About | Participate | Program | REGISTER

STEW Program

Thursday January 21st

09:00 Welcome
Introduction to the day.
Each participant introduces themselves shortly.
09:50 Break
10:00 AI-engineering: what is needed for a successful development and operation of AI-systems?
Helena Holmström Olsson, Professor at the Department of Computer Science and Media Technology, Malmö University
Artificial intelligence (AI) and machine learning (ML) are increasingly broadly adopted in industry. However, based on well over a dozen case studies, we have learned that deploying industry-strength, production quality ML models in systems proves to be challenging. Companies experience challenges related to data quality, design methods and processes, performance of models as well as deployment and compliance. We learned that a new structured engineering approach is required to construct and evolve systems that contain ML/DL components. In this presentation, we provide a conceptualization of the typical evolution patterns that companies experience when employing ML as well as an overview of the key problems experienced by the companies that we have studied. We define ”AI engineering” as a set of methods, processes and technologies that are necessary for a successful development and operation of AI-based systems. We present and discuss the key engineering challenges surrounding ML solutions and give an overview of open items that need to be addressed by the research community, and needed to be acquired by companies.
Visual software analytics: challenges and opportunities
Rafael M. Martins, PhD, Senior Lecturer in Computer Science and Media Technology, Linnaeus University
Software engineering and development are data-intensive activities that generate large scale, multidimensional, and heterogeneous data: text from requirements, bug reports, or developer communication, the complete history of code from versioning systems, and quality measurements from both code (static) and tests (dynamic) are just a few examples. The quick-growing discipline of software analytics has the goal of improving software engineering practices by analyzing and making sense of all of this data using machine learning, data science, and visualization techniques. In this presentation we focus on the last part: the use of visualization to support software analytics processes when fully-automated techniques reach their limits. Visual software analytics has been applied to the investigation of large-scale software systems from different structural, behavioral, and evolutionary perspectives, with the use of interactive interfaces between the developer/analyst and the systems under analysis. We will present an overview of the recent research, a few practical examples of the state of the art from both our own work and the research community in general, and a summary of currently open challenges and opportunities for the future. We hope to demonstrate the potential of the area and to spark interest and ideas from the audience.
Workshop
Presentation – Group findings
11:50 Lunch with networking opportunity
13:00 Explainable artificial intelligence for predictive maintenance
Slawomir Nowaczyk, Professor in Machine Learning, Halmstad University
Real-world applications of Predictive Maintenance are increasingly complex, with intricate interactions of many components. Artificial Intelligence solutions are a widespread technique in this domain, and especially the black-box models based on deep learning approaches show promising results in terms of predictive accuracy and capability of modelling complex systems. However, the decisions made by these black-box models are often difficult for human experts to understand – and therefore to act upon. The complete repair plan and maintenance actions that must be performed based on the detected symptoms of damage and wear often require complex reasoning and planning process, involving many actors and balancing different priorities. It is not realistic to expect this complete solution to be created automatically – there is too much context that needs to be taken into account. Therefore, operators, technicians and managers require insights to understand what is happening, why it is happening, and how to react. Today’s mostly black-box AI does not provide these insights, nor does it support experts in making maintenance decisions based on the deviations it detects. The effectiveness of the PM depends less on the accuracy of AI alarms than on the relevancy of actions performed based on them.
Intellectual property as a tool for constructing data openness
Pavel Kopylov, IP Business & Legal Consultant, Synergon AB
Creating openness requires established rules of the game for data sharing. When using external data sources organizations often experience uncertainty regarding legal status of the external datasets. Some data is released as public domain, some is shared under a myriad possible open source terms, some data is proprietary. In practice, this uncertainty triggers questions like: “Am I allowed to use this data?”, “What are the legal risks if I utilize this dataset for training my algorithms?”, “Am I allowed to modify this data?”. The presentation provides a practical view on the main intellectual property issues related to data. Understanding these concepts will assist in interpreting the legal implications of using external datasets as well as provide the guidelines for making your own data available to others.
Workshop
Presentation – Group findings
14:50 Break with networking opportunity
15:20 Making Program Analysis Useful Together
Emma Söderberg, Associate Senior Lecturer, Lund University
We need to consider many different aspects of a program in software development, for instance, security, maintainability, runtime behavior, performance, privacy, to name a few. Program analysis can help us manage this complexity by highlighting issues in our code, and perhaps even suggest improvements. However, program analysis is hampered by usability issues due to too many false positive results, poor integration into the developer workflow, and incomprehensible results. Developers many times lack trust in these tools or get overwhelmed by too many results. The effect is that developers end up actively ignoring the results from apparently useful tools. In this presentation, we will describe an approach that tackles these usability issues by gathering usage data in order to adapt how program analysis is presented to users. The presented approach has been realized in an extensible open-source system integrated into code review, and deployed in a pilot study at Axis Communications in a M.Sc. thesis project. During this presentation we will show examples of what this system looks in practice, along with encouraging initial results. We will also outline the vision for how a system like this can be used to build a community around useful program analysis.
Workshop – Group discussions on last presentation and recap from the day
Presentation – Group findings
End of day one

