KES International is an organization that provides conferences in the world and gets scientists and sometimes practitioners together. Last week there was multi-conference in Sorrento, Italy, consisting of four conferences:
- Agents and Multi-agent Systems – Technologies and Applications (AMSTA-15)
- Intelligent Decision Technologies (IDT-15)
- Intelligent Interactive Multimedia Systems and Services (IIMSS-15)
- Smart Education and E-Learning (SEEL-15)
I decided to attend the conference since with Big Data and HANA quantitative methods for decision making get more and more important in my opinion. I wanted to keep my knowledge up to date to learn about new methods in data analysis, modeling, prediction and drawing conclusion from them even in automated processes.
Jann Müller from SAP UK and me proposed a paper for KES-IDT15 and it was accepted. The conferences took part at the conference center in Sorrento in rooms without windows. The schedule was very tough and I took the chance to visit some talks of other conferences, f.e. about Flip teaching, Anthropic Oriented Computing – especially simulation of emotions in the computer and Interacting with music. Also very inspiring have been semantic methods for expert systems, where people managed to formalize problems, approaches and algorithms for those problems using Ontologies. This is in fact an approach that is also used (but not presented at this conference) by SAP f.e. in UX Explorer. In this conference researchers presented more advanced approaches like interaction and use of reason techniques to find appropriate answers – but the idea is the same since it’s all about making complex domain and information models accessible.
But my main focus was of course Intelligent Decision Making Technologies. The papers of this conference have been published by Springer in the series “Smart Innovation, Systems and Technologies”.
Intelligent Decision Making Technologies
In this conference Decision Making was presented in an interdisciplinary way and researcher from the following areas presented their work, mostly:
- Logic: logical foundations, algebra, abstract argumentation, rules…
- Business Intelligence: Detection of Rules, classifiers, classification of rare sets…
- Decisions in Financial Mathematics f.e. investments and risk management
- Pattern Recognition
From a mathematical point of view most of talks dealt with nonlinear and sometimes discrete models and optimization problems based on real-world use cases from many different areas.
The conference center was at Hilton Hotel in Sorrento and it was possible to attend talks of other conferences, too.
Rules Systems in Process Automation that can deal with ambigous data
Our contribution was about a model developed from a real world use cases. In a project I had to implement rule sets that classified scanned questionnaires of accident reports. When implementing those rule sets I learned that many of the accident reports contained contradicting data which is an endemic for all real world-data. Automatic processing relies on the highly standardized business objects like invoices and purchase orders but for other business cases it becomes challenging. The reason is that an accident isn’t standardized and if people start to describe it they tell a story and of course this story can be interpreted by an insurance expert or in the context of highly regulated domain of statutory health insurance (think of legal compliance) completely in a different way. And then there is always the problem when the scanning process fails of produces bad results.
So in fact our rule sets have been nearly as good as human beings since we also added data from the SAP backend systems. But ambiguity of data was still a problem. So I started to think about the following questions:
- Can we detect the cases where the rule sets will fail and can we improve them?
- Can we detect additional rules that can be applied as pre-checks when the data comes from a different channel like an online application?
- And last but not least: Even when the data is highly ambiguous can we improve the process by finding actions that should be started automatically in the insurance claims system?
The latter aspect is most interesting since different domain experts can disagree on a certain data set: one can say it is a staircase accident and the other one that it is a household accident for example. But there are cases even where experts disagree they share an (at least small) consensus and in this cases we can automate the whole business process by perhaps defining additional actions (like sending an email, starting workflows and so on). But how can we formalize this concept of consensus between different experts?
Here Jann Müller had a great idea and suggested to use Abstract Argumentation Theory to analyze the rule systems. At this time I was in contact with domain experts and when they gave me specifications for the rule systems which I implemented in BRFplus I gave feedback and also suggested some improvements. It was funny because at this time I wasn’t familiar with the business process but I drew my argument graphs and looked at conflicts. So mathematics helped me to understand and to improve rule systems. Together with Jann who is expert in this area due to his Ph.D. studies at the University of Belfast we developed a small theory combing description logics and argumentation and decided to publish it.
The main Challenge for Scientists: Where are the Data?
I was lucky in my business rules project since could work with (of course anonymized test data sets). Unfortunately most researchers don’t have access to those data. Often they use some of the very few published data sets or generated test data. I am worried about this:
- There are no possibilities to validate your model.
- Can you really draw conclusions about it? Supposed you have a model about investment decisions and compare it with a 5 year old data set from China and a European data set? I don’t believe that most models can deal with that since economy is quite different in both cases and many explicit and implicit variables of this model will differ dramatically.
So sometimes researchers presented beautiful models with no practical relevance and sometimes they seemed to be too complex models but there remained questions about their practical relevance In fact the researchers are well aware about this problem.
So in my opinion researchers should take the chance and get in contact with the industry to get real-world data sets and can do real experiments. I think this would lead to an improvement of the models as well and hopefully the solutions methods could be improved, too.
Last but not least: some facts about Sorrento
I liked the conference and I felt accepted as “industry guy”. I loved to spend three days in rooms without windows despite the beautiful surrounding. For the ones who want to stay and Sorrento I can give you some tips.
Lemon trees are everywhere:
Try out fresh lemon juice and the famous Limoncello.
Traffic rules are only suggestions (if they are relevant at all). This is the main road to the central bus station and used by everyone:
Learn to defend your meal! Guess what happens next here:
The “Capri sunset” is beautiful:
Big Data and HANA are a chance to make enterprise resource planning smarter. Therefore industry people like me have to refresh their skills. But also researchers especially in applied sciences should use this chance to learn about real world use cases.
Working on a scientific publication is hard and requires lots of work. But it is a great feeling to present it at the conference and to answer questions from the scientifical community. People from academia have a completely different point of view and will give you very important advice. The interdisciplinary approach of KES is really inspiring but following scientific talks is much more exhausting than learning well prepared lectures of SAP TechEd for example.
Last but not least: KES organizes their conferences at beautiful places and has social events in the evening. Nevertheless attending a conference is serious work and during the day you won’t see much of the location.