Introduction to Neural Networks - Part 1
Why should you learn Machine Learning? It’s no secret that machine learning and AI is one of the most popular buzzwords of this decade, but is the hype justified? Lets take a closer look at what exactly we should expected from Machine Learning and AI followed by a closer look at what exactly Neural Networks really is and why they have become so popular.
Even though Machine Learning and the algorithms used in todays applications was introduced several decades ago, it's first within the last 10-15 years that we have started to see the real benefit of these. The reason why the area of Machine learning has seen such a growth can be boiled down to the data available. The amount of data we are able to collect nowadays with the introduction of the internet, Social media platforms and Internet of things just to mention some.. has lead to a revolution of the Machine Learning field, as these algorithms are in need of large datasets to achieve high accuracy.
Examples of areas where we will see MI revolutionize the current standard can be seen in the following articles:
https://vinsloev.blogspot.com/2019/07/neuralink-merging-of-brain-and-machine.html
https://vinsloev.blogspot.com/2019/04/artificial-intelligence-how-it-could.html
The articles above showcase just a tiny percentage of the fields where Machine Learning is having an impact now and will continue to have so in the future. But to fully see the possibilities of a new technology one most understand its details, so the rest of this article will be centered around Neural Networks and their inner details. The level of detail will be an overall introduction as this is the first of many Articles covering Neural Network and Deep Learning on Vinsloev Academy.
A Neural Network also referenced to as NN is a set of algorithms which takes inputs x and produce an output y. The special trade of a Neural network is that it is modeled after the human brain, which consists of neurons capable of recognizing patterns. A Neural Network can be divided into three types of layers namely Input Layer, Hidden Layer and Output Layer.
Based on our understanding of the different layers within a NN it's time to dive deeper into the notations of a network as this is used as part of the calculation. A layer within a NN can consist of one or more nodes (also referenced to as Neurons from the human brain), each of these takes a given input in order to produce an output. Lets try dive deeper into a single Node, from our NN model.
As illustrated above when looking at a Node, we see that for this particular Node, two types of calculation is to happen as this NN is based around Logistic Regression. A Node in logistic regression really represents two steps of computation. This is the case as we want each Node to output a value between 0 and 1, such that our final output of the NN states the probability that y is equal to one of the input features x:
This however is difficult to compute using Linear regression in it's normal form, using the formula:
Based on the sigmoid function the final calculation of a given node can be summed to:
Based on this we can focus on learning the parameters w and b such that our output of the sigmoid function becomes close to being equal to one. The illustration below should illustrate how this can be seen within a NN where we try to predict whether we're looking at a Dog or Not. Keep in mind this is very simplified and there is much more to image recognition than this.
In part 2 of this article we will look at how we can learn the parameters w and b and change thoese based on a cost function.
As extra material you can view this Deep Learning Series on the Vinsloev Academy YouTube channel:
Even though Machine Learning and the algorithms used in todays applications was introduced several decades ago, it's first within the last 10-15 years that we have started to see the real benefit of these. The reason why the area of Machine learning has seen such a growth can be boiled down to the data available. The amount of data we are able to collect nowadays with the introduction of the internet, Social media platforms and Internet of things just to mention some.. has lead to a revolution of the Machine Learning field, as these algorithms are in need of large datasets to achieve high accuracy.
Examples of areas where we will see MI revolutionize the current standard can be seen in the following articles:
Neuralink - The merging of Brain and Machine
The whole idea behind Neuralink is to make a fully working Human-Brain Interface(HBI), which will allow a human to control a computer or a smartphone directly from the brain..https://vinsloev.blogspot.com/2019/07/neuralink-merging-of-brain-and-machine.html
Artificial Intelligence - How it could lead to a greener and healthier world
With the use of AI farmers can now use sensors and data from satelites to better understand the health of their crops and thereby increase the producing..https://vinsloev.blogspot.com/2019/04/artificial-intelligence-how-it-could.html
The articles above showcase just a tiny percentage of the fields where Machine Learning is having an impact now and will continue to have so in the future. But to fully see the possibilities of a new technology one most understand its details, so the rest of this article will be centered around Neural Networks and their inner details. The level of detail will be an overall introduction as this is the first of many Articles covering Neural Network and Deep Learning on Vinsloev Academy.
Figure 1 - Neural Network (NN)
A Neural Network also referenced to as NN is a set of algorithms which takes inputs x and produce an output y. The special trade of a Neural network is that it is modeled after the human brain, which consists of neurons capable of recognizing patterns. A Neural Network can be divided into three types of layers namely Input Layer, Hidden Layer and Output Layer.
- Input Layer: The Input Layer consist of nodes which brings the initial data into the network.
- Hidden Layer: The term Hidden Layer refers to the nodes placed in a layer in which the true values are not observed. Hence you only know the values of the Input layer and Output Layer. This also means that you can't find these values in the training set which explains the name Hidden Layer.
- Output Layer: consist of a single node, which will make the final decision of what to output based on the layer before it and it's connected nodes.
Based on our understanding of the different layers within a NN it's time to dive deeper into the notations of a network as this is used as part of the calculation. A layer within a NN can consist of one or more nodes (also referenced to as Neurons from the human brain), each of these takes a given input in order to produce an output. Lets try dive deeper into a single Node, from our NN model.
Figure 2 - Node
As illustrated above when looking at a Node, we see that for this particular Node, two types of calculation is to happen as this NN is based around Logistic Regression. A Node in logistic regression really represents two steps of computation. This is the case as we want each Node to output a value between 0 and 1, such that our final output of the NN states the probability that y is equal to one of the input features x:
To solve this we will take the calculated value of the linear regression formular and use it as input to a Sigmoid Function. The Sigmoid Function will level out high and low values such that it fit in between the value space of 0 and 1.
A graphical illustration of the sigmoid function can be seen below:
 |
| https://en.wikipedia.org/wiki/Sigmoid_function |
In part 2 of this article we will look at how we can learn the parameters w and b and change thoese based on a cost function.
As extra material you can view this Deep Learning Series on the Vinsloev Academy YouTube channel:
Kommentarer
Send en kommentar