Dr. Thomas Hußlein is the managing director of OptWare GmbH.
OptWare GmbH was founded in 1999 to support companies with modern optimization methods in improving and restructuring their processes. Today the company successfully combines instruments of mathematical optimization with engineering and the most advanced technology. Their goal is to find the optimal solution for every customer for a wide variety of problems.
Together with Thomas Pfadenhauer, I manage the OptWare GmbH in Regensburg, which we founded more than 20 years ago. In addition to the administrative activities as managing director, I mainly undertake tasks in sales and manage the areas of mathematical modeling and optimization (MMO) as well as planning software and operations (PSO).
In my role in sales, together with our customers, I investigate new ways of mastering existing and future challenges using quantitative methods and data-driven planning and control. Usually we have to delve deeply into the technical processes in order to find the causes of severe (entrepreneurial) pain, which we then heal in projects.
In the solutions that we implement together with our customers, we transfer sophisticated mathematical optimization algorithms and state-of-the-art software technologies into the productive systems and continuously generate benefits.
In the solutions that we implement together with our customers, we transfer sophisticated mathematical optimization algorithms and state-of-the-art software technologies into the productive systems and continuously generate benefits.
We develop individual solutions for our customers – e.g. automobile manufacturers and their suppliers, machine and plant manufacturers as well as representatives from the financial and pharmaceutical industries – using methods from mathematical optimization and data science. For many problems we use huge amounts of data, mostly from different sources and in completely different formats. In addition, most problems are inherently extremely complex, so that we have to fall back on the wide range of methods of current research on AI and optimization.
AI is currently finding its way into all corporate processes in numerous facets. Occasionally, AI is reduced to a few methods such as artificial neural networks, but we see a much larger field of applications and methods. Already today, decisions are intelligently prepared or even made autonomously by algorithms at many points in companies. For example, we develop sequencing algorithms that automatically determine the optimal sequence of workpieces and write the results back to the productive systems of our customers. Just recently, a previous customer told us that an algorithm we had developed has been in continuous use for eight years. Thus, a central task of operational planning was completely taken over by the machine.
We like to use AI in the stricter media sense, i.e. with approaches such as artificial neural networks or RandomForest, for forecasts and predictive maintenance. Most recently, we predicted certain aspects of engine behavior and were extremely close to the values that occurred later. In addition to algorithms, there are a number of other challenges in this area of AI. For example, we usually use very large amounts of data and process them. The variety of systems, concepts and formats and their continuous rapid further development requires that we always have to keep our skills and tools for data handling and software development up to date in order to maintain our competitiveness.
AI is currently finding its way into all corporate processes in numerous facets. Occasionally, AI is reduced to a few methods such as artificial neural networks, but we see a much larger field of applications and methods. Already today, decisions are intelligently prepared or even made autonomously by algorithms at many points in companies. For example, we develop sequencing algorithms that automatically determine the optimal sequence of workpieces and write the results back to the productive systems of our customers. Just recently, a previous customer told us that an algorithm we had developed has been in continuous use for eight years. Thus, a central task of operational planning was completely taken over by the machine.
We like to use AI in the stricter media sense, i.e. with approaches such as artificial neural networks or RandomForest, for forecasts and predictive maintenance. Most recently, we predicted certain aspects of engine behavior and were extremely close to the values that occurred later. In addition to algorithms, there are a number of other challenges in this area of AI. For example, we usually use very large amounts of data and process them. The variety of systems, concepts and formats and their continuous rapid further development requires that we always have to keep our skills and tools for data handling and software development up to date in order to maintain our competitiveness.
OptWare was originally founded as a spin-off from the University of Regensburg and is still well connected with the university. We collaborate in particular with the Business Informatics sector in the form of internships and bachelor/ master theses as well as with Biophysics in the field of neural networks.
In addition, we are very closely connected with the OTH Regensburg, especially with the Faculty of Computer Science and Mathematics. The collaboration includes internships, bachelor/ master theses and doctorates. I also support the practical exchange by taking on the lecture on Operations Research in the computer science course of study.
