ERC Ongoing Projects

Afeyan Research Grants

PI: Suren Khachatryan

The past is never silent. It speaks to us in many ways, often quite literally – through handwritten manuscripts or carved stone inscriptions. Armenian monuments are particularly rich in the latter. These inscriptions serve as vital records of cultural and linguistic heritage, offering insights into the lives, beliefs, and traditions of Armenians during the Middle ages. However, reading, comprehending, and even detecting these inscriptions pose significant challenges. Due to weathering, vandalism, erosion, and the complexity of ancient scripts, many of these texts remain unreadable. Yet, the few existing studies indicate that deciphering these messages from the past is feasible with technological advancements. The modern inscription recognition methods are dominated by a two-phase approach – a character detection followed by their classification. We study a unique, newly created and unexplored collection of digital twins of Armenian tapanakars (tombstones) and khachkars (cross-stones), and process both 2D images and full 3D models aiming at the improvement of the detection accuracy. The hierarchical segmentation is implemented at three levels using the detected geometrical and statistical features – multi-line texts, individual text lines and characters or strokes in the text line. Next, we apply the results to character classification and estimate the accuracy of the generated character images. Finally, we discuss the extensions and further developments. Since the detection stage of the algorithm is universal for any kind of shapes, it opens up new research avenues that extend beyond text recognition alone. The same pipeline can be adapted to identify decorative motifs, geometric symbols, and other visual patterns commonly found on tapanakar surfaces.

Afeyan Research Grants

PI: Hrachya Kocharyan

Numerous numerical and experimental studies in the literature have extensively investigated how the geometric properties of granular particles modify the properties of granular matter, such as flowability and penetration resistance. However, all of these reports either experimentally study the macroscopic behavior of the matter or make simplifications in numerical analysis, which can result in significant distortion of real-world physical behavior. One of the major issues in numerical simulations is the inherent inaccuracy of the simple contact laws used, which approximate the contact behavior between spherical particles with high accuracy but fail to do so for complex shapes. There is a worldwide effort to improve the leading numerical technique used in granular matter simulations: the Discrete Element Method (DEM). One such effort is within the European COST Action CA22132 Open Network on Discrete Element Method (DEM) Simulations (ON-DEM), where the PI serves as one of the working group leads. The proposed research collaboration stems from this action and will help better model complex shapes while attempting to improve existing contact models. Another significant outcome of the project will be the creation of a substantial experimental base for ongoing research in materials science.

Afeyan Research Grants

co-PIs: Aram Butavyan, Varduhi Yeghiazaryan

A number of recent studies, including works by the members of our team, indicate the great potential of traditional 3D hyperspectral imaging (HSI) and spectrally enhanced 4D HSI in revealing otherwise invisible surgical targets. However, the high dimensionality and resolution of HSI data makes automated processing and analysis of such data difficult or intractable. Furthermore, while there is an extensive body of literature on the automated segmentation/classification of HSI data for the field of remote sensing, the automated processing of HSI data for biomedical applications remains under-explored. The aim of the project is to design and implement deep-learning-based procedures for fully automated parallel segmentation and pixel-level classification of 3D and 4D hyperspectral images of atrial tissue undergoing radiofrequency ablation procedure, incorporating spatio-spectral information from the original image and segmentation output. Determine the optimal 3D/4D data configuration for the best tissue classification results. Our prior successful application of deep learning for the classification of 3D HSI of radiofrequency-ablated atrial tissue is described in our IEEE Access article published in February.

The team has currently utilized eight datasets, produced at the L.A. Orbeli Institute of Physiology, and established an initial experimental setup. Six deep learning models have been chosen for evaluation. Preliminary findings spectrally enhanced 4D data are available. A conference paper is in preparation, to be submitted to the Workshop on Hyperspectral Image and Signal Processing: Evolution in Remote Sensing (WHISPERS) 2025.

