What Is An Artificial Neural Network?

Computers assist us in everything, such as work, entertainment, communication, watching movies, and more. Every year, we develop computers with better performance. However, this is far from enough for humanity; we dream of having supercomputers that surpass imagination and can help scientists process more data in a short time. Scientists draw inspiration from the way the brain is structured: neurons are interconnected in a vast network, and electrical impulses transmit data between neurons. They are trying to create a digital neural network. Unlike the electrical impulses in the human brain, in this network, data is represented digitally in electronic circuits. Artificial neurons connect to form a vast artificial neural network that can effectively perform tasks such as image recognition. Perhaps one day, computers will be able to think like humans!

Did you know? In just 1 second, the brain can receive billions of signals, and the number of signals flowing through the brain every second is unprecedented! The human brain is like an onboard computer; it is the most complex machine ever created, smaller than a soccer ball, yet contains more cells than there are stars in the Milky Way! If we compare the body to a ship, the brain is the captain. The human brain can quickly adapt to new and unfamiliar situations and can recognize objects faster than the most advanced computers in the world. When you see your best friend, the brain recognizes faces faster than a mosquito flaps its wings. Computers can recognize various complex patterns, but even the most powerful computers today cannot surpass the human brain. Scientists from different fields are still striving to understand the mechanisms of the brain.

As we mentioned earlier, neurons are brain cells that can conduct electrical impulses and transmit information to other cells. They consist of a cell body containing a nucleus and long, thread-like structures extending from the cell body to transmit electrical signals. These thread-like structures can be divided into two parts: one is the dendrite, which receives data from other neurons and transmits it to the cell body; the other is the axon, which transmits data from the cell body to other neurons. Dendrites are much shorter than axons, and axons are usually covered by a layer of fat called myelin; it acts like an insulator wrapped around electrical wires, helping electrical signals flow and transmit. The point where two neurons meet and transfer signals from one cell to another is called a synapse (see Figure 1).

Even when neurons are not sending electrical impulses to other neurons, there is always a weak intensity of electrical activity flowing along the neuron, which scientists refer to as “noise”. When devices are used to map the electrical activity of neurons, the “noise” appears as a slightly distorted line; when neurons transmit electrical signals, the image will show spikes or peaks (see Figure 2). Thus, we can consider that neurons exist in two states: “off” (noise) or “on” (sending sharp electrical signals) [1]. This modality can be represented by two mathematical symbols: “0” (representing off) and “1” (representing on). The language of using 0 and 1 is called binary language, which is the language of computers!

Figure 2 Electrical impulses can be represented in binary language, consisting of a series of “1” and “0” numbers.

Imagine constructing a large three-dimensional structure using pipes of various shapes and sizes, where each pipe connects to different pipes and has a valve that can be “on” or “off”. Ultimately, we assemble a structure where hundreds of pipes connect. This sounds quite complex, doesn’t it? Now, we connect these pipe devices to a faucet; the varying sizes of pipes will allow water to flow at different speeds. If a valve is closed, the water will stop flowing. Water represents the data transmitted in our brain, while the pipes represent neurons, and the valves act like synapses connecting neurons. Based on the binary coding of neurons mentioned in the previous section, scientists are trying to create an accurate and intelligent “digital brain”, connecting digital neurons just like we imagined with the pipes, ultimately forming a large, efficient, and reliable artificial neural network that collaborates with digital neurons.

The digital neurons that make up the artificial neural network are called nodes. Developers program each node with a special function called weight. We can compare the weight of a node to the valves in the imagined pipe structure, which are the synapses in the brain; the valves regulate the intensity of incoming signals. Let’s imagine that the pipes in the structure lead to a water tank, which represents an artificial neuron. Each valve regulates the amount of water entering the tank; the total amount of water collected from various pipes is considered the “input” of the tank, known as “signal input” in the artificial neural network; the valves represent the weights of the nodes, which regulate the signal intensity entering the node from the external environment and other neurons. Meanwhile, the tank is filled with images, signals, sounds, and other information received by the brain from the outside world, known as “data output”.

Each node in the artificial neural network has multiple “inputs” representing the input signals from the surrounding environment and other neurons. When the network is active, nodes receive different data through each “input” (represented numerically), multiply the data by the corresponding assigned weights, and then sum all the “input signals” to obtain a total, which is the “output signal”. Remember how we explained earlier that neurons may often receive weak electrical noise interference, which hinders signal transmission? In the artificial neural network, if the “output signal” is below a preset threshold, it is considered noise, and the node will not pass the data to the next layer; if the amount exceeds the threshold, the node will send the “output signal” to the next layer, similar to how when electrical activity is sufficiently high, the brain’s electrical signals transmit information through synapses. All of the above processes are recorded in the binary language represented by “1” and “0”.

Inspired by the human brain, to design the next generation of supercomputers, we need to construct a large neural network based on artificial neurons (see Figure 3). However, without mathematics to simulate the operation of real neurons, creating artificial neural networks would be a fantasy.