Karin Rathsman, ESS

Causal learning challenge at ESS

ESS is a research facility under construct and commission in Lund. When finished in 2027 researchers from all over the world are welcome to bring their specimens to ESS for neutron scattering experiments. Since ESS is dedicated for external users, the demand on availability is a challenge. This talk will brief how machine learning applied to the ESS integrated control system could address operational challenges in general. In particular we will present a dataset with control system data from a cryogenics system at ESS, which is accessible through the WARA-OPS portal and intended for causal learning studies.

Mikael Lindberg, Saab-Kockums

IT-Operations aspects of submarine and surface ship design

The marine platforms of today are complex system-of-systems constructs with significant interactions between components, both local and remote. Operations while at sea must be carried out by crew, sometimes in very adverse conditions. This talk attempts to present some of the specific challenges this introduces in the designs and how we attempt to employ machine learning and AI to overcome them. CLICK HERE FOR PRESENTATION SLIDES.

Thomas Olsson, Bosch

Modern software engineering with LLMs

At the R&D Center in Lund, we develop software for the next generation mobility solutions. We work in international teams with a DevOps setup, having a highly automated and integrated toolchain for developing, testing, and deploying software. Our developers spend a lot of time analyzing log files, from various sources such as compilers, servers, test simulations and live products in the field. Analyzing log files is a time consuming and far from trivial task, that and can block the progress of development and deployment.

The Lund DevOps teams want to improve the tools around log file analysis by using LLM and Machine Learning. As Bosch is a truly international company with teams across the world, any changes to the toolchain typically requires a lot of education and planning, usually project-specific. In this prototype project, we therefore aim to integrate a log analyzer service as an unobtrusive advisor through a convenient chat interface. We have a large amount of log files for different systems, programming languages, etc.

Despite initial positive results, we are not yet sure how applicable our initial protype is to a wider context. We are now extending our initial work to further research how we can improve performance, compatibility, sensitive to drift, etc. .

Rickard Moller, Sony

Weirdness Detector

Finding and fixing software defects early lowers the cost and increases quality. Field trials during development of mobile phones generate large volumes of event logs. This work investigates an analysis technique that uses the collected data for early defect prevention and reduced need of manual testing.

In the past we have only used a tiny part of all these event logs. Most often only one kind of event has been analyzed at the time, after a defect has been found by other testing. It has also been practically impossible to check if seemingly unrelated series of events could give clues to why some problem occurred.

The analysis task is formulated as anomaly detection applied to a collection of time series. An LSTM autoencoder is trained on event logs collected from field trials of past software versions. It is then used to evaluate the present software version. A metric that measures the difference in behavior and a list of the most significant outliers (event sequences) are produced.

This is work in progress. In an initial experiment, we looked at a subset of the data (thermal events) and demonstrated the feasibility and potential benefit of the approach. More work is needed before conclusive results can be presented. CLICK HERE FOR PRESENTATION SLIDES.

Per Runeson, Lund University

Growing a data sharing community

With the introduction of machine learning to address various problems, gradually the supply of data for training and validation has raised as a major concern. While some data are very specific to a problem, other data become commodity. Still data are costly to clean, label and maintain. The idea of sharing data - and thereby share costs and benefits - emerges naturally from open source software, where code sharing communities have been established since several decades. This talk introduces Open Data Ecosystems, as a conceptual framework for data sharing communities. It is derived from empirical observations of emerging data sharing practices in different domains. Opportunities and challenges, particularly for operational data is address, relating this research to the establishment of WARA-Ops. CLICK HERE FOR PRESENTATION SLIDES.

Daniel Landström, Ericsson

Collecting live data from a 5G network

Great things grow out of passion, democratization of data is one major step towards letting people flourish through theirs!

Ericsson Garage Lund is an collaboration arena, focusing on enabling people to reach their next level, sharing knowledge and innovate. With the north star of being “The worlds kindest mobile phone operator”, Ericsson Garage Lund is operating their own 5G network, to be a playground for experimentation. The latest is addon is superior network observability and data collection, in the shape of a Sidecar. CLICK HERE FOR PRESENTATION SLIDES.