Friday January 22nd

08:50 Check- in – drink coffee and talk
09:00 Recap from yesterday
09:15 From beauty in code to beauty in data
Markus Borg, Senior Researcher, RISE Research Institutes of Sweden AB
In the book Your Code as a Crime Scene, we introduced the concept “Beauty in code”. Inspired by studies on human attractiveness – showing that average is beautiful – we argued that the source code analogy means “no surprises”. This leads to an actionable design principle: beautiful code shall have a consistent level of expression free from special cases. Enter machine learning (ML). As features now are trained rather than coded, the beauty concept must expand to cover also data. It turns out that a fundamentally different guiding principle must be embraced when developing an ML-based system. Average can no longer be the target – “Beauty in Data” requires diversity. In this talk, we will elaborate on the transition of the beauty concept from code to data. Finally, we will share some recommendations on how to support both types of beauty to allow symbiotic coexistence.
Open data and open innovation around roads
Thomas Olsson, Senior researcher,  RISE Research Institutes of Sweden
Open innovation – collaborative innovation and reducing cost by sharing assets – is practiced in open source software. With increasing reliance on and availability of data, open innovation communities around data are increasingly important. Such communities have the potential to unlock innovation areas when data from different sources are combined. In RoDL – Road Data Lab – we are building an open innovation platform, integrating different datasets to support open innovation to improve safety, quality, and planning around roads. We aim to provide both tools and methods to enable large-scale data analysis in general and machine learning in particular. RoDL will also address non-technical issues such as establishing an efficient governance and tackling the barriers to data sharing, such as legal and privacy concerns. We will talk about open data and open innovation, and specifically show examples of benefits and challenges from the RoDLs project partners and reference group. We report company needs and expectations on data sharing in RoDL, as well as results from our pilots currently running. The partners in the project are RISE Research Institutes of Sweden AB, AI Innovation of Sweden, Lund University, Zenuity, Mapillary, and Chalmers Industriteknik.
Workshop
Presentation – Group findings
11:05 Break with networking opportunity
11:35 The petabyte project
Sven Nilsson, Software Innovation Specialist, SAAB AB
Creating and maintaining functionality based on Machine Learning and Deep Learning (ML/DL) takes massive amounts of data. Everyone who tries to move into ML/DL eventually comes to the conclusion that they need more data to achieve the kind of results they want and need. This presents a number of surprising challenges: • How do you collect all that data? • How do you store all that data? • How do you share all that data? • How much data can you store in one place before it becomes sensitive? This is a presentation about how SAAB’s X Innovation Lab has approached these challenges to build a data lake of one petabyte of radar data to use as a foundation for building and maintaining future AI based radar functionality.
12:05 Lunch with networking opportunity
13:00 Workshop – Group discussions on last presentation and recap from the full conference
Presentation – Group findings
14:00 End of conference

Do you want to know more about STEW?

Please contact the secretariat.