Artificial Intelligence - Neural Networks

Artificial intelligence - the neural network

Yet another area of ​​research in AI, the neural network, is inspired by the natural nervous network of the human nervous system.

What are Artificial Neuro Networks (ANNs)?

The inventor of the first neuro computer, Dr. Robert Hett-Nielsen defines a neuro network as follows:
"... A computing system is made up of a very simple, highly interconnected number of processing elements, which process information by their dynamic state response to external inputs."

The infrastructure of ANN

ANN's belief is based on the belief that by making the right connections to the human brain, silicon and wires can be mimicked as living neurons and dendrites.

The human brain is made up of 86 billion nerve cells and neurons. They are connected to other thousand cells by Axons. The exchange of stimuli or sensory organs from the external environment is accepted by dendrites. These inputs generate electric impulses, which travel through a fast neural network. The neuron can then send the message to other neurons to handle the issue or not forward it.

ANN is made up of multiple nodes, which mimic the biological neurons of the human brain. Neurons are connected by links and they communicate with each other. Nodes can take input data and perform simple operations on the data. The results of these operations are transmitted to other neurons. The output in each node is called its activation or node value.

Each link is related to weight. ANNs are able to learn, which is achieved by changing the weight value. The following example shows a simple ANN:

A specific ANN

Types of artificial neural networks

There are two artificial neural network topologies - feedforward and feedback.

Feedforward ANN

In this ANN, information flow is unidirectional. One unit sends information to another unit from which it does not receive any information. There is no response loop. These patterns are used to generate / recognize / classify. They have fixed inputs and outputs.

Feedforward ANN

Feedback ANN

Here, the feedback loop is allowed. They are used in content-addressable memories.

Feedback ANN

Working for ANN

In the topology diagram shown, each arrow represents the connection between two neurons and indicates the path for the flow of information. Each connection has a weight, an integer number that controls the signal between two neurons.

If the network produces "good or desired" output, there is no need to adjust the weight. However, if the network produces "poor or unnecessary" output or error, then the system changes the weight to correct the result.

Machine education in ANNs

ANNs are capable of learning and they need training. There are many learning strategies -

Supervised Learning - This involves a teacher who is more knowledgeable than ANN. For example, the teacher feeds some example data about which the teacher already knew the answers.

For example, pattern recognition. ANN comes with estimates and estimates. Then the teacher provides the answers. The network then guesses with the teacher's "correct" answer and makes adjustments based on errors.

Unsurveyed Learning - This is required if no instance data is set with known answers. For example, search for hidden patterns. In this case, the cluster i is divided into groups of elements based on an existing data set that divides into some unknown format.

Reinforcement Learning - This strategy was built on observation. ANN makes decisions based on its environment. If the observation is negative, the network adjusts its weight to be able to make different decisions next time.

Back promotion algorithm

This is a training or learning algorithm. It learns from the example. If you specify in the algorithm what you want the network to do, it changes the weight of the network so that it can produce the expected output for a particular input after training.

Backlinks are ideal for simple pattern recognition and mapping tasks.

Bayesian Network (BN)

These are graphical structures that represent the possible relationship between a set of random variables. The Bayesian network is also called a trust network or base net. The argument about BNS uncertain domain.

In these networks, each node represents a random variable with specific propositions. For example, in a medical diagnostic domain, node cancer proposes that the patient has cancer.

The edges connecting the nodes represent the potential dependence between those random variables. If two nodes are outside, one is affecting the other, they must be directly connected in the direction of the effect. The probability of the relationship between the variables is measured by the probability associated with each node.

The only obstacle in arcs in a BN is that you cannot return to the node just by following the directed arcs. Hence BNs are called guided ascetic graphs (DAGs).

BNs are able to handle multivariate variables simultaneously. The BN variable is made up of two dimensions -

Range of forecast

Feasibility proposals have been assigned to each.

Consider a separate set X = {X1, X2,…, Xn Consider Separate random variables, where each variable Xi can take a value from the finite set, denoted by Val (Xi). If there is a link from the variable Xi to the variable, Xj, then the variable Xi will be the parent of the variable Xj showing a direct dependence between the variables.

The structure of the BN is ideal for a combination of prior knowledge and observation data. B.N. Causes can be used to learn relationships and understand different problem domains and to predict future events, even in the case of missing data.