Vincent Hardion & Mirjam Lindberg, MAX IV

Towards the Excellence Operation of MAX IV Laboratory Control System

The MAX IV Laboratory, a leading-edge 4th generation synchrotron radiation facility located in Lund, Sweden, has been at the forefront of scientific research since its first operation in 2016. The facility can be described as a super microscope that reveals details about nature’s smallest building blocks. More than 2,000 visiting researchers annually conduct their experiments at MAX IV across its multiple beamlines and experimental stations

As the facility transitions from initial operations to a phase focused on operational excellence, the experiment control systems, crucial interfaces for user interactions, faces increasing complexities. Not only it has to accommodate the uniqueness of diverse experiments but also enhance reliability and performance to minimize operational interruption and maximize research quality. This presentation explores the evolution of MAX IV's operational strategies, particularly the integration of data-driven methodologies within the control systems. CLICK HERE FOR PRESENTATION SLIDES.

Fanny Söderlund, Ola Angelsmark, Advenica

Hidden in plain sight - Intrusion detection in systems logs

Advenica is a Swedish cybersecurity firm quietly safeguarding critical infrastructure, government agencies, and business enterprises alike. Given our role in network segmentation, we often find ourselves at the border between networks of differing levels of sensitivity, putting us in a great position to watch for suspicious activities. In this presentation we will look at how system logs can show evidence of intrusion attempts, hacker reconnaissance, and data exfiltration. CLICK HERE FOR PRESENTATION SLIDES.

Peng Kuang, Lund University

Developers' Perspective on Today's and Tomorrow's Programming Tool Assistance

Software development is a complex activity that needs a lot of tool assistance. Over the years there has been a lot of effort put into development of automated assistance to help with activities such as detection of issues via program analysis, or refactoring of code. Recently, the landscape of developer tool assistance is being disrupted with the entry of AI tools, such as co-pilot and ChatGPT, powered via large language models. Other kinds of tools assistance, for instance, gaze-driven assistance, is around the corner. What are programmers’ perceptions on tool assistance today? What do they see as good directions for the future? We present the results of a survey where we asked developers about their programming practices, experience with program analysis, and attitudes and views on enabling technologies, like AI and eye-tracking.

Matthias Wagner, Lund University

Deciphering the EU AI Act: Implications for Software Engineering and Industry Compliance

As AI technologies become increasingly integral to our lives, the European Union seeks to mitigate associated risks through the introduction of the AI Act, establishing harmonized regulations tailored to AI. This landmark legislation adopts a risk-based framework, imposing stringent requirements on high-risk and general-purpose AI systems alike. Considering this regulation, companies are expected to overcome the experimental nature of ML development if it is to be applied to production software. With the AI Act receiving the European Parliament's approval in March, this talk aims to dissect the legislation from a software engineering lens.

Chanh Nguyen, Umeå University

Enhancing the Resilience of Machine Learning Models in the Cloud

Machine Learning (ML) models serve as crucial tools in enabling intelligent decision-making within cloud systems management. However, the ever-evolving nature of operational data streams in cloud environments can introduce shifts in both historical training data and real-time inference data. These shifts often result in performance degradation of ML models over time, adversely affecting the efficiency and cost-effectiveness of cloud systems.

In this presentation, we will explore the challenges posed by this phenomenon, along with proposed solutions and preliminary research endeavors aimed at preserving the resilience of deployed ML models within cloud systems.

Konstantin Malysh, Lund University

Inter-organizational Data Sharing Processes - an exploratory analysis of incentives and challenges

Businesses across different areas of interest are increasingly depending on data, particularly for machine learning (ML) applications. To ensure data provisioning, inter-organizational data sharing is proposed, e.g. in the form of data ecosystems. The aim of this talk is to present our study in a form of exploratory investigation into the data sharing practices that exist in business-to-business (B2B) and business-to-customers (B2C) relations, in order to shape a knowledge foundation for future research. We launched a qualitative survey, using interviews with the aim of finding key practices, differences and similarities between approaches, so we could formulate the future research goals and questions.

