研究実績の概要 |
Our research achievements for this period are mainly focused on defining the basic principles of efficient algorithms and method to compute flow of data and knowledge in classical and information networks. Two major problems were encountered that are common to all the network considered, one being the size of available data and the other presence of cycles and strongly connected components in real life networks. Our existing framework to compute a fair and efficient metric on information networks assumed that those networks were acyclic. However, such an assumption is rare in real life information networks. We first consider the case of pseudo directed acyclic graph (PDAG). For instance, traditional publication networks can be considered PDAG whereas pre-print publication networks such as arXiv have relatively large strongly connected components. We designed and analyzed a set of procedures called “decycling” to reproduce the behavior of the transmission of knowledge in PDAG as if the graph was acyclic. Those procedures can be extended to evaluate accurately or approximately the transmission of knowledge in the general case under certain conditions (approximation rate, order on the network links or nodes such as versioning). With the aim to improve the computation of classical algorithm to explore graphs and search spaces, we extended two existing concepts. First, we extended our Compressed Stack algorithms framework to compressed dequeue and lists. Second, we introduced the concept of limited memory structure for geometric algebra to use in (very) high-dimension context.
|
今後の研究の推進方策 |
Implementation or finalization, and evaluation of afore-mentioned achievements, if not already available: flow of knowledge on PDAG and general acyclic networks; general black-box algorithm for compressed data structure such as stacks, deques, lists and graphs; compressed search structure for optimization solver; smart and light geometric algebra implementation in high-dimension context. Completion of the Network Interdiction framework for classical network flow using various optimization techniques adapted to small and large networks. Extension of this framework to the flow of knowledge in information networks with the aim of discover weak links and possible fake news, in particular in dynamic environment.
|