CANDY: Cyberinfrastructure for Accelerating Innovation in Network Dynamics

The CANDY project's goal is to create the first parallel, scalable, extendable, and user-friendly software platform for updating important properties of dynamic networks and requisite functionalities to modify existing algorithms or create new ones. The main components of the CANDY framework correspond to three steps: (i) Sparsification to identify parts of the network likely to be affected by changes; (ii) Selection to evaluate each change whether it affects the property under consideration; and (iii) Update to apply necessary mechanisms for updating the property. CANDY to be implemented in C++ with MPI (distributed memory), OpenMP (shared memory), and CUDA or OpenCL (GPUs), will provide interfaces at the beginner, intermediate, and advanced levels. Major CANDY infrastructure components include: (1) a novel hierarchical taxonomy of network analysis algorithms that allows for the layered specification of the algorithms based on multiple parameters; (2) efficient methods to implicitly estimate the impact of change on different network properties; (3) innovative application of tree-decomposition methods to develop a generalized template for creating new scalable algorithms for dynamic network analysis; (4) novel evaluation metrics based on the invariant features of network properties, in order to resolve the problem owing to non-uniqueness of results; (5) novel performance metrics that measure the work based on relative the size of the network and the amount of change; and (6) a set of benchmarks from real-world applications and generators, that are representative of the variations in network dynamics.NSF Award Numbers: 2104076, 2104115, and 2104078