How TIGER Works
TIGER is a Python-based toolbox designed for evaluating graph vulnerability and robustness, providing a structured approach to analyze how networks respond to failures, attacks, or propagation events. It integrates multiple state-of-the-art methods to quantify network resilience, simulate disruptions, and test defensive strategies. The tool operates on graph-structured data, allowing users to assess structural weaknesses, simulate cascading failures, and model the spread of phenomena such as epidemics or information diffusion. By combining analytical measures with simulation capabilities, TIGER enables researchers to evaluate network performance under controlled conditions, ensuring that results are tied to specific operational questions rather than generic robustness scores.
The workflow begins with defining the study question, such as identifying critical nodes or edges that, when compromised, could disrupt network functionality. Users then select appropriate measures—such as connectivity, reachability, or spectral properties—to evaluate network vulnerability. TIGER supports both node and edge removal attacks, including targeted strategies like NetShield or PageRank-based attacks, as well as random failures. Defenses, such as node protection or edge rewiring, can be tested against these threats to determine their effectiveness. For propagation modeling, TIGER includes synchronous SIS (Susceptible-Infected-Susceptible) and SIR (Susceptible-Infected-Recovered) epidemic simulations, as well as influence models like independent cascade or threshold dynamics. Cascading failure simulations, such as Motter-Lai or Crucitti models, allow users to observe how localized failures propagate through the network, affecting overall efficiency and connectivity.
TIGER is built to handle both CPU and GPU acceleration, with optional CUDA support for performance-critical operations. The tool dispatches computations to the most efficient backend—whether CPU or GPU—based on the workload, ensuring compatibility with NetworkX graphs and CuPy for GPU-accelerated measures. Installation is modular, allowing users to opt into GPU support for CUDA 12 or CUDA 13 environments, with verification tools like tiger-gpu-status to confirm hardware compatibility.
The tool’s design emphasizes operational relevance, requiring users to specify failure mechanisms, attack budgets, and desired outcomes (e.g., largest connected component size or epidemic containment). This approach ensures that analyses are grounded in practical scenarios rather than abstract metrics. TIGER also supports capacity-aware cascade models, such as the Wei local load-sharing policy, which accounts for dynamic load redistribution during failures.
Key Use Cases
TIGER is primarily used in research and testing environments where graph robustness and vulnerability are critical. One of its core applications is in network security and resilience testing, where it helps identify critical infrastructure vulnerabilities. For example, utilities, transportation networks, or communication systems can use TIGER to simulate targeted attacks—such as removing high-betweenness nodes—to assess how quickly failures propagate and which defensive measures (e.g., rewiring or node protection) mitigate damage. The tool’s cascading failure simulations, such as the Motter-Lai model, are particularly useful for studying how localized overloads or congestion spread through power grids or traffic networks, potentially leading to systemic collapse.
Another major use case is in epidemic and information diffusion modeling. TIGER’s SIS and SIR simulations allow researchers to study how viruses, misinformation, or propaganda spread through social or contact networks. By varying parameters like infection rates, recovery probabilities, and intervention strategies, users can evaluate the effectiveness of containment measures, such as targeted node removal or vaccination campaigns. The tool’s support for competitive-message dynamics and voter models further extends its applicability to political science, marketing, and behavioral research, where understanding influence propagation is key.
TIGER is also employed in red teaming and adversarial testing, where attackers simulate real-world threats to identify network weaknesses. For instance, a red team might use TIGER to test how a network responds to a series of targeted edge removals or node attacks, comparing the results to random failure scenarios. Additionally, TIGER’s defense mechanisms—such as edge rewiring or capacity-aware load redistribution—enable researchers to explore proactive strategies for improving network robustness before disruptions occur.
In critical infrastructure protection, TIGER’s measures and simulations are used to assess the vulnerability of large-scale systems like power grids, water distribution networks, or the internet. For example, spectral radius and effective resistance metrics help identify networks that are highly susceptible to targeted attacks, while cascading failure models reveal how initial failures can trigger cascades that disrupt entire systems. By integrating these analyses with defensive strategies, TIGER provides actionable insights for policymakers and engineers aiming to harden networks against both natural and malicious threats.
Key Features and Capabilities
TIGER implements a comprehensive suite of vulnerability and robustness measures, each derived from peer-reviewed literature to ensure accuracy and relevance. These measures include vertex and edge connectivity, diameter, average distance, clustering coefficients, and spectral properties such as spectral radius, spectral gap, and natural connectivity. GPU acceleration is available for several measures, including spectral radius, spectral gap, and generalized robustness index, significantly speeding up computations for large graphs. For example, the spectral radius—linked to a network’s vulnerability to targeted attacks—can be computed efficiently on GPU-enabled systems, making it feasible to analyze networks with millions of nodes.
The tool supports diverse attack strategies, including node removal based on degree, betweenness, PageRank, or Eigenvector centrality, as well as edge removal attacks. TIGER also includes recalculated degree and betweenness attacks, where priorities are updated dynamically as the network degrades, providing a more realistic simulation of adaptive adversaries. Random removal attacks serve as baselines for comparison, helping users quantify the severity of targeted threats.
For defensive measures, TIGER allows users to test node protection, edge addition, and graph rewiring strategies. These defenses can be evaluated against fixed threat models to determine their operational effectiveness. For instance, rewiring edges to reduce path concentration or adding redundant connections can improve a network’s resilience to attacks. TIGER’s capacity-aware cascade models, such as the Wei local load-sharing policy, further enable users to simulate how networks redistribute load during failures, ensuring that defensive strategies account for dynamic stress conditions.
TIGER’s propagation and diffusion models include synchronous SIS and SIR epidemic simulations, as well as independent cascade, threshold, and competitive-message dynamics. These models are parameterized to reflect real-world processes, such as infection rates, recovery probabilities, and influence thresholds. For example, the SIS model can simulate how a virus spreads through a social network, with users able to test the impact of interventions like node removal or vaccination.
The cascading failure simulations in TIGER are based on established models such as Motter-Lai (permanent overload failures) and Crucitti (dynamic edge-efficiency congestion). These simulations help users understand how localized failures can escalate into systemic disruptions. For instance, the Motter-Lai model simulates node overloads where failed nodes redistribute their load to neighbors, potentially triggering further failures. TIGER provides metrics like largest connected component size and network efficiency to quantify the impact of cascades, with optional visualization of simulation snapshots or animations.
TIGER’s GPU acceleration is a key feature for handling large-scale graphs efficiently. Release 0.8.0 introduced CuPy implementations for spectral measures, connected-component analysis, NetShield attacks, and synchronous state-transition workloads. The tool automatically selects the optimal backend (CPU or GPU) based on the workload, ensuring compatibility with NetworkX and CuPy. Users can install GPU support for CUDA 12 or CUDA 13 via optional extras, with the tiger-gpu-status command verifying hardware and software compatibility. Benchmark suites included in the release ensure numerical and state parity between CPU and GPU results, guaranteeing reproducibility.
Documentation provides detailed guides on loading graphs, configuring simulations, and interpreting results, with examples covering robustness measures, attack types, defenses, and visualization.