Network analysis concerns modeling the interactions of complex systems, such as those arising in biology, social science and software engineering, as networks (or graphs) and analyzing these models to understand the properties of the underlying application. Like all computations involving real-world systems, all steps of network analysis from data collection to modeling to analysis are influenced by experimental, subjection and computational choices. However, network analysis algorithms are primarily based on graph theory and therefore assume the inputs to be exact — not approximate.

In this talk, I will demonstrate using examples from biology, social networks and literature how user choices and computational limitations can affect network analysis results. I will discuss some of our ongoing work on ameliorating the effect of noise in network analysis and how concepts from numerical analysis such as conditioning and stability can be extended to evaluate the effect of noise in this domain.