Artificial Neurons: Building Blocks of Intelligent Machines

Imagine a world where machines can think and learn like humans. Where self-driving cars can navigate the roads without a human driver, performing medical diagnoses with unmatched accuracy, and even the most complex problems can be solved in seconds. While this may seem like science fiction, the development of artificial neurons has brought us closer to this reality than ever before. Artificial neurons are the building blocks of artificial neural networks, which can process vast amounts of information and can make intelligent decisions. In this article, we'll explore the fascinating world of artificial neurons, how they work, and their potential to revolutionize the way we live and work.

We know about neurons, which are the basic building block of the human brain. A neuron, and effectively a network of neurons, is responsible for all the functions of the nervous system, which controls everything from our breathing, emotions and control and coordination in our body.

The whole idea of modern concepts like Machine Learning and artificial intelligence revolves around simulating the neuron to help machines make decisions like human brains but with computers.

Now, whenever we talk about simulating a neuron using a computer, we need to do that using binary, the language a computer can understand. Now, talking about artificial neurons superficially, we need to do a mathematical representation of neurons, which computers can process and understand.

Now let us first have a glance at a biological neuron.

Each biological neuron consists of three main parts: the dendrites, the cell body, and the axon. The dendrites receive signals from other neurons, the cell body processes these signals, and the axon transmits the processed signals to other neurons.

The image above gives a fantastic analogy comparing artificial and biological neurons. Now you must have got a gist about the biological neuron, so let us finally dive into the topic of our concern.

The Artificial Neuron!

The artificial neuron is like a mathematical function of inputs given to that neuron. If you’re unfamiliar with this, a mathematical function is like a machine that takes an input (a number or a set of numbers) and applies a set of instructions or rules to it to produce an output. Think of it like a recipe: you put in some ingredients, follow the instructions, and get a tasty dish as a result.

For example, the function f(x) = 2x + 3 takes an input value x, multiplies it by 2, adds 3 to the result, and gives you a new number as the output. If you put in x = 4, the function would give you

f (4) = 2(4) + 3 = 11 as the output.

Now a neuron has several properties which decide, what will be the output, when we give a specific set of values as an input, these properties or these parameters are called “weights” and “biases”.

Now let us understand what these weights and biases are.

A mathematical representation of an artificial neuron looks something like this:

Here the σ represents the activation function, which we will discuss further, first focusing on what is inside the activation function.

The term inside simply says that it is the sum of all inputs multiplied by their respective weights, for example, if there are 3 inputs, the term insider activation function will look something like this, when we expand that term:

W1X1 +W2X2 + W3X3 + b

Here the terms W1, W2, W3 are weights.

Weights are values that the computer assigns to each neural network input. These values help the computer make decisions based on the input it receives. The values of these weights are decided based on the previous data we have.

These weights are different for different neurons, depending on the function we are expecting to be performed from a neuron.

Now, what are biases?

Remember the “b” in the mathematical representation of the neuron above?

Biases are a type of parameter that work together with weights to decide the output of an artificial neuron. Weights are used to determine how much importance each input should be given, while biases function as a sort of offset that helps to fine-tune the output of the neuron.

Now let us talk about an activation function,

In simple terms, the activation function is a mathematical function that takes in the weighted sum of inputs of an artificial neuron and produces an output signal, which is then passed to the next layer of the neural network.

The activation function is like a filter that decides whether the neuron should "fire" and send a signal to the next neuron or layer. The activation function helps to introduce non-linearity into the neural allowing it to simulate complex relationships between inputs and outputs, activation function plays a crucial role in determining the output of an artificial neuron, and thus the performance of the neural network.

Sigmoid, ReLu, and SoftMax are some of the activation functions which are popularly used and can be studied in further depth.

We’ve talked about neurons for a while now, but how these neurons are used in the applications we know?

The single neuron itself is not capable of doing complex things like image recognition, speech processing, and other applications of machine learning.

Artificial neurons are used in the form of networks of neurons, called Artificial neural networks.

Artificial Neural Networks, popularly called as ANNs, consist of layers of interconnected artificial neurons, where each neuron in a layer receives input from the previous layer, processes that input using its weights and biases, and passes the output to the next layer. The output of the final layer is the result produced by the neural network; They can be represented like this in pictorial form:

ANNs are used for a range of applications, for example, in self-driving cars, neural networks are used to identify pedestrians, other cars, traffic signs, and other objects on the road. The neural network is trained on a large dataset of images labelled with the objects in the image. Once trained, the network can accurately recognize and classify objects in real time, enabling the car to make decisions about its movements and actions. There are many more applications including image recognition, speech recognition, and natural language processing, they could be trained to detect objects, recognize spoken words, and make predictions on past data.

Natural language processing is another amazing application of these neural networks, which are trained to generate text and speech in human-like language.

You now have a clear understanding of how neurons work, and superficially know, how they contribute together to the applications we discussed in this article.

The development of artificial neurons has brought us one step closer to creating machines capable of thinking and learning like humans, with the continued development of artificial neurons and ANNs, we may soon be able to create machines that can revolutionize the way we live and work.