Business Process Modeling as Complex Networks
In this early stage Project we want to translate the BPM notation into graph language in order to perform structural analysis of business processes together with classical functional analysis from Operations Research. The process network metrics will help us to the process analysis and optimization:
- Graph diameter: the maximum number of activities that must be carried out to end the processes.
- Shortest path: the minimum number of activities that must be carried out to end the processes.
- Average path length and clustering coefficient can give information about the parallelization and recursion degree of the processes.
- Indegree and Outdegree of the process components, identify the process sources and sinks, the ration between them influence the process controllability.
- Hubs: Critical activities.
- Clusters: Sets of very interrelated activities, will possible coincide with the designed process and can be used as a process design assessment tool.
- Structural attacks: removing critical activities would make the process collapse? Measure robustness of the process.
Corporate Sustainability World Map
The General Reporting Initiative (GRI) is the standard for Corporate Sustainability Reporting. The relations for a set of global enterprises interrelationships are built with the Mutual Information (MI) between enterprises (GRI) reporting data. We describe the enterprises interrelation from the MI topology using chord diagrams to represent the inter and intra connectivity between geographical regions and economic sectors globally, by continents and countries. Detailed maps are presented for Europe and Asia regions/sectors as well as the relationship for Europe and Asia top reporting countries. Our findings reinforce previous research about the role of Europe as a driver of sustainability and its influence worldwide. The GRI maps outline the behavior of the involved economic sectors, for all the studied regions. Also, a measure relating the inter to intra connections is presented in order to describe the relationship inside and between regions/sectors.
In the image below you can appreciate a schematic representation of the GRI network and identified clusters at microscopical level of enterprises. Human society is “good” at clustering, at right you see a geographical representation of the GRI map at continent scale. We can study the system inter and intra relations at different scales, continents, regions, countries.
Mapping the Global Offshoring Network through the Panama Papers
The aim is to map the offshoring network between regions and countries worldwide through the Panama Papers. The Panama Papers 2016 divulgence is the largest leak of offshoring and tax avoidance documentation. The leaked documents contain 2.6 Terabytes of information involving more than two hundred thousand of enterprises in more than two hundreds countries. Using the Offshore leaks database (https://offshoreleaks.icij.org) we related entities around the world through different types of relationships. These relationships were used in order to build an offshoring network at countries and geographical regions scales. The network will be characterized and described to map the intra and inter relation between the countries and regions, discovering which of them are the more prominent in the worldwide offshoring scenario.
Image: https://offshoreleaks.icij.org
Ensemble learning using Attractor Neural Networks
One of the main trends of research in Machine Learning, along with “deep learning”, is the study of “networks of networks”. One central question of interest is comparing a net of nets with overall connectivity of similar degree with only one single net with the same size and check their different performances in terms of storage/load, computational complexity, etc. Attractor networks as associative memory models have many desirable properties. As dynamical systems and associative memory systems the attractor networks can be used as denoising and pattern completion. They have been used in neuroscience to model for example spatial working memory and in Engineering applications such as fingerprint recognition and automotive traffic content retrieval systems. However, one of their drawbacks is limited capacity of storage. Using an Ensemble of diluted Attractor Neural Networks for pattern retrieval we have increased the network storage capacity by a divide-and-conquer approach of subnetworks. With the Ensemble approach we can deal with Engineering applications to limited memory systems: embedded systems or smartphones. Also, applications with intensive storage needs can be approached.
A Bandwidth Auction Mechanism to Enable Affordable Internet Access
Although technological developments have provided momentum to extend the frontier of commercially feasible network deployments, the latest Millennium Development Goals Report from the United Nations shows that the digital divide between the rich and poor is increasing due to the lack of affordable services. Therefore, an economic framework is needed to create conditions for affordable network services. In this paper, we introduce a set-aside mechanism that can satisfy this need by reserving resources for targeted groups and resolving the practical problem of having greedy users that rationally compete for cheaper resources. In this mechanism, prices are tailored to users’ budget capacities and quality needs. The simulation results indicate that it is possible to increase the resource allocation for delivering services to the poorest while simultaneously maintaining operators’ revenues by inducing users to discriminate among themselves.
Contact: Sergio Cabrales
Generalized additive model with embedded variable selection for bankruptcy prediction: The case of retail industry in Colombia.
This paper explores the properties of using a generalized additive model with embedded variable selection for the prediction of bankruptcy. During the last two decades, many machine learning and artificial intelligence algorithms have been implemented on this problem with the main purpose of provide higher prediction accuracy. However, there exists a duality between interpretation and prediction that have prevented the massive use of these novel methods. Using an additive model allows to incorporate nonlinear effects for each predictor enhancing the predictive power over classical linear models, but at the same time, keeping separated the marginal effects for interpretation. Also, an innovative penalization likelihood approach (GAMSEL, Chouldechova and Hastie (2015)) automatically selects important financial ratios and classifies them between linear and nonlinear effects, improving the interpretation of the estimations.
Contact: Carlos Valencia, Sergio Cabrales