In addition, we are very closely connected with the OTH Regensburg, especially with the Faculty of Computer Science and Mathematics. The collaboration includes internships, bachelor/ master theses and doctorates. I also support the practical exchange by taking on the lecture on Operations Research in the computer science course of study.
We follow the latest trends of knowledge and tools. We can easily try out new ideas through various theses such as Bachelor, Master or PhD. In addition, the close contact with the students often creates the opportunity to get to know future employees at an early stage.
Artificial intelligence has already become an integral part of the planning, processes and products of many companies. Today’s students will work with Artificial Intelligence at all levels of their professional lives. Just as the use of a pocket calculator was inevitable 50 years ago or how Excel, Word and Power Point became standard in many companies more than 20 years ago, interaction with autonomous and self-learning systems is expected to be just as natural in the future. Students should be prepared for how to use these new tools optimally. They should ask the right questions to the systems and be able to take a constructively critical attitude towards the results. Furthermore I think it is essential to deal universally with the possibilities, but also the limits of artificial intelligence, today.
We deal regularly, also currently and in the future, with fundamental topics from optimization and data science as well as current research results on efficient software development. We are always open to inquiries about cooperation.
Current work is subject to confidentiality, which we are unfortunately not (yet) allowed to talk about in this context.
Current work is subject to confidentiality, which we are unfortunately not (yet) allowed to talk about in this context.
We have positioned ourselves very broadly in research.
Over the past few years, I have published a number of research articles in recognized specialist journals together with former colleagues. In addition to basic problem solving methods, we mostly deal with the implementation of the algorithms in corporate planning. Fundamental research on implementation is imperative, because even the most powerful algorithms can only be used effectively if the company processes are coordinated with them.
Last but not least, our wealth of knowledge in numerous theses: I would like to highlight two papers on artificial intelligence from recent years. In a master’s thesis that was supervised jointly with the University of Regensburg, we examined deep learning approaches in supply chain management. In a recently completed bachelor thesis in conjunction with the OTH Regensburg, we achieved very good results with a reinforcement learning approach to a scheduling problem.
Over the past few years, I have published a number of research articles in recognized specialist journals together with former colleagues. In addition to basic problem solving methods, we mostly deal with the implementation of the algorithms in corporate planning. Fundamental research on implementation is imperative, because even the most powerful algorithms can only be used effectively if the company processes are coordinated with them.
- M Feldmeier, T Husslein Backbone Strategy for Unconstrained Continuous Optimization, 2017, ECMS, 529-533
- T. Husslein, J. Breidbach, Anwendung und Anwendbarkeit von Optimierungsalgorithmen in der Praxis in: Produktionsplanung und –Steuerung, Springer, 227-239 (2015)
- T. Husslein, C. Danner, M. Seidl, J. Breidbach, W. Lauf Deriving A Mathematical Model Of A Paint Shop From Data Analysis ECMS 2014: 670-675
- S. Müller, C. Danner, J. Breidbach, M. Seidl, T. Hußlein, Wolfgang Lauf Stochastic Modeling of Throughput Times of Supplier Parts for an Automotive Plant Prozesse, Technologie, Anwendungen, Systeme und Management (2014), 35-47
- J. Schneider, C. Froschhammer, I. Morgenstern, T. Husslein, J. Singer Searching for backbones — an efficient parallel algorithm for the traveling salesman problem Comp. Phys. Comm.: 2-3/96 (1996) 173-188.
Last but not least, our wealth of knowledge in numerous theses: I would like to highlight two papers on artificial intelligence from recent years. In a master’s thesis that was supervised jointly with the University of Regensburg, we examined deep learning approaches in supply chain management. In a recently completed bachelor thesis in conjunction with the OTH Regensburg, we achieved very good results with a reinforcement learning approach to a scheduling problem.
