Debarun Bhattacharjya

Value of information of seismic amplitude and CSEM resistivity

We propose a method for computing the value of information in petroleum exploration, a field in which decisions regarding seismic or electromagnetic data acquisition and processing are critical. We estimate the monetary value, in a certain context, of a seismic amplitude or electromagnetic-resistivity data set before purchasing the data. The method is novel in the way we incorporate spatial dependence to solve large-scale, real-world problems by integrating the decision-theoretical concept of value of information with rock physics and statistics. The method is based on a statistical model for saturation and porosity on a lattice along the top reservoir. Our model treats these variables as spatially correlated. The porosity and saturation are tied to the seismic and electromagnetic data via nonlinear rock-physics relations. We efficiently approximate the posterior distribution for the reservoir variables in a Bayesian model by fitting a Gaussian at the posterior mode for transformed versions of saturation and porosity. The value of information is estimated based on the prior and posterior distributions, the possible revenues from the reservoir, and the cost of drilling wells. We illustrate the method with three examples.

Track Geometry Defect Rectification Based on Track Deterioration Modelling and Derailment Risk Assessment

Analysing track geometry defects is critical for safe and effective railway transportation. Rectifying the appropriate number, types and combinations of geo-defects can effectively reduce the probability of derailments. In this paper, we propose an analytical framework to assist geo-defect rectification decision making. Our major contributions lie in formulating and integrating the following three data-driven models: (1) A track deterioration model to capture the degradation process of different types of geo-defects; (2) A survival model to assess the dynamic derailment risk as a function of track defect and traffic conditions; (3) An optimization model to plan track rectification activities with two different objectives: a cost-based formulation (CF) and a risk-based formulation (RF). We apply these approaches to solve the optimal rectification planning problem for a real-world railway application. We show that the proposed formulations are efficient as well as effective, as compared with existing strategies currently in practice.

Formulating Asymmetric Decision Problems as Decision Circuits

Decision analysis problems have traditionally been solved using either decision trees or influence diagrams. Although decision trees are better at handling asymmetry, prevalent in many reliability and risk analysis problems, influence diagrams can solve larger real-world problems by exploiting conditional independence. Decision circuits are graphical representations that combine the computational benefits of both graphical models. They are syntactic representations, i.e., they depict the summation, multiplication, and maximization operations required to solve a decision analysis problem. Previous work on decision circuits has focused on compiling them automatically from influence diagrams and describing the ways in which they can be used for efficient solution and sensitivity analysis. In this paper, we show how a decision circuit can be formulated directly, with or without the preprocessing of numbers that are assessed from the decision maker. By constructing two decision circuits for a nuclear reactor example, one using probabilities in inferred form and the other using probabilities in assessed form, we show how decision circuits generalize decision trees. The notion of coalescence is also made more explicit because computations for decision analysis can be saved and then reused in several ways. Because of their generality, decision circuits provide the analyst with a great deal of flexibility in problem formulation.

The Value of Information in Portfolio Problems with Dependent Projects

In the portfolio problem, the decision maker selects a subset from a set of candidate projects, each yielding an uncertain profit. When the projects in the portfolio are probabilistically dependent, further information regarding any particular project also provides information about other projects, and there is thus an opportunity to improve value through prudential information gathering. In this paper, we study the value of information in portfolio problems with multivariate Gaussian projects, analyzing the effect of parameters such as the expected values and standard deviations of profits from each project, the accuracy of the information, and dependence among projects. We are particularly interested in the role that dependence plays, illustrating the results using examples from the earth sciences, where there is spatial dependence among physical locations.

A Culinary Computational Creativity System

Compared to artifacts in expressive or performance domains, work products resulting from scientific creativity (including culinary recipes) seem much more conducive to data-driven assessment. If such products are viewed as an assembly of constituents that follow certain association principles, one could apply computationally intensive techniques to generate many possible combinations and use automated assessors to evaluate each of them. Assembly work plans for the selected novel products could subsequently be inferred from existing records. In this chapter, we report on our efforts to build a computational creativity system for culinary recipes. After gathering data and creating a knowledge base of recipes and ingredients, the system generates ingredient combinations that satisfy user inputs such as the choice of key ingredient, desired dish, and cuisine. Once a combination has been selected with the help of novelty and quality evaluators, the system further recommends ingredient proportions using a distributional conformance method and generates recipe steps using a subgraph composition algorithm. The time durations or efforts of atomic steps are estimated by solving an inverse problem from data on complete recipes. The example of culinary recipes could be generalized and applied to other scientific domains; manufacturing products and business processes could potentially follow a similar recipe for success.