If you're interested in getting involved, or just have questions about WARA-Ops, please get in touch!

WARA-Ops Co-Manager: Paul Townend
WARA-Ops Co-Manager: Johan Eker

 

Karin Rathsman, ESS

Causal learning challenge at ESS

ESS is a research facility under construct and commission in Lund. When finished in 2027 researchers from all over the world are welcome to bring their specimens to ESS for neutron scattering experiments. Since ESS is dedicated for external users, the demand on availability is a challenge. This talk will brief how machine learning applied to the ESS integrated control system could address operational challenges in general. In particular we will present a dataset with control system data from a cryogenics system at ESS, which is accessible through the WARA-OPS portal and intended for causal learning studies.

Mikael Lindberg, Saab-Kockums

IT-Operations aspects of submarine and surface ship design

The marine platforms of today are complex system-of-systems constructs with significant interactions between components, both local and remote. Operations while at sea must be carried out by crew, sometimes in very adverse conditions. This talk attempts to present some of the specific challenges this introduces in the designs and how we attempt to employ machine learning and AI to overcome them.

Thomas Olsson, Bosch

Modern software engineering with LLMs

At the R&D Center in Lund, we develop software for the next generation mobility solutions. We work in international teams with a DevOps setup, having a highly automated and integrated toolchain for developing, testing, and deploying software. Our developers spend a lot of time analyzing log files, from various sources such as compilers, servers, test simulations and live products in the field. Analyzing log files is a time consuming and far from trivial task, that and can block the progress of development and deployment.

The Lund DevOps teams want to improve the tools around log file analysis by using LLM and Machine Learning. As Bosch is a truly international company with teams across the world, any changes to the toolchain typically requires a lot of education and planning, usually project-specific. In this prototype project, we therefore aim to integrate a log analyzer service as an unobtrusive advisor through a convenient chat interface. We have a large amount of log files for different systems, programming languages, etc.

Despite initial positive results, we are not yet sure how applicable our initial protype is to a wider context. We are now extending our initial work to further research how we can improve performance, compatibility, sensitive to drift, etc. .

Per Runeson, Lund University

Growing a data sharing community

Talk abstract will be added shortly.

Daniel Landström, Ericsson

Collecting live data from a 5G network

Talk abstract will be added shortly. 

Vincent Hardion & Mirjam Lindberg, MAX IV

Towards the Excellence Operation of MAX IV Laboratory Control System

The MAX IV Laboratory, a leading-edge 4th generation synchrotron radiation facility located in Lund, Sweden, has been at the forefront of scientific research since its first operation in 2016. The facility can be described as a super microscope that reveals details about nature’s smallest building blocks. More than 2,000 visiting researchers annually conduct their experiments at MAX IV across its multiple beamlines and experimental stations

As the facility transitions from initial operations to a phase focused on operational excellence, the experiment control systems, crucial interfaces for user interactions, faces increasing complexities. Not only it has to accommodate the uniqueness of diverse experiments but also enhance reliability and performance to minimize operational interruption and maximize research quality. This presentation explores the evolution of MAX IV's operational strategies, particularly the integration of data-driven methodologies within the control systems

Fanny Söderlund, Ola Angelsmark, Advenica

Hidden in plain sight - Intrusion detection in systems logs

Advenica is a Swedish cybersecurity firm quietly safeguarding critical infrastructure, government agencies, and business enterprises alike. Given our role in network segmentation, we often find ourselves at the border between networks of differing levels of sensitivity, putting us in a great position to watch for suspicious activities. In this presentation we will look at how system logs can show evidence of intrusion attempts, hacker reconnaissance, and data exfiltration.

Konstantin Malysh, Lund University

Inter-organizational Data Sharing Processes - an exploratory analysis of incentives and challenges

Talk abstract will be added shortly.

If you're interested in getting involved, or just have questions about WARA-Ops, please get in touch!

WARA-Ops Co-Manager: Paul Townend
WARA-Ops Co-Manager: Johan Eker