Created by Akanksha

Neural Network — MADALINE to Self-Driving Cars

Features that are relevant to prediction using Machine Learning Concepts and Data Quantification process by Domain Experts or by using some algorithm to divide the data in two parts and then test the accuracy of model that

Akanksha Singh
May 20 · 11 min read

Initially in era 1837, Charles Babbage have proposed Mechanical Computing Machines for Calculations and Computations. Those machines had basic properties of ALU, Flow Control, and Integrated Memory. Although computers over the decades of evolution have changed a lot but human interruption is needed as computers don’t have intelligence and can only perform the tasks we ask. Computers can’t Predict and can’t schedule themselves as per the conditions.

Computer intelligence could only be programmed into these machines using some Data Set, based on that we have some Formula called as Model. Predictions that we get over the Experience of large amount of Data then finding Features that are relevant to prediction using Machine Learning Concepts and Data Quantification process by Domain Experts or by using some algorithm to divide the data in two parts and then test the accuracy of model that behind the scene giving intelligence to computer.

The above concept of model introduced by Data Scientists from ML (Machine Learning). Machine Learning States that, “Whatever Data Set We have in Storage Just Learn from them by doing some Processing and the later use the Learning in taking Decisions for the other Similar Data Set.” Also we know that we are here to discuss Neural Network then why we are discussing Machine Learning. Then another Question will arise 🤔 “Are Machine Learning Model using Neural Networks for Learning?”

So lets jump on discussing Neural Network : Layer of Neurons that combines to give an analytical function to learn and predict using Logics and can retrain the model whenever the accuracy goes down. This in turn introduces a new concept of Deep Learning. Neural network are like human brains and helps computer to take decisions among themselves without human help i.e. Retrain model whenever data Set arrive and enable machines to take accurate decisions. Whereas in Machine Learning we need to check the accuracy and manually need to retrain model.

Hence in Artificial Intelligence that we are providing to our Machine is subdivided in two categories, First is Machine Learning where we need to write program that retrieve correct learning rate and create model for further decisions. Second is Deep learning structures algorithms in layers to create an “artificial neural network” that can learn and make intelligent decisions on its own and it’s Subset of Machine Learning.

Fig 1. AI Neurons Simulates our Brain Neuron

Neuron :

Fig 2. Neuron

Neurons are just like a learning cell which contain numbers. A network of neuron corresponds to human brain and the end results\output that we get after the prediction\experience\operational learning is directed to some output medium be it speaker\monitor\printer. The intelligence in Computers are due to the evolution of these Neural Networks as we have understood in first part of this blog.

I want to represent this section like a human nervous system which give different results in different type of input\situation. Within an artificial neural network, a neuron is a mathematical function that model the functioning of a biological neuron. Typically, a neuron compute the weighted average of its input, and this sum is passed through a nonlinear function, often called activation function, such as the sigmoid.

Perceptron:

Fig 3. Perceptron with hardlim function

This is a special case of Neural Network which has only one neuron (with mathematical function) is associated with an Activation Function. Look at [fig2.] Perceptron , the number of input element could be R and each would have some weight which is defined after processing into the perceptron containing hardlim function that states that A = hardlim(W,b), here W=Weight and b=bias.

Fig 4. Hardlim (Hard Limit Transfer) function

Hard-limit transfer function gives perceptron ability to classify input vectors by dividing the input space into two regions. Outputs will be 0 if the net input n is less than 0, 1 if the net input n is 0 or greater.

Fig 5. AI — ML — DL — NN

🤔 What is the use of neuron in AI\ML learning model?

The output of the neuron further can then be sent as input to the neurons of another layer, which could repeat the same computation (weighted sum of the input and transformation with activation function).

When a data Set put in our AI model to train the neuron and create their own functions with appropriate activation functions. Each feature of the data is one input in the INPUT LAYER, then the further DENSE LAYER/HIDDEN LAYER holds some mathematical formula and Random Initializer that in turn find some Weight/Bias now these would be transferred to another layer of Neurons. More Neuron would result in better learning and accurate Weight and Bias and will provide Accurate Output in the OUTPUT LAYER. In later part of this blog we will discuss the Architecture of Neural Network.

