focus on exact features.

Neural Networks

Deep learning is implemented through neural networks. It exactly works just like the neurons of our brains. As we say statistical machine learning techniques that’s why from input to output it has some hidden layers. In this case, artificial neurons came. Just like the neurons, the artificial neurons separate the output according to the dataset.

Process

Input layers —–> hidden layers —–> Output layers

Perception

The perception learning algorithm is a type of artificial neuron. Through this, we first initialize the inputs to the transfer function to calculate output, then transfers to the activation function for update and then get separated output. The activation functions are the step function, sigmoid function and sign function.

Similarly, it can be implemented through logic gates.

Problems before Neural networks

A specific step which computer needs to follow as well as the computer can’t solve a problem. Additionally, this restricts the problem-solving capability of the computer. Especially, which we know how to solve. Consequently, the computer follows a set of instructions.

In fact, with Neural networks computers can do these things.

Frameworks

Deep learning frameworks use to solve problems. Similarly, it uses for designing, training etc. purposes

In brief, frameworks applied for libraries, sources, performance etc.

Few are CNTK, Mxnet, Keras, Pytorch, Tensorflow etc.

A simplified diagram for implementation

In addition to the summary of deep learning framework implementation, we have two parts

1. Pre-processing of the dataset

start -> reading -> defining -> encoding -> dividing into train and test

2. making predictions of the test data

from the last step, we then apply framework data structure -> implement the model -> train -> Output

According to the framework sometimes the diagram changes. But not too much.

Deep learning applications

Of course, it has not been implemented fully in all the industries and sectors but the few applications are

1. self-driving cars
2. voice control
3. Auto image caption
4. Machine translation

and so on.