Researchers Identify Intruders in Noise

A group of scientists from MIPT and Kazan National Research Technical University is developing a mathematical apparatus that could improve network security. The results have been released in the journal Mathematics.

Complex systems, such as network traffic or living microorganisms, do not have deterministic physical laws to properly define them and predict future habits. In this instance, an important part is played by correlation analysis, which details the behavior of the system in terms of sets of statistical parameters.

These complex systems are explained by trendless sequences, commonly specified as long-term time series or “noise.” They are variations generated by various resources and are among the most challenging data to assess and extract reputable, stable information.

One of the metrics utilized in economics and natural sciences in time series analysis is the Hurst backer. It suggests whether the trend available in the data will persist: as an example, whether values keep increasing or whether the growth will resort to decrease. This assumption holds for several natural processes as well as is discussed by the inertia of natural systems. For example, lake level change, which follows predictions derived from the Hurst exponent value’s analysis, is figured out by the current quantity of water and evaporation rates, precipitation, snowmelt, etc. Everything above is a time-consuming process.

Capturing a cyber attack

The quantity of traffic passing through network devices is extraordinary. This holds for the end devices – house PCs, but particularly so for intermediate devices such as routers, along with high-volume servers. Part of this traffic, such as video conferencing, needs to be sent with the greatest possible concern while sending files can wait. Or perhaps it is torrent traffic that is blocking a narrow channel. Alternatively, at worst, a network attack takes place, and it needs to be obstructed.

Traffic analysis needs computational resources, storage space (barrier), and time, resulting in latency in transmission. All of these are in short supply, specifically when it pertains to low-power intermediate devices. It is currently either a reasonably basic machine learning method, which experiences a lack of accuracy, or deep neural network methods, which require relatively robust computing stations with ample memory to release the infrastructure to run, not to mention the analysis itself.

The principle behind the work of the team of researchers led by Ravil Nigmatullin is relatively simple: generalize the Hearst exponent by including more coefficients to get a much more complete description of the changing data. This idea makes it feasible to discover patterns in the data that are generally regarded as noise and were formerly impossible to examine. It is possible to extract major features on the spot and use simple machine learning methods to look for new attacks on the network. Combined, they are more accurate than hefty neural networks, and also the technique can be deployed on low-power intermediate devices.

Noise is generally discarded, but determining patterns in noise can be highly beneficial. For example, the researchers have analyzed the thermal noise of a transmitter in a communications system. This mathematical tool allowed them to isolate from the data a series of parameters defining a specific transmitter. This could be a solution to one of cryptography’s problems: Alice sends out messages to Bob, Chuck is an intruder that tries to impersonate Alice and send Bob a message. Bob needs to differentiate a letter from Alice from a message from Chuck.

Data handling is penetrating deeply into all sectors of human life, with picture and speech recognition formulas having long since moved from the realm of science fiction to something we experience daily. This description method produces signal functions that can be utilized in machine learning, substantially simplifying and quickening recognition systems and enhancing the accuracy of decisions.

Alexander Ivchenko, a member of the Multimedia Equipments and Modern Technology Lab at MIPT, one of the authors of the advancement, statesed that the advancement of this mathematical tool can address the concern of parameterization and also analysis of procedures for which there is no specific mathematical description. This opens massive prospects in defining, evaluating as well as forecasting complicated systems.

Originally published by Phys.org. Read the original article.

Reference: Raoul Nigmatullin et al, Generalized Hurst Hypothesis: Description of Time-Series in Communication Systems, Mathematics (2021). DOI: 10.3390/math9040381