History of Neural Networks and some industrial Projects:

In 1959, Bernard Widrow and Marcian Hoff of Stanford developed models called “ADALINE” and “MADALINE.” In a typical display of Stanford’s love for acronymns, the names come from their use of Multiple Adaptive Linear Elements. ADALINE was developed to recognize binary patterns so that if it was reading streaming bits from a phone line, it could predict the next bit.

Fig 6. Madaline Neural Network Model

MADALINE was the first neural network applied to a real world problem, using an adaptive filter that eliminates echoes on phone lines. While the system is as ancient as air traffic control systems, like air traffic control systems, it is still in commercial use.

Many other Neural Network Products Listed below:
1.
Traffic Sign Classification
2.
Breast Cancer Classification
3.
Music Genre Classification
4.
Chat Bot
5.
Gender and Age Detection
6.
Driver Drowsiness Detection
7.
Human Activity Recognition
8.
Image Caption Generator
9.
Colorize Black & White Image
10.
Handwritten Digital Recognition
11.
Iris Flower Classification
12.
Shock Front Classification
13.
Predicting Bird Species
14.
Detecting Trend in Fashion
15.
Unlocking Phone Using Face ID
16.
Forecasting Earthquake
17.
Human Brain Project
18.
3d Convolutional Speaker Recognition
19.
Image Super Resolution
20.
Self Driving Car
and Many more. Continuously we are making our systems autonomous and human independent day-by-day. Even we quote that Need of Today is Automation hence IOT, Autonomous Vehicles and many more things are added to the stack of Development and Growth — Awesome thing is that main Unit of Intelligence is Neuron.

Objective of Neural Network are:

Fig 7. How Computer detect image

With the human-like ability in problem-solving — and apply that skill to huge datasets — neural networks possess the following powerful objectives that they fulfill:

  • Adaptive Learning: Like humans, neural networks model non-linear and complex relationships and build on previous knowledge. For example, software uses adaptive learning to teach math and language arts.
  • Self-Organization: The ability to cluster and classify vast amounts of data makes neural networks uniquely suited for organizing the complicated visual problems posed by medical image analysis.
  • Real-Time Operation: Neural networks can (sometimes) provide real-time answers, as is the case with self-driving cars and drone navigation.
  • Prognosis: NN’s ability to predict based on models has a wide range of applications, including for weather and traffic.
  • Fault Tolerance: When significant parts of a network are lost or missing, neural networks can fill in the blanks. This ability is especially useful in space exploration, where the failure of electronic devices is always a possibility.
  • Classification: Neural Network organize patterns or datasets into predefined classes. This is a powerful feature as it implies to binary data analysis.
  • Prediction: They produce the expected output from given input. The main objective of any model is to provide accurate prediction for this we have the Optimizers and Magic Confusion Matrix.
  • Clustering: They identify a unique feature of the data and classify it without any knowledge of prior data.
  • Associating: You can train neural networks to “remember” patterns. When you show an unfamiliar version of a pattern, the network associates it with the most comparable version in its memory and reverts to the latter.

Architecture of Neural Networks:

Fig 8. Architecture of Neural Network

Understanding the architecture of an artificial neural network, we need to understand what a typical neural network contains. In order to describe a typical neural network, it contains a large number of artificial neurons (of course, yes, that is why it is called an artificial neural network) which are termed units arranged in a series of layers. Let us take a look at the different kinds of layers available in an artificial neural network:

Input layer:

The Input layers contain those artificial neurons (termed as units) which are to receive input from the outside world. This is where the actual learning on the network happens, or recognition happens else it will process.

Output layer:

The output layers contain units that respond to the information that is fed into the system and also whether it learned any task or not.

Hidden layer:

The hidden layers are mentioned hidden in between input layers and the output layers. The only job of a hidden layer is to transform the input into something meaningful that the output layer\unit can use in some way.

Most of the artificial neural networks are all interconnected, which means that each of the hidden layers is individually connected to the neurons in its input layer and also to its output layer leaving nothing to hang in the air. This makes it possible for a complete learning process and also learning occurs to the maximum when the weights inside the artificial neural network get updated after each iteration.

