Plectics Laboratories

~/Research

Research Directions

Plectics Laboratories is dedicated to advancing interdisciplinary research that addresses fundamental questions in complex systems.

Physics of Algorithmic Complexity

This program investigates the fundamental relationship between data, algorithms, and learning. We use the lens of statistical mechanics, particularly the study of disordered systems and energy landscapes, to understand why certain high-dimensional models are effective. By analyzing phase transitions and the geometry of non-convex loss landscapes, we aim to uncover the core principles governing generalization and data compression.

Neural Representations

This program investigates the formation of neural representations in artificial and biological neural networks, with an emphasis on how structured information is encoded, transformed, and stored. We draw on insights from neurodynamics, information theory, and geometric and topological methods in data analysis to develop approaches that clarify the complex internal representations within neural systems.

Advanced Models for Social Sciences

This research focuses on designing novel computational frameworks for complex social systems. By employing agent-based modeling, network theory, and deep learning, we investigate how individual behaviors and group dynamics aggregate to shape large-scale social patterns, with the aim of contributing to the quantitative study of human culture.

Research Projects

Our research directions are put into practice through the specific projects highlighted below

Predicting Multiagent Behavior

As part of our inquiry in computational social sciences, we investigate how individual decisions aggregate into collective dynamics, by employing multi-agent reinforcement learning (MARL). Heterogeneous simulated agents run RL-based policies, continuously adapting to both environment and peers, with shifting alliances, deception, and competition. We represent their states and interactions as time-varying graphs, which we feed into advanced machine-learning pipelines to forecast system trajectories. With this framework we hope to provide some insight into drivers of emergent multiagent phenomena, such as market crashes, opinion polarization, subculture formation, and to deliver a novel way to benchmark predictive models at the cutting edge of computational social science.

[ Coming Soon → ]

Applied Sheaf Theory

Understanding complex systems demands a framework that handles heterogeneous and data and captures higher-order relationships. We tackle these challenges using sheaf theory, to rigorously track local information and specify when and how it can be integrated into a coherent global picture. Informed by these theoretical advantages, we actively leverage Sheaf Neural Network (SNN) architectures, with the aim of providing superior tools for interpretable modeling of complex systems.

[ Coming Soon → ]

Page updated

Google Sites

Report abuse