DIS Springboard Projects
2027
Bell's inequality in social systems
How can we disentanble non-classical contextuality from classical direct causal influences in human decision-making and social interactions? We want to evaluate the applicability of Bell-type, CHSH, and Leggett-Garg inequalities in this context. By pairing mathematically rigorous frameworks – such as Contextuality-by-Default – with controlled behavioral experiments and observational social media analysis, the project tests whether cognitive and social choice dynamics violate classical Kolmogorovian probability bounds. Ultimately, the project replaces uncritical “quantum social science” with an empirically grounded, mathematically precise taxonomy for modeling decision-making under ambiguity and context dependence.
Complexity in games
Can competitive games —ranging from discrete turn-based board games to real-time table soccer — be quantitatively modeled as complex dynamical systems exhibiting phase space trajectories, state entropy, and chaotic sensitivity? By combining formal metrics like evaluation divergence on neural embeddings with empirical analysis of 2,500+ table-soccer match logs and chess variation data, we want to establish a mathematically rigorous framework for measuring game tension and strategic complexity. The project bridges non-linear dynamics, game theory, and empirical sports analytics to distinguish genuine non-linear complexity from mere metaphorical chaos.
Data-driven dependency mapping of European energy infrastructure
European energy markets are deeply interconnected through cross-border electricity flows, shared generation resources, and coupled pricing mechanisms. However, the systemic dependencies between countries and regions remain insufficiently mapped from a data-driven perspective. Openly accessible market data (e.g., prices, generation, load, and cross-border flows) provide a foundation for estimating how strongly events in one region propagate to others. This project explores whether such data are sufficient to construct a dependency graph of European energy infrastructure that reveals plausible cross-border contagion pathways.
Economic complexity with imports included
A major limitation in classic economic complexity algorithms is the artificial inflation of country fitness scores, caused by assembly trade and high intermediate imports. Can we develop an import-adjusted fitness framework? By integrating disaggregated trade data with multiregional input-output matrices, the project aims to construct product-to-product requirement tensors and reformulate non-linear capability metrics to isolate true domestic value creation from global value chain assembly. The project combines complex systems modeling, network science, and empirical trade economics to advance economic complexity theory and growth forecasting.
Laplacian matrices to model asymmetric flow
How does local network asymmetry propagate into large-scale statistical patterns? In many real-world networks – such as urban road systems – flow moves differently one way relative to the other, yet most methods connecting local to global structure assume symmetry. We aim to use magnetic Laplacians to model asymmetric traffic flow and spatial directional structure. By extending complex Hermitian operator theory to spatial graphs, the goal is to replace heuristic network metrics with a rigorous mathematical physics framework for urban diffusion and transport. The project combines spectral graph theory, large-scale OpenStreetMap data analysis, and transport dynamics to connect local topological flow chirality with macroscopic urban scaling laws across cities globally.
LLM agents in financial markets: a simulation of structural change
Autonomous LLM-based agents are evolving from research prototypes into deployed financial actors that perceive market data, reason, and execute trades with minimal human oversight. In contrast to classical algorithmic traders, these agents share highly correlated priors from overlapping training data and can be instantiated at scale within seconds. This project uses agent-based simulation to investigate how the progressive substitution of human traders by LLM-based agents changes structural market properties such as price stability, liquidity, and correlated failure modes.
LLM-based systematic detection of data manipulation in digital forensics
Ensuring the integrity of digital evidence is a core challenge in forensic investigations; manipulated data can take many forms (e.g., altered file contents, inconsistent metadata, backdated documents) and may be deliberately designed to evade detection. Existing forensic tools typically rely on rule-based approaches that capture only a fraction of possible manipulation patterns. This project investigates whether large language models can systematically detect signs of data manipulation across file content, metadata, and structural properties. The expected outcome is a proof-of-concept system evaluated on datasets containing both authentic and manipulated samples.