Building a Bayesian network

A knowledge engineer can build a Bayesian network. There are several steps that a knowledge engineer must take to build it.

Example Problem - A patient with lung cancer is breathing. He comes to see the doctor and he is suspected to have lung cancer. Doctors know that in addition to lung cancer, there are other possible diseases that can affect patients, such as TB and bronchitis.

Gather relevant information about the problem

Is the patient a smoker? If so, then high chances of cancer and bronchitis.

Is the patient exposed to air pollution? If so, what kind of air pollution?

Take an X-ray positive X-ray indicates either TB or lung cancer.

Identify interesting variables

The knowledge engineer tries to answer the questions -

Which nodes to represent?

What values ​​can they take? In what state can they be?

For now, let's just consider nodes with different values. The variable must take one of these values ​​at a time.

The most common types of inactive nodes are -

Boolean nodes - represent those offerings, taking the binary values ​​TRUE (T) and FALSE (F).

Sorted values ​​- A node represents contamination and can describe the degree of risk of patient contamination from values ​​{low, medium, high from.

Integral Value - The node called age can represent the age of the patient with possible values ​​ranging from 1 to 120. Even at this early stage, modeling options are happening.

Create arcs between nodes

The topology of the network should capture the qualitative relationship between the variables.

For example, what causes a patient to get lung cancer? - Pollution and smoking. Then add arcs from node contamination and node smoking node lung-cancer.

Similarly, if the patient has lung cancer, the X-ray result will be positive. Then add the arcs from the node page-cancer to the node x-ray.

BNN Arch Creation

Specify the topology

Traditionally, BNs are placed so that the arcs are centered from top to bottom. The set of parent nodes of node X is given by parents (x).

Cancer nodes have two guardians (causes or causes): pollution and smoking, while node smoking nodes are the ancestor of X-rays. Similarly, the X-ray node is the child (results or effects) of lung cancer and the node is the inheritor of smoking and pollution.

Conditional possibilities

Now the amount of connection between the connected nodes: This is done by specifying the conditional probability distribution for each node. Here only passive variables are considered, this takes the form of a conditional probability table (CPT).

First, for each node, we must look at possible combinations of the values ​​of those parent nodes. Each such combination is called the instant of the parent set. For each separate instance of parent node values, we need to determine the probability that the child will take.

For example, the parents of lung-cancer nodes are pollutants and smokers. They take possible values ​​= {(H, T), (H, F), (L, T), (L, F). The CPT determines the probability of cancer for each of these cases <00 of, 0.02, 0.03, 0.01, respectively.

Applications of neural networks

They can do things that are easy for humans but difficult for machines -

Aerospace - autopilot aircraft, aircraft fault detection.

Automotive - Vehicle guidance system.

Military - weapons orientation and steering, target tracking, object discrimination, facial recognition, sign/image recognition.

Electronics - Code Sequence Prediction, IC Chip Layout, Chip Failure Analysis, Machine Visual, Voice Synthesis.

Financial - Real Estate Valuation, Loan Advisors, Mortgage Screening, Corporate Bond Rating, Portfolio Trading Programs, Corporate Financial Analysis, Currency Price Forecasts, Document Readers, Credit Application Evaluators.

Industrial - production process control, product design and analysis, quality inspection system, welding quality analysis, paper quality prediction, chemical product design analysis, dynamic model of the chemical process system, machine maintenance analysis, project bidding, planning, and management.

Medicine - Cancer cell analysis, EEG and ECG analysis, artificial design, transplant time optimization.

Speech - Speech recognition, speech classification, conversion to text form.

Telecommunications - image and data compression, automated information services, real-time spoken language translation.

Traffic - Truck Brake System Diagnosis, Vehicle Table, Routine Systems System.

Software - Pattern recognition in facial recognition, optical character recognition, etc.

Time series forecasting - ANNs are used to predict stocks and natural disasters.

Signal Processing - Neural networks can be trained to process audio signals and filter them properly in hearing aids.

Control - ANNs are often used to make staircase decisions.

Detecting anomalies - Since ANNs specialize in identifying patterns, they can also be trained to produce output when something strange appears to be abusing the pattern.