Obtaining overview of large text document data sets remains a challenging and important task across multiple research and application fields. In this paper, we extend our previously proposed prevalence-aware method for topic extraction and successfully apply it to a substantially larger data set than in the previously published examples. This data set has been constructed by scraping the abstract texts of articles from arXiv that contain the keyword ‘LLM’ (denoting large language models), allowing us to obtain insights on the development of this fast moving field of high research interest. We also introduce new functionality within our prototype visual analytics tool, which allows the analyst to combine the results from several different runs and search for common patterns. By extending the maximum size of the input corpus and by providing new strategies for selecting or compiling the best possible result, we position our methodology and tool as a promising candidate for many real-world analysis tasks and scenarios. The main goal is to provide the analyst with the best possible answer to the question “What are the most prevalent topics within this corpus?" and build the trust for the yielded results.

Recent advances in artificial intelligence (AI), data, and robotics (ADR) have pushed the boundaries of the benchmark performance of the respective methods and have already started to change the landscape in various application domains. Some of these domains are mission critical, with control activity that must match running processes. For those domains, a number of questions and challenges related to safety, robustness, and trustworthiness of AI and ADR methods and models still remain open, especially in the scenarios involving human operators. In this chapter, we provide an overview of human-centered perspectives on ADR with an emphasis on human-AI interaction, interactive visualization, and visual analytics. We explain the relationship of these fields to the related disciplines and fields, including human factors and human-computer interaction. We introduce the readers to basic concepts from these fields and discuss how the prior work fits with ADR principles, focusing on examples in visualization for explainable AI, cognitive systems engineering for joint human-AI control, and evaluation approaches for human-AI decision support systems. We argue that the techniques and frameworks proposed in these human-centered fields can and should be integrated with ADR methods.

In order to increase performance, resilience, and safety of critical infrastructures such as railway and air traffic, much attention has been recently paid to modern artificial intelligence (AI) methods and solutions. Data-driven methods, combined with large quantities of training data, have shown impressive results across multiple application domains. However, the unique challenges and high stakes associated with critical infrastructures comprise a rich sociotechnical context that must be taken into account at all stages of the AI solution lifecycle. Design and implementation of AI algorithms and models, validation of proposed decision-making assistant tools, and analyses of human-centered concerns among operators and stakeholders all constitute multi- and interdisciplinary challenges from both academic and applied perspectives.

AI for Real-World Network Operation (AI4REALNET) is an international project funded by the European Union’s Horizon Europe Research and Innovation programme and it involves 17 partners from academia and industry across 8 European countries. 

The project works on and across three critical domains with network operations: electricity grids operations, rail, and air traffic management. The goals of the project include design, development, and validation of AI solutions implemented with supervision and reinforcement learning approaches