Our researchers discover path selection rules in the information flow of networks

The latest findings have been recently published in Nature Scientific Reports.

“We too were surprised to find how similar the topological characteristics of the various networks were at the level of abstraction”, stressed Zalán Heszberger, associate professor of the Department of Telecommunications and Media Informatics at the Faculty of Electrical Engineering and Informatics. He and his colleagues recently published a paper on their latest research in Nature Scientific Reports.

A crucial part in this is played by the nodes and their connections, which not only influences the appearance, but also, in a self-fulfilling way, the operation of a system. Previous studies have mainly focused on the structure of networks, however, we decided to go a step further and look at the dynamism of the processes, more precisely the information flow within a system. The main motivation for this article came from the questions that arose during earlier work by our research team.

Our researchers studied four different networks: the Internet, air traffic, paths in the human brain and a word morph game. Three common characteristic rules can be identified in these very different networks”, explained Zalán Heszberger. The first finding, which can be easily justified and is probably not very surprising, is that physical networks, such as the Internet, aim to use the minimum amount of energy possible: the path between the start and endpoint of the information flow is usually the shortest”, noted Heszberger.

We also found that these information paths tend to avoid so-called “valleys”, which means that nodes with more connections do not prefer the exchange of information through their subordinates. “It is the same in the workplace also, since those higher up in the hierarchy, such as bosses, do not like to communicate with each other through their subordinates”, as the researcher explained with an everyday example, which is also true for airport networks: two large airports are not likely to organise transfers through a smaller one, even if it means initially travelling in the opposite direction.

The third finding is that information tends to avoid cores where nodes at higher levels of the hierarchy are concentrated. “I won’t talk to my boss unless it is absolutely necessary, because it would cost me and it could mean a commitment”, the researcher continued with his workplace example.

Their research regarding the Internet is partly built on their previous work. “We previously studied path-selection within the Internet extensively, so we were fairly at home in this topic”, recalled Zalán Heszberger. “After discovering these selection rules, we wanted to find out if they also exist in other networks.”

Researchers identified an interesting phenomenon on the web: in spite of being an artificial physical system, the Internet shows similarities to natural networks created through evolution. Higher level service providers with many nodes do not prefer to exchange information with lower level service providers. “The key to this is that at a higher level of communication, such as that of the service providers, we are talking about natural networks based on human decisions, which are often influenced by corporate processes. Top executives of large companies negotiate with each other and do not like to involve smaller companies in this for business strategy reasons. This is an example for the second rule.”

“When it comes to information transfer between parts of the brain, there are still many unanswered questions”, emphasised József Bíró, professor of the Department of Telecommunications and Media Informatics at the Faculty of Electrical Engineering and Informatics and member of the research team, adding that when analysing the paths in the human brain they consulted noted experts in the field from the US and Switzerland. “fMRI is used to examine the activity of various brain regions and the neural pathways connecting them. Finding a link between functionality and the structural observations is the “Holy Grail” of brain researchers. The two must be combined in order to discover how information is transmitted in the brain and which networks enable us to actually think.”

They met an American-Russian scientist through a previous publication, who sent his data on these brain regions to his Hungarian colleagues. The dataset comprises the fMRI data of forty human subjects: first cooperation between eighty, then one thousand parts were analysed, while also measuring the density and strength of the neural pathways. “Brain researchers helped us to interpret these processes, because it is a completely different area of science with different concepts”, recalled József Bíró.

“We discovered that the previously mentioned rules are also present in the way these networks in the brain work”, added Zalán Heszberger. Although it is only an idea at the moment, it seems that the so-called Hebb rule can help us understand how these connections are built. According to the Hebbian theory, in very simple terms, the increase in active connections between neurons or brain regions strengthens these neurons or regions. In other words, ‘rich club’ models can also develop between brain regions.” (Editor note: The research team led by András Gulyás discussed this phenomenon in a previous publication, which was also covered by bme.hu.)

The researcher divided the topics into two groups: one usually focuses on the theoretical approach, while the other examines a more tangible subject, namely the data set of brain activities. “Data sets based on large and extensive measurements are extremely valuable”, he stressed.

Thanks to their existing connections they were the first to receive these fMRI images from their foreign colleagues, enabling them to work together in the hope of a joint publication. They also set out to create their own data set by using the „fit-fat-cat” word ladder game, the concept of which was developed together with Marian Slíz, a linguist from ELTE. The goal of the game is to transform a source word into a target word through meaningful intermediate words by changing only one letter at a time, using a pool of over a thousand three-letter words.

“Fortunately we had the developer background for creating a free downloadable mobile app”, recalled József Bíró. “The words chains were collected anonymously from the users, enabling us to compare the paths they used and display them graphically. This allowed us to get valuable data with little effort and without having to ask a lot of people to come and visit us for the tests.”

The researchers are planning to publish another separate paper on the highly valuable data set in one of Nature’s sub-journals, called Scientific Data, which was established as a forum for such research activities. József Bíró thinks that the research project and the publication of the study can also be interpreted using network analysis. “First of all we needed connections with the foreign brain researchers, for which we also needed the expertise of our colleague, András Gulyás. Instead of approaching someone at the top of the research team’s hierarchy, which probably would have been ignored, he targeted the level, a researcher, who he had a better chance of convincing to work together with him. This way more people became interested in the collaboration, which led to the article and is still ongoing. Networks give us plenty of ideas to work on, so we are already preparing the research for our next publication.”

HA - GI

Photo: Philip János

Illustration: Nature Scientific Reports