I have always been fascinated by how decisions and the decision-making process can be mapped in a computer. I’ve been dealing with this since I was 16 years old. I was taught the basics of neural networks and their connection to optimization algorithms during my studies in Computational Physics. In my further career at the IBM Watson Research Center and the University of Pennsylvania, I dealt intensively with AI, which ultimately led to the founding of OptWare and further in-depth work on the topic of optimization.
At OptWare I have repeatedly given impulses for the productive use of AI. The personal supervision of master’s theses as well as the contact with the University of Regensburg offer me excellent opportunities for a deep insight into current research.
Question 10: Do you personally find it easy or difficult to deal with topics that are related to AI?
Basically, the access to current AI topics is relatively easy for me due to my theoretical background and many years of application in a company context. However, AI is developing extremely quickly in many directions, so that individuals can hardly keep track of it. It therefore seems almost impossible to me to be a “universal expert” in all systems at the same time.
Therefore, we distribute the diverse know-how on AI to our employees in OptWare in order to stay up-to-date in terms of breadth and depth.
As already mentioned above, the theoretical basics and the fascination have been with me for several decades, especially the mathematical approaches how to learn from data and generate knowledge. For some years now, hardware and development options as well as the understanding in companies have matured to such an extent that AI can be implemented and productively used with a manageable project scopes. With the help of AI, we are now developing new and, above all, holistic solutions for problems on which individuals previously have failed because of their complexity or size. This makes it possible to objectify the decision-making process in companies and even to automate them in parts.
This is confidential information, which we are unfortunately not allowed to talk about in this context.
We are currently encountering the topic of AI almost exclusively among mathematics, physics and computer science students as well as in some ethical discussions. I would like the universities to contribute to creating a broad understanding of AI in all courses and to promote cooperations with AI. Understanding breaks down prejudices and allows it to interpreted the results appropriately. Ideally, students recognize and accept the strengths of the machine in some areas and understand how to profitably link these with their own – human – strengths.
In addition to technical aspects and the possibilities of economic application, students should also be able to discuss the ethical, legal and political implications of a comprehensive introduction of artificial intelligence. In general, I think that the “dangers” and changes from and through AI decisions must be communicated as early as possible. In addition, the legal framework for the transfer of responsibility from a decision by humans to a decision by AI must finally be provided.
In all discussions, everything stands and falls with the correct collection and processing of the training data as well as the consistently critical questioning of the models and approaches. And so that discussions between all disciplines involved can be conducted on an equal footing, the universal advice applies: Keep it simple!
In addition to technical aspects and the possibilities of economic application, students should also be able to discuss the ethical, legal and political implications of a comprehensive introduction of artificial intelligence. In general, I think that the “dangers” and changes from and through AI decisions must be communicated as early as possible. In addition, the legal framework for the transfer of responsibility from a decision by humans to a decision by AI must finally be provided.
In all discussions, everything stands and falls with the correct collection and processing of the training data as well as the consistently critical questioning of the models and approaches. And so that discussions between all disciplines involved can be conducted on an equal footing, the universal advice applies: Keep it simple!
I am very enthusiastic about the interdisciplinary research approach, the contact with practice and of course the involvement of students. I see a lot of overlap between my personal interests and the work of OptWare. I take part with this interview very directly. But in the future, I will look forward to working with “Knowledge Creates² Data”.
For several years now, I have been observing that the barriers to entry to develop your own AI have fallen rapidly. For example, image recognition (classification) can be tried out with numerous online tools without programming. But the programming itself is also greatly simplified. In corresponding courses at the university, on online learning platforms or, of course, in the crash courses of the “Knowledge creates2 data” project, an introduction to the relevant languages, above all R and Python, is given in a very short time. With the help of freely available libraries for R/Python, even beginners can train their own artificial intelligence for an individual problem with a few dozen lines of code and thus possibly get a worthy opponent for their favorite board game.
Finally, my second (and last) advice in this interview: Just give it a try!
Thank you for taking the time to answer these questions. We wish you a nice day!
Finally, my second (and last) advice in this interview: Just give it a try!
Thank you for taking the time to answer these questions. We wish you a nice day!