IBO – Intelligent Business Optimization
Agile Team “Intelligent Business Optimization”
|Partner (Fraunhofer IIS)
|Prof. Dr. Alexander Martin, Research Group: Analytics & Mixed Integer Optimization
|Prof. Dr. Freimut Bodendorf, Research Group: Management Intelligence Services
|Qualified partners from industry and research
|Dr. Andreas Bärmann
A clear potential for business improvement can be found in combining the approaches of artificial intelligence, in particular machine learning, with optimization approaches. Descriptive methods look into the past and analyze what happened, how, when, and for what reason. Predictive approaches go in the opposite direction in time, applying methods to understand the future and describe what will happen. In the area of optimization, prescriptive approaches use mathematical methods and simulations to model possible scenarios, to find better or best solutions, and to make recommendations for action on this basis.
Within the agile “Intelligent Business Optimization” team, a symbiosis between the research areas of machine learning and mathematical optimization leverages the mutual capabilities, combining both areas in research projects and use cases.
The keywords with a color background identify essential tasks in the data analysis and optimization pipeline that describe “what” is to be achieved. The keywords with a white background outline important activities for processing the tasks and thus “how” the goals are pursued. In the Agile Team, the focus is on the “Analytics”and “Optimization” phases. The preliminary phases of “Data Governance” and “Data Foundation” must be taken into account and the “Trust” phase in particular is of great importance for the acceptance of the analysis and optimization results, e.g., by managers who base decisions on them.
Data Governance is comprised of processes and policies relevant to the collection, sharing, quality and security of data used in an organization.
Data Foundation ensures that the data necessary for the Analytics and Optimization stages are in the required form, and that the needed hardware and software infrastructure is in place.
Insight Engineering and Method Design & Engineering – The combination of analytics and optimization methods is the primary research goal of the Agile Team. Intelligent analytics methods and advanced optimization algorithms and models build on a data foundation that is intended to facilitate or promote the symbiosis of analytics and optimization. In this respect, the combination of these two research fields is also an important challenge for the Data Governance and Data Foundation stages. Intelligent analytics methods use machine learning, deep learning, agent-based modelling, natural language processing, text mining and process mining for descriptive, diagnostic and predictive analytics. These approaches aim to gain insights from large amounts of empirical data. The ultimate goal, however, is to develop new methods on the basis of this knowledge, which calculate solutions for operational tasks. For example, the company’s restrictions, together with the analytics and predictions obtained through insight engineering, are to be translated into a solution space in which optimizations are carried out using mathematical processes. Conversely, the results of the optimization models can be validated by heuristics from business analytics. In this way, “Optimization Intelligence” approaches learn through the heuristics of “Business Analytics” and compute solutions in a solution space that is specified using advanced mathematical modeling including realistic limitations.
On the one hand, a detailed modeling of relevant practical relationships often results in very extensive models, the creation of which is usually time-consuming and, on the other hand, often lies at the limit of the modeling possibilities of mixed-integer programs. Furthermore, such detailed models are usually also difficult to solve, for example when they explicitly include physical effects that are described by differential equations. An interesting research question arises from integrating such complex or even unknown relationships into the model in a data-driven way. It is precisely at this point that the information obtained through “AI engineering” could be used successfully to adequately describe the solution space of the problem in a different way.
Trust is a final essential task in connecting “Optimization Intelligence” and “Business Intelligence”. The aim is to promote or ensure the comprehensibility and traceability of the developed intelligent approaches, methods and their results (“Transparency Engineering” in Figure 1). This topic is related to research in the field of Explainable Artificial Intelligence (XAI). XAI has established itself as a research topic in the context of complex models known as black-box models. This means that neither the developers nor the decision makers can fully understand and explain the reasoning behind the analysis and optimization results of the model. The main problem is finding a trade-off between performance and explainability.
The core competencies of the research group “Analytics & Mixed-Integer Optimization” (Prof. Martin) lie in the development of mathematical optimization models for real-world problems, their theoretical analysis, and the implementation of efficient algorithms to solve them. Special focus is put on the optimization of different types of networks, such as energy supply or transport networks. Further research activities are dedicated to the analysis and solution of scheduling problems in various domains e.g., finding optimal train schedules, processes in air traffic management, healthcare, or production planning. Since a few years, the group has also been working on machine learning problems such as approaches for combining optimization and online learning.
The “Management Intelligence Services“ research group (Prof. Bodendorf) has been active for almost 30 years, including in many industry collaborations, in the areas of machine learning, agent-based simulation and network-based analytics. These methods are used in companies and organizations in the healthcare sector to support strategic decisions by top management. Managers in the consumer goods sector are consulted on improving production processes or product development using text mining and process mining. In e-commerce, in addition to other methods, causal analytics as a combined approach consisting of graph models, statistical methods and deep learning, are currently supporting marketing and sales decision-makers, e.g., when evaluating and planning advertising campaigns.