The new research shows how an algorithm which governs how much of the Internet functions is also active, in a similar way, in the human brain. The research group, at the Salk Institute, behind this observation hopes the insight will improve human understanding of both engineered and neural networks. The finding could, in addition, help with research into learning disabilities.
The Internet
The Internet is term applied to the global system of interconnected computer networks that use the Internet protocol suite (TCP/IP) to link devices worldwide. In essence the Internet is a network of networks, covering the private, public, academic, business, and government spheres, both local and global. While it’s easy to make crude anatomical comparisons between the brain (on one hand) and the Web and the Net (on the other), many neuroscientists are finding areas of commonality. This is perhaps clearest with neural networks.
Biological networks
A neural network, in biological science, is a series of interconnected neurons whose activation defines a recognizable pathway. Here the activation of neurons sends signals and this forms the basis of brain activity. In computing, biological neural networks have inspired the design of artificial neural networks. Artificial neural networks are processing devices (like algorithms) typically made up of hundreds or thousands of processor units. The networks are used for functions like capturing associations or discovering regularities within a set of patterns.
Similar functions for the human brain and Internet?
As to why the human brain might be similar to the Internet, it is put forward by some neuroscientists that a person inputting information into a computer is similar to a neuron receiving electro-chemical stimuli in the brain. Here the computer, like the neuron, houses a body which processes information. In addition, sending information from one computer to another can require the use of a connection like a fiber optic cable. This can be regarded as similar to the axon of a neuron, transmitting information along a mylinated sheath. Also, with both biological and computer systems, information is placed into packets: with the Internet it is IP packets containing bits of information, and with the neurological network it is the synaptic vesicles containing neurotransmitters.
With the new study, Professor Saket Navlakha explains that “the founders of the Internet spent a lot of time considering how to make information flow efficiently…finding that an engineered system and an evolved biological one arise at a similar solution to a problem is really interesting.” Professor Navlakha makes his connection based on common rules shared by both biological and artificial systems.
With computers, Professor Navlakha explains controlling information flow in order to avoid information routes from becoming clogged is based on the Internet utilizing an algorithm called “additive increase, multiplicative decrease”. Here a computer sends a packet of data and then waits for an acknowledgement from the receiver. When the data packet is rapidly acknowledged, this is a sign that the network is not overloaded and data can be transmitted at a fast rate. With each successive successful packet, a computer ‘understands’ that it is safe to increase its speed by one unit (the “additive increase.”). Where an acknowledgement to a data packet is delayed the computer knows that there is congestion and slows down. The slowdown is often by 50 percent (which is the “multiplicative decrease” aspect of the algorithm).
This process, Professor Navlakha puts forward in a research paper, is similar to how the brain processes and manages information. Here his researches found the neuronal equivalent of additive increase to be something called “long-term potentiation”. This occurs when one neuron fires closely after another. This strengthens the synaptic connection and makes it more likely the first neuron will trigger the second in the future. Moreover, the neuronal version of multiplicative decrease happens when the firing of two neurons is reversed (that is the second neuron triggers before first). This weakens their connection, with the end result being that the first neuron is less likely to trigger the second in the future. The biological term applied for this is “long-term depression”. The synapses in the brain’s neural network weaken or strengthen according to this rule, and the entire neural network adapts and learns.
A summary of the research is shown in the following video:
So, while the brain and the Internet are very different in the way they are constructed, there appears to be simple local rules, of a similar nature, that govern how both process information.
The findings are published in the journal Neural Computation. The study is called “Learning the Structural Vocabulary of a Network.”
In related news, neural networks are often used today to analyze complex data (such, in medicine, to find clues to illnesses in genetic information). While successful, no one knows precisely how these networks work. To find ways to increase understanding researchers are developing software to look into neural networks more fully.
Essential Science
This article is part of Digital Journal’s regular Essential Science columns. Each week Tim Sandle explores a topical and important scientific issue. Last we examined new research which showed how an imbalance of the wrong microbes in the gut can cause blood pressure to rise. The week before we weighed up the latest data from NASA and predicted what landing on the surface of Pluto might be like.
