Scientists from St Petersburg University train a neural network to "think" like an ancient man
Experts from St Petersburg University, engaged in historical geography, have developed a neural network capable of finding new ancient religious structures in the territory of South-Eastern Altai based on landscape features of already studied archaeological sites.
The territory of South-Eastern Altai is one of the most promising area for historical and geographical research. Thousands of religious constructions from different eras, from the Aeneolithic Age (4th−3rd millennia BC) to the Ethnographic Age (15th−19th centuries), have now been discovered here. These finds make it possible to study questions of human settlement and migration. However, the localisation of monuments poses a problem for researchers. This is because the area covers a large territory and consists of a variety of landscapes and natural conditions, including mountain ranges and river valleys. Also, researchers register an acute lack of written sources describing life in this area at different times.
The research findings were presented at the international scientific and practical conference "Fifth Landscape-Ecological Readings dedicated to Grigorii Grishankov. Nature and Society: Integration Processes".
To solve this problem, geographers from St Petersburg University trained a neural network to recognise landscape features of known locations of ancient crop traces as part of a historical-geographical approach. A number of landscape features are used to build the neural network model, including: true altitude; a position relative to watercourses; inclination; exposure; intensity of solar radiation; and visibility of mountain peaks to name just a few. This data allows for the most complete investigation of the area and highlights locations where new traces of ancient cultures can be found. The accuracy of detection with the developed neural network reaches 80%. This can be improved by: increasing the knowledge base of already investigated sites; and adding new landscape criteria to the analysis.
‘Our neural network identifies patterns based on pre-existing data. In this case, the landscape is crucial because human settlement in ancient times was strongly influenced by environmental conditions. For example, the tribes that lived in South-Eastern Altai preferred to settle along river valleys, mainly at the confluence of two rivers in the south-eastern and south-western directions, as well as along river valleys in the intermontane Kuray Steppe and on the periphery of the Chuya Steppe, where the conditions for living were most comfortable. We have studied landscapes of about 900 archaeological sites found there. This data made it possible for us to teach a neural network to literally "think" in environmental categories as ancient people could do. All this makes it possible to predict the location of new archaeological sites with great accuracy,’ said the research participants, Igor Sergeev, Associate Professor in the Department of Geomorphology at St Petersburg University, and Anastasiia Glebova, Assistant Professor in the Department of Physical Geography and Landscape Planning at the University.
The analytical work was supported by a grant from the Russian Foundation for Basic Research.
As a result, the neural network discovered traces of kurgans not yet explored by archaeologists − places of possible human settlements and their ceremonial activities. These predictions were confirmed by satellite images. As noted by Igor Sergeev, the developed neural network can be used in different geographical areas, depending on the scientific goals and objectives.
Additionally, neural network analysis allows a better understanding of the settlement patterns of ancient peoples and the cultural preferences of people. Thus, this data can be used not only to search for monuments, but also in analytical work and interpretation of certain finds. In future, the researchers plan to use the data obtained by the neural network to conduct fieldwork during scientific expeditions.