Temporal graphs provide a robust framework for modeling and analyzing systems that evolve over time, offering unique advantages in addressing dynamic real-world challenges. In anomaly detection, they capture time-varying patterns to identify irregularities across domains such as finance, social networks, and cryptocurrency markets. For example, temporal graphs can highlight fraudulent transactions, uncover outliers in trading patterns, or detect sudden shifts in network behavior. In epidemic modeling, temporal graphs accurately represent evolving human contact networks, enabling more precise predictions of disease spread by accounting for temporal dependencies and behavioral changes, such as social distancing or vaccination. They also enhance forecasting in transportation systems like flight networks, tracking delays, cancellations, and disruptions over time to optimize scheduling and reduce congestion. By incorporating both structural and temporal dynamics, temporal graphs empower more realistic, data-driven approaches to solving complex, time-sensitive problems across diverse fields.
Selected Publications
Static graph approximations of dynamic contact networks for epidemic forecasting
In this work, we introduce new, more rigorous evaluation procedures for link prediction in dynamic graphs, addressing the challenges and real-world considerations that are often overlooked in static graph analysis. The authors propose tools to enhance the evaluation process, including new datasets, innovative negative sampling strategies, and a strong baseline model. These contributions aim to better compare the strengths and weaknesses of different methods, highlighting the importance of robust evaluation frameworks in advancing the field of dynamic graph learning
Fast and Attributed Change Detection on Dynamic Graphs with Density of States
We introduce a novel spectral method called Scalable Change Point Detection (SCPD) to address the limitations of current solutions in detecting anomalous change points in dynamic graphs. SCPD generates an embedding for each graph snapshot by efficiently approximating the distribution of the Laplacian spectrum, and it can capture shifts in node attributes by tracking correlations between attributes and eigenvectors. Through extensive experiments using synthetic and real-world data, the authors demonstrate that SCPD achieves state-of-the-art performance, is significantly faster than existing methods, can handle large quantities of node attributes, additions, or deletions, and effectively discovers interesting events in large real-world graphs
A Strong Node Classification Baseline for Temporal Graphs
Here we introduce a robust and effective baseline method for node classification in temporal graphs, serving as a benchmark for evaluating more complex models. This approach is characterized by its simplicity and strong performance in categorizing nodes based on their evolving characteristics within the graph. By addressing the unique challenges posed by temporal graphs, the method provides a clear point of comparison for researchers and practitioners, enhancing the understanding and development of advanced techniques in temporal graph analysis. Its relevance spans various applications, including link prediction and node attribute inference in dynamic networks.
SigTran: Signature Vectors for Detecting Illicit Activities in Blockchain Transaction Networks
SigTran is an efficient graph-based method for identifying illicit nodes in blockchain networks. SigTran constructs a graph from blockchain transaction records, representing nodes based on their structural and transactional characteristics to differentiate between legitimate and illicit activities. This method is versatile and can be applied to various blockchain networks.
Incorporating dynamic flight network in SEIR to model mobility between populations
In this paper, we introduce a significant enhancement to the standard SEIR (Susceptible, Exposed, Infectious, Recovered) epidemiological model by integrating dynamic flight networks to better capture the mobility of populations, particularly in the context of disease spread like COVID-19. The authors propose a modified model called Flight-SEIR, which accounts for the movement of individuals through air travel, thereby estimating imported cases based on air traffic volume and test positive rates. This modification allows for more accurate modeling of disease transmission between populations, enabling early detection of outbreaks, more precise estimation of the reproduction number, and better evaluation of the impact of travel restrictions. By incorporating real-world flight data, Flight-SEIR provides a more realistic and comprehensive approach to epidemiological modeling, crucial for navigating pandemics in an interconnected world.