eZoneToday

Neural Networks: Mimicking the Brain for AI

Artificial Neural Networks (ANNs) are computational systems inspired by the structure and function of biological nervous systems. These networks attempt to replicate the workings of the animal brain, utilizing interconnected processing elements, or neurons, to solve complex problems.

Both biological and artificial neural networks share neurons as fundamental building blocks. The conceptual groundwork for AI and neural networks was laid before the advent of modern computing, with computer emulations accelerating advancements in the field.

Currently, artificial intelligence and neural networks find applications in information processing, pattern recognition, and medical diagnosis. The field continues to evolve, presenting numerous opportunities for future development.

Biological vs. Artificial Neurons

The human brain comprises billions of neurons and specialized cells, forming the most complex organ in the human body. Although current knowledge of the brain is still limited—with estimates suggesting that less than 1% of its functions are fully understood—we possess a foundational understanding of its learning mechanisms.

A neuron receives signals from other neurons through a host of fine structures called dendrites. The neuron sends out spikes of electrical signals through a long, thin stand known as an axon, which splits into thousands of branches. At the end of each branch, a structure called a synapse converts the activity from the axon into electrical effects that inhibit or excite activity in the connected neurons. When a neuron receives excitatory input that is sufficiently large compared with its inhibitory input, it sends a spike of electrical activity down its axon. Learning occurs by changing the effectiveness of the synapses so that the influence of one neuron on another changes. In other words, information is stored in these junctions between each neuron.

Biological to artificial neurons

Artificial neurons mimic this biological function. An artificial neuron receives multiple inputs and produces a single output. Inputs replace dendrites, a processing unit emulates the cell body, and an output line functions as the axon. The processing unit can be programmed to generate specific outputs based on given inputs.

Neuron Operation

Neurons operate in two modes: training and usage. During training, a neuron learns to produce a specific output for a given set of inputs. In usage mode, the neuron outputs the associated output when a trained input pattern is detected. When an unrecognized input pattern is encountered, a firing rule determines the output state.

Firing rules enable a neuron to generalize from known input-output pairs to unknown ones, allowing the neuron to classify inputs based on their similarity to known inputs.

Artificial neural network example

For example, a 3-input neuron is taught to output 1 when the input (X1,X2 and X3) is 111 or 101 and to output 0 when the input is 000 or 001. Then, the truth table is;

X1:	0	0	0	0	1	1	1	1
X2:	0	0	1	1	0	0	1	1
X3:	0	1	0	1	0	1	0	1
OUT:	0	0	?	?	?	1	?	1

Output for the inputs 010,011,100 and 110 is not known to neuron. Here comes the firing rule Take the pattern 010. It differs from 000 in 1 element, from 001 in 2 elements, from 101 in 3 elements and from 111 in 2 elements. Therefore, the ‘nearest’ pattern is 000 which belongs in the 0-taught set. Thus the firing rule requires that the neuron should not fire when the input is 001. On the other hand, 011 is equally distant from two taught patterns that have different outputs and thus the output stays undefined (0/1).

By applying the firing in every column the following truth table is obtained;

X1:	0	0	0	0	1	1	1	1
X2:	0	0	1	1	0	0	1	1
X3:	0	1	0	1	0	1	0	1
OUT:	0	0	0	?	?	1	1	1

Therefore the firing rule enables neuron to respond ‘sensibly’ to patterns not seen during training.

More complicated neuron architecture

Above neuron doesn’t do anything that conventional computers don’t do already. In a neural network output of one neuron is given as input to another neuron. So, more sophisticated and meaningful neuron can be designed obtained by giving weight to each input. Each input’s influence on decision making depends on its assigned weight.  The weight of an input is a number which when multiplied with the input gives the weighted input. These weighted inputs are then added together and if they exceed a pre-set threshold value, the neuron fires. In any other case the neuron does not fire.

Output=f(X1W1+X2W2+………..+XnWn)

Neural network

A neural network is an interconnection of large number of neurons. Output of one neuron will be fed to input of one or more neurons. Apart from the input and output layers , hidden layers are include so that output will be fine tuned and error free. As the number of neurons in a network increases, intelligence of that network will also improve(as in biological neural system).

Artificial Neuron: Example

In above figure neuron is trained to recognise the patterns X and Y. The associated patterns are all black and all white respectively as shown below.Let us assume input is captured by 3 X 3 sensor array.

If we represent black squares with 1 and white squares with 0 then the truth tables for the neuron is.

A1	A2	A3	A4	A5	A6	A7	A8	A9	o/p
1	0	1	0	1	0	1	0	1	X

The truth table for above pattern is as follows.

A1	A2	A3	A4	A5	A6	A7	A8	A9	o/p
1	0	1	1	1	1	0	1	0	Y

Above two input sets are the taught input sets. Now, let us see what will be the output of the neuron if another set of input is introduced.

Truth table for the above pattern is given below. Initially output of the above pattern unknown to the neuron, but, on application of firing rule neuron will conclude above pattern is similar to taught input corresponds to ‘X’.

A1	A2	A3	A4	A5	A6	A7	A8	A9	o/p
1	1	1	0	1	0	1	1	1	X

Neuron will look for the similarity of this input with the taught inputs. This truth table is similar to first truth table in 7 ways(7 inputs are similar) and to the second table in 4 ways only. So, neuron will give the output as ‘X’.

By applying firing rule neuron will conclude above pattern is ‘Y’ since it’s 5 input are similar to taught inputs of ‘Y’ and only 3 inputs are similar to taught inputs of ‘X’.

A1	A2	A3	A4	A5	A6	A7	A8	A9	o/p
1	0	1	0	1	0	0	1	0	Y

Similarly observing other patterns given below you can see that , with only two taught input sets , system can identify up to five set of inputs. If a single neuron can do this much, what would be the capability of a network of tens, thousands or millions of neurons.? Today this algorithm can be implemented using simple computer programs.

Some more patterns and their truth table

A1	A2	A3	A4	A5	A6	A7	A8	A9	o/p
1	0	1	0	1	0	1	1	1	X

A1	A2	A3	A4	A5	A6	A7	A8	A9	o/p
1	1	1	0	1	0	1	0	1	X

A1	A2	A3	A4	A5	A6	A7	A8	A9	o/p
1	0	1	1	1	1	0	0	1	Y

Artificial neural network application

• Neural networks are best in pattern identification, identifying particular trends in a set of data; they are well suited for prediction or forecasting needs in sales, marketing, customer research, planning, etc.
• Medicine: Artificial neural nets are capable of disease diagnosis and treatment suggestions. Neural networks learn by example, so the details of how to recognize the disease are not needed. What is needed is a set of examples that are representative of all the variations of the disease. Their problem-solving ability and precision improve after each task (like humans).
• Humanoid Robots: Artificial intelligence can be incorporated into robotic brains using neural networks.
• Software: Pattern recognition in facial recognition, optical character recognition, etc.
• Self-driving vehicles and autopiloting.

Many applications of Neural Networks are yet to be explored. Their ability to learn by example makes them very flexible and powerful. With the advancement of high power computing we have a lot to gain from neural networks. Scientists believe that someday a ‘conscious’ networks might be produced. For that neural network will be the biggest contributor along with other technologies.