A48

• WebPage

Adaptive Foraging in Dynamic Environments Using Scale-Free Interaction Networks

We explored the influence of scale-free networks and alternative networks with long-range interactions on an artificial swarm. The task of the swarm was foraging in an environment with dynamically changing food availability. <br>To obtain our insights, we ran simulations based on C++ and analyzed the data in Python and Matlab. The simulations ran in parallel, remotely using the IDLab cloud infrastructure.

• PublicationIssue

Coherent collective behaviour emerging from decentralised balancing of social feedback and noise

This paper demonstrates how a collective multi-agent system is able to maintain coherence through adaptive communication. Mathematical models that describe the relationship between coherence and communication are proposed and validated by physics-based computer simulations.

• PublicationIssue

Collective Decision-Making on Triadic Graphs

In this work we compare different types of triadic graphs and their influence on collective decision-making. Triadic graphs are synthetic networks constructed from triadic motifs, i.e. network sub-patterns that were previously discovered to occur in various real-world systems such as gene-regulatory networks, food webs or neuronal networks. As a case study, we consider a special type of artificial collective systems, inspired by marching locust swarms.

• PublicationIssue

Ilja Rausch, PhD

• Person

Influence of triadic motifs on the dynamics of epidemic networks

Networks have become an important tool for the analysis of a vari-<br>ety of dynamic systems such as financial markets, coupled oscillators<br>or epidemics. Looking at structure properties such as clustering or<br>overrepresentation of certain motifs can lead to new insights on how<br>networks evolve in time. In my research I want to investigate how the<br>occurrence of specific types of triadic motifs can influence the dynam-<br>ics of disease spreading on Steiner Triple Systems. In order to model<br>the infection spreading, I consider the SIR model where susceptibles<br>(S) interact with the infected (I) at rate β and are removed (R) due to<br>recovery or death at rate γ. I use numerical simulations and recently<br>developed techniques such as Node Specific Pattern Mining (NoSPaM)<br>and Z-Score analysis to investigate the effect of overrepresented triadic<br>motifs on the clustering coefficient and the number of infected individuals for different<br>values of β and γ.

• PublicationIssue

Modeling the Influence of Social Feedback on Altruism using Multi-Agent Systems

We studied the impact of various complex networks on the resource sharing behavior of a distributed AI.<br>To obtain our insights, we ran simulations based on C++ and analyzed the data in Python and R. The simulations ran in parallel on a remote server, using the IDLab cloud infrastructure.<br>The presentation of our contribution during the ALIFE 2020 is publicly available https://www.youtube.com/watch?v=gIvbc4YCN7M

• PublicationIssue

Scale-Free Features in Collective Robot Foraging

In this study we perform a rigorous statistical analysis of various time and space related features of an artificial collective system. In particular, we use statistical tests to examine whether a robot swarm performing the foraging task, develops scale-free properties (i.e. those that follow a power law distribution).

• PublicationIssue

Towards effective continued pre-training of EU institutional LLMs on EuroHPC supercomputers

We studied the limitations of current Large Language Models (LLMs) in meeting the specific needs of the public sector and other entities in Europe, particularly focusing on language coverage gaps. Our research, conducted in collaboration with the European Commission's Directorate-General for Translation (DGT) and DG CONNECT, involved leveraging high-quality multilingual data from EU institutions to enhance the multilingual capabilities of open-source LLMs through continued pre-training on EuroHPC JU supercomputers. Our findings, based on human evaluations and automated benchmarks, demonstrate the potential of this approach in adapting LLMs to European linguistic diversity.

• PublicationIssue