Neural Networks — Self Driving Cars

Fig 9. Self Driving Autonomous Cars

Let’s Now Start discussing about Self Driving Cars:

“Autonomous cars are no longer beholden to Hollywood sci-fi films”
— Elon Musk

Without any Human Inputs managing the functions of running a car is the feature of an autonomous car but when it comes to directions and road mapping to reduce and accident there comes the role of AI/DL. These are meant for the monitoring. Self-driving cars combine a variety of sensors to perceive their surroundings, such as radar, lidar, sonar, GPS, odometry and inertial measurement units. Advanced control systems interpret sensory information to identify appropriate navigation paths, as well as obstacles and relevant signage.

Self Driving Cars have 5 key in a Pipeline combining which it powers its built by big automobile companies like Toyota, Tesla and some more big names in market:

  1. Computer Vision:

Self driving cars have cameras to check the lane lines and find other vehicles these visual inputs are given to the car by not only one but more then 6 cameras (like: Tesla provides 8 cameras). Super vision abilities are there with these cars of around 150 meters what surrounds them. There are so many tasks that cameras enable like Lane detection, Road Curvature Estimation, Obstacle detection and classification and Traffic Light Detection. Deep Learning as immerged with some accurate approaches for pedestrian/object Detection and Classifications. The features and model training is done for this using Neural Networks that further works for color extraction, size and other features to detect a variety of object.

Fig 10. Computer Vision

2. Sensor Fusion:

It holds the data and sense them from another device like RADAR, LIDAR, CAMERAS. These sensors measures and senses the distance and velocity accordingly our sensors sense the data and make decisions of moving car. Each Sensor device has their own abilities like RADAR would tell something is in front of or back of the car — That would then be detected by LIDAR as it’s work is of Object detection, It create 3D imaging of the object. These both devices i.e. RADAR and LIDAR track the objects from far. A state transition model is used to estimate the other car movements and speed.

Fig 11. Sensor Fusion

3. Localization:

The GPS are fit into these cars to get the location they are at currently and the destination they need to go following which all areas globally. Since GPS are accurate about 2 meters which can result in major accidents. To reduce this we have created optimized algorithm to make it accurate by 2cm and for that we use post and mail boxes for the identification and planning.

Fig 12. Localization of Self Driving Cars

4. Path Planning:

Using these pathway maps the car can locate where the next steps are going to be put for which cars chart a trajectory. It decides this by checking what the surrounding vehicle movements will go and then how our chart map is going to design to create a map of vehicles. Next the trajectory is build to run our call safely.

Fig 13. Path Planning

5. Control:

The final and most complex thing is Control which is the reason of evolvement of these cars. Once the trajectory meant by path planning algorithms we have to work on the movement of the steering wheel. Similarly Accelerations and breaks are controlled. The major issue are because of Hard turns and High Speeds. The machines are really good and honest with the streams of algorithms we provide them to follow and will never do the uncontrolled behavior that may lead to car damage or any kind of accidents.

Fig 14. Control of Self Driving Cars

When it comes to creating algorithms for these above Pipeline creation we can use Keras (a library of Python)to create Neural Network models for replicating the human behavior. The control would be done by taking the camera’s input and then accordingly moves the steering to move in the pathway and sensing the objects then accordingly movements takes place. The frame of model that works in Self Driving cars are image mapping putting labels and then creating maps and decisions to take next control of our car. These Five Step Pipelines made the autonomous car more accurate and reliable for todays need.

A Convolutional Neural Network are used here with lots of dense layers and activation functions at each layer of data input. Each Ride data are stored and using that our car retrain itself to evolve as more accurate model for next drive and predictions. CNN makes an accurate set of codes which are Simple and Easy to understand and customized. CNN learns from those images and for building a generalized model from image dataset it get.

Last Words to this discussion would take these quotes:

“Autonomous cars are future of transportation, they have already started being deployed and will one day become commonplace.”

Geek Culture

Proud to geek out. Follow to join our +1M monthly readers.