From Reliability Block Diagrams to Fault Tree Circuits

Reliability block diagrams (RBDs) depict the functional relationships between components comprising a system, whereas Bayesian networks (BNs) represent probabilistic relationships between uncertain variables. Previous research has described how one can transform an RBD into a BN. In parallel, developments in the artificial intelligence literature have shown how a BN can be transformed into another graphical representation, an arithmetic circuit, which can subsequently be used for efficient inference. In this paper, we introduce a new graphical representation that we call a fault tree circuit, which is a special kind of arithmetic circuit constructed specifically for an RBD. A fault tree circuit can be constructed directly from an RBD and is more efficient than an arithmetic circuit that is compiled from the BN corresponding to that RBD. We develop several methods for fault tree circuits, highlighting how they can aid the analyst in efficient diagnosis, sensitivity analysis, and decision support for many typical reliability problems. The circuit framework can complement tools that are popular in the reliability analysis community. We use a simple pump system example to illustrate the concepts.

Towards Effective Business Process Availability Management

In this paper, we present a decision-support framework for managing the business process unavailability caused by the planned maintenance activities of the underlying infrastructure components such as software and hardware upgrades and patch application. The framework has two parts: first part provides an approach for estimating the business impact of business process outages. Four business impact factors are defined consisting of quantitative and qualitative metrics. Total business impact involves combining the individual impacts on business for various countries affected by the outages within a company. We apply multi-attribute utility theory to combine the values of impact factors. In the second part, optimal schedules for process maintenance activities are computed to minimize the overall estimated business impact using a sliding-window based scheduling algorithm. Finally, the schedules and their business impact are presented to the decision maker in a visual environment that is conducive for effective decision making. A prototype of the system has been developed and is currently being evaluated for use in a large Dow Jones (Top 10) company.

Using influence diagrams to analyze decisions in 4D seismic reservoir monitoring

Exploration and production decisions are complex due to several uncertainties: uncertainty in the geologic properties, seismic imaging, repeatability, reservoir structure, rock and fluid properties, etc. Furthermore, several decisions need to be made over the entire life cycle, and often it is not clear how current decisions might affect the future bottom line. In the exploration and characterization stage, more information is acquired about the reservoir structure, reservoir rock and fluid properties, and the spatial distributions of lithology, porosity, and saturations. Exactly how much information needs to be acquired is a critical decision at this stage. Later, decisions have to be made about the technical and economic feasibility of seismically monitoring the reservoir during production. Simple tools like decision trees have become popular (e.g., Newendorp and Schuyler, 2000) but complex situations demand the use of more sophisticated decision-analysis tools and their integration with the existing to...

Towards effective business process availability management

We present a decision-support framework for managing business process unavailability caused by the planned maintenance activities of the underlying infrastructure components such as software and hardware upgrades and patch application. The framework has two parts: the first part provides an approach for estimating the business impact of business process outages. Four business impact factors are defined, involving both quantitative and qualitative metrics. Total business impact involves combining the individual impacts on business for various countries affected by the outages within a company. We apply multi-attribute utility theory to combine the values of impact factors. In the second part, optimal schedules for process maintenance activities are computed to minimize the overall estimated business impact using a sliding-window based scheduling algorithm. Finally, the schedules and their business impact are presented to the decision maker in a visual environment that is conducive for effective decision making. A prototype of the system has been developed and is currently being evaluated for use in a large Dow Jones (Top 10) company.

Sequential information gathering schemes for spatial risk and decision analysis applications

Several risk and decision analysis applications are characterized by spatial elements: there are spatially dependent uncertain variables of interest, decisions are made at spatial locations, and there are opportunities for spatial data acquisition. Spatial dependence implies that the data gathered at one coordinate could inform and assist a decision maker at other locations as well, and one should account for this learning effect when analyzing and comparing information gathering schemes. In this paper, we present concepts and methods for evaluating sequential information gathering schemes in spatial decision situations. Static and sequential information gathering schemes are outlined using the decision theoretic notion of value of information, and we use heuristics for approximating the value of sequential information in large-size spatial applications. We illustrate the concepts using a Bayesian network example motivated from risks associated with CO2 sequestration. We present a case study from mining where there are risks of rock hazard in the tunnels, and information about the spatial distribution of joints in the rocks may lead to a better allocation of resources for choosing rock reinforcement locations. In this application, the spatial variables are modeled by a Gaussian process. In both examples there can be large values associated with adaptive information gathering.