HOW INDUSTRY 4.0 RESHAPES THE WORLD: RECOMMENDATIONS BASED ON COMPLEX GRAPH NETWORK ANALYSIS

Industry 4.0 is a great opportunity and a tremendous challenge for every role of society. Our study combines complex network and qualitative methods to analyze the Industry 4.0 macroeconomic issues and global supply chain, which enriches the qualitative analysis and machine learning in macroscopic and strategic research. Unsupervised complex graph network models are used to explore how industry 4.0 reshapes the world. Based on the in-degree and out-degree of the weighted and unweighted edges of each node, combined with the grouping results based on unsupervised learning, our study shows that the cooperation groups of Industry 4.0 are different from the previous traditional alliances. Macroeconomics issues also are studied. Finally, strong cohesive groups and recommendations for businessmen and policymakers are proposed.

A Deeper Dive into Pattern-Aware Subgraph Exploration with PEREGRINE

Graph mining workloads aim to extract structural properties of a graph by exploring its subgraph structures. PEREGRINE is a general-purpose graph mining system that provides a generic runtime to efficiently explore subgraph structures of interest and perform various graph mining analyses. It takes a 'pattern-aware' approach by incorporating a pattern-based programming model along with efficient pattern matching strategies. The programming model enables easier expression of complex graph mining use cases and enables PEREGRINE to extract the semantics of patterns. By analyzing the patterns, PEREGRINE generates efficient exploration plans which it uses to guide its subgraph exploration. In this paper, we present an in-depth view of the patternanalysis techniques powering the matching engine of PEREGRINE. Beyond the theoretical foundations from prior research, we expose opportunities based on how the exploration plans are evaluated, and develop key techniques for computation reuse, enumeration depth reduction, and branch elimination. Our experiments show the importance of patternawareness for scalable and performant graph mining where the presented new techniques speed up the performance by up to two orders of magnitude on top of the benefits achieved from the prior theoretical foundations that generate the initial exploration plans.

Inference and Search on Graph-Structured Spaces

How do people learn functions on structured spaces? And how do they use this knowledge to guide their search for rewards in situations where the number of options is large? We study human behavior on structures with graph-correlated values and propose a Bayesian model of function learning to describe and predict their behavior. Across two experiments, one assessing function learning and one assessing the search for rewards, we find that our model captures human predictions and sampling behavior better than several alternatives, generates human-like learning curves, and also captures participants’ confidence judgements. Our results extend past models of human function learning and reward learning to more complex, graph-structured domains.

Some New Bounds of Weighted Graph Entropies with GA and Gaurava Indices Edge Weights

Motivated by the concept of Shannon’s entropy, the degree-dependent weighted graph entropy was defined which is now become a tool for measurement of structural information of complex graph networks. The aim of this paper is to study weighted graph entropy. We used GA and Gaurava indices as edge weights to define weighted graph entropy and establish some bounds for different families of graphs. Moreover, we compute the defined weighted entropies for molecular graphs of some dendrimer structures.

An Automated Model Reduction Method for Biochemical Reaction Networks

We propose a new approach to the model reduction of biochemical reaction networks governed by various types of enzyme kinetics rate laws with non-autocatalytic reactions, each of which can be reversible or irreversible. This method extends the approach for model reduction previously proposed by Rao et al. which proceeds by the step-wise reduction in the number of complexes by Kron reduction of the weighted Laplacian corresponding to the complex graph of the network. The main idea in the current manuscript is based on rewriting the mathematical model of a reaction network as a model of a network consisting of linkage classes that contain more than one reaction. It is done by joining certain distinct linkage classes into a single linkage class by using the conservation laws of the network. We show that this adjustment improves the extent of applicability of the method proposed by Rao et al. We automate the entire reduction procedure using Matlab. We test our automated model reduction to two real-life reaction networks, namely, a model of neural stem cell regulation and a model of hedgehog signaling pathway. We apply our reduction approach to meaningfully reduce the number of complexes in the complex graph corresponding to these networks. When the number of species’ concentrations in the model of neural stem cell regulation is reduced by 33.33%, the difference between the dynamics of the original model and the reduced model, quantified by an error integral, is only 4.85%. Likewise, when the number of species’ concentrations is reduced by 33.33% in the model of hedgehog signaling pathway, the difference between the dynamics of the original model and the reduced model is only 6.59%.

Inference and search on graph-structured spaces

How do people learn functions on structured spaces? And how do they use this knowledge to guide their search for rewards in situations where the number of options is large? We study human behavior on structures with graph-correlated values and propose a Bayesian model of function learning to describe and predict their behavior. Across two experiments, one assessing function learning and one assessing the search for rewards, we find that our model captures human predictions and sampling behavior better than several alternatives, generates human-like learning curves, and also captures participants’ confidence judgements. Our results extend past models of human function learning and reward learning to more complex, graph-structured domains.

Practical DDoS Attack Group Discovery and Tracking with Complex Graph-Based Network

In recent years, a large number of users continuously suffer from DDoS attacks. DDoS attack volume is on the rise and the scale of botnets is also getting larger. Many security organizations began to use data-driven approaches to investigate gangs and groups beneath DDoS attack behaviors, trying to unveil the facts and intentions of DDoS gangs. In this paper, DDoSAGD - a DDoS Attack Group Discovery framework is proposed to help gang recognition and situation awareness. A heterogeneous graph is constructed from botnet control message and relative threat intelligence data, and a meta path-based similarity measurement is set up to calculate relevance between C2 servers. Then two graph mining measures are combined to build up our hierarchical attack group discovery workflow, which can output attack groups with both behavior-based similarity and evidence-based relevance. Finally, the experimental results demonstrate that the designed models are promising in terms of recognition of attack groups, and evolution process of different attack groups is also illustrated.