Real-time supply chain signal intelligence and geopolitical vulnerability
Official trade statistics are published with significant delays, while geopolitical events that reshape supply chains such as export bans, sanctions, factory disruptions, regulatory shifts unfold in real time. This project develops methods to address this gap: building an early warning system that detects supply chain threats from open-source intelligence, including how structured information (affected companies, products, regions, and disruption types) can be reliably extracted from unstructured news data and other open sources, and how detected signals can be validated against structural supply chain vulnerability assessments to separate noise from genuine strategic threats. Using such mappings, the aim is then to develop production network models aimed to anticipate the potential consequences of such disruptions and to assess how specific policy interventions might reduce vulnerabilities.
Re-designing scientific publishing
Traditional scientific publication practices are facing a crisis. How can the research community move from legacy commercial publishing to zero-cost, decentralized journals built on preprint repositories (e.g., arXiv) powered by algorithmic reputation systems? Using game theory and high-performance Agent-Based Modeling, the project aims to design Sybil- and collusion-resistant reviewer incentives and simulate the micro-behavioral dynamics of authors and peer reviewers. The ultimate objective is to map out the phase transitions and systemic conditions required to trigger a widespread migration toward an open, equitable, and sustainable research dissemination ecosystem.
Social media influence on cryptocurrency prices
Social media has become a primary information channel in cryptocurrency markets, where highly followed accounts routinely post directional market calls reaching millions of retail investors. However, the relationship between individual influencer activity and realized price changes remains poorly understood. This project empirically investigates whether public statements by influential accounts precede cryptocurrency price movements, and how an account’s position in the influence network amplifies or dampens this effect.
Spatiotemporal dynamics of interactomes
Biological communities can be characterized by their interactome, the network of inter- and intraspecific interactions that determine ecosystem functioning and spatiotemporal dynamics. To understand interactomes, multilayer networks and their emergent properties will be investigated in light of abiotic and biotic drivers of system dynamics and interaction types. As ecosystems and their interactomes are rapidly changing in the Anthropocene, insights from such analysis will inform management and mitigation approaches, but also help us understand eco-evolutionary dynamics and their spatiotemporal scaling.
Stress-testing production networks of strategic and emerging industries
Geopolitical competition has turned supply chains into instruments of statecraft, yet most risk assessments still rely on simple bilateral import shares that ignore the cascading nature of supply disruptions. Understanding supply chain vulnerability requires more than knowing which countries trade with which; it requires knowing how products are actually made, which inputs are irreplaceable, and where production bottlenecks create strategic chokepoints. This project develops methods to reconstruct the multi-stage production networks of strategic industries from open data to map the full input-output structure from raw materials through processing stages to finished products. The reconstructed networks will then be stress-tested: how does the system respond when a specific node (a country, a firm, or a processing step) is disrupted? The project applies these methods to a set of emerging and strategically critical sectors and technologies, where production networks are evolving rapidly and geopolitical control is actively contested.
Systemic risk and tropical algebra
Systemic risk analysis on large networks, such as supply chains, is currently limited to dynamic, simulation-based – and thus computationally costly – approaches. We investigate max-plus (tropical) algebra as an ultra-fast computational framework for modeling systemic risk and failure propagation in complex networks. By reformulating discrete-event disruptions as linear tropical matrix operations, we aim to develop spectral algorithms to identify critical bottlenecks and compute cascade bounds without costly step-by-step simulations. New algorithms can be benchmarked using proprietary, real-world economic data against state-of-the-art CSH supply chain models.
Transforming business: agentic-AI testbed
Given recent advances in generative AI, businesses are transforming internal processes (e.g., marketing, controlling, user support) into agentic setups. The central challenge lies in selecting the most effective model for a given business task. This project develops a testbed in which users define business areas and tasks, provide ground-truth data samples, and automatically evaluate available AI models against these tasks. The expected result is a prototypical evaluation framework.