Afeyan Research Grants

PI: Habet Madoyan

As autonomous systems become increasingly central to global technological innovation, Armenia has an opportunity to lead in specialized areas like unmanned aerial vehicle (UAV) control. This project proposes the development of an intelligent, reinforcement learning (RL)-based system for real-time adjustment of PID (proportional-integral-derivative) coefficients—crucial for stable and responsive UAV flight.
Traditional PID tuning methods are manual, time-consuming, and require domain expertise. Our proposed approach leverages cutting-edge machine learning to automate this process, enabling UAVs to dynamically adapt to changing flight conditions such as wind, payload, or mission type. The outcome will be a fully functional, open-source module integrated into UAV firmware, with improved flight performance demonstrated in both simulated and real-world environments.
The project is led by Habet Madoyan (PI), in collaboration with Davit Ghazaryan of AUA and Aram Harutyunyan from Airworker, one of Armenia’s leading UAV manufacturers. This academic-industry partnership ensures both research rigor and direct impact on Armenia’s growing UAV sector. Importantly, the initiative also provides AUA students with applied research experience at the intersection of artificial intelligence, robotics, and control systems.
This proposal reflects the core mission of the Afeyan Foundation: to support transformative, locally driven innovation with global relevance. Through this project, we aim to position Armenia as a contributor to the next generation of intelligent flight systems.

Afeyan Short-Term Collaboration Grants

co-PIs: Narine Sarvazyan, Ethan Bourdeau

In recent years, the focus of preservation efforts has expanded beyond physical structures and visuals to encompass the auditory dimension of cultural heritage. Armenia, known for some of the earliest examples of Christian architecture, offers a unique opportunity to study this heritage. Architectural acoustics, combined with advanced technologies such as LiDAR and 3D modeling, can deepen our understanding of how music, liturgy, architecture, and nature intertwine. This project aims to create a complete digital representation of both spatial and sound features of three different Armenian cathedrals, advancing knowledge in both the fields of architecture and heritage conservation and providing unique educational opportunities to the AUA students. This project showcases a robust international collaboration. Ethan Bourdeau, an Assistant Professor at Columbia University and founder of Bourdeau Acoustical Design, joins forces with three faculty members from the American University of Armenia, including Mihran Gurunian, Satenik Mnatsakanyan, and Narine Sarvazyan.

Afeyan Short-Term Collaboration Grants

co-PIs: Varduhi Yeghiazaryan, Irina Voiculescu

Hand-annotating medical images with segmentation masks requires an immense amount of time and effort from clinical experts. Replacing full masks with a simpler annotating gesture can mitigate annotation costs. This can come in the form of a scribble, and leads to weakly supervised training scenarios. Scribble-supervised segmentation typically utilises advanced neural architectures to compensate for the limited training data. Instead of just relying strictly on the pixels from each scribble, we also enhance each scribble by spreading, i.e. propagating, annotation labels through the image. We use a hierarchical partitioning of the image, produced with watershed/waterfall transforms, and propagate the individual pixel labels through the waterfall regions. We propose that a semantic label can be propagated to all other pixels in the same waterfall region. This increases the number of pixels that can be used for training supervision. We show experimentally that this technique greatly boosts the performance of established neural architectures on public semantic segmentation datasets like ACDC and MSCMRseg.

Our first paper got accepted to be presented at Medical Image Understanding and Analysis (MIUA) 2025 conference. The paper will be included in the conference proceedings volume, published in the Springer Lecture Notes on Computer Science (LNCS) series. We are currently working towards an extended version to be submitted to the IEEE Transactions on Medical Imaging.

Afeyan Short-Term Collaboration Grants

co-PIs:

Arsen Gasparyan

Robert Lücking

Armenia, a part of the Caucasus Biodiversity Hotspot as designated by Conservation International, hosts substantial fungal biodiversity. However, this diversity is largely overlooked and undervalued. Globally, less than 10% of fungi species have been described, and about 2–3 million species of funga yet to be discovered․ One of the most recent modern techniques that allows scientists to expand their understanding of biodiversity is eDNA metabarcoding. It enables the detection of multiple species from environmental DNA (eDNA) present in soil, water, air, or other types of samples. The eDNA metabarcoding approach has been applied to various purposes in funga research. It has proven to be an effective complementary tool for monitoring lichen and fungal biodiversity at diverse scales․ For this project's pilot study, the Dilijan National Park has been selected. The dominating temperate mountainous forests and other ecosystems of the Dilijan National Park host rich a flora, fauna and funga with high species diversity. They serve as key habitats for more than 480 known species of fungi and at least 130 species of lichens. We will study the lichen and fungal biota using eDNA metabarcoding, with a focus on identifying threatened species and overlooked habitats (e.g. deadwood), as well as applying conservation strategies. In the frame of the project, we aimed to establish and further develop collaboration between AUA Acopian Center for the Environment and the Botanic Garden and Botanical Museum Berlin (BGBM), which is affiliated with the Freie Universität Berlin (FU).

Siemens

PI: Hrachya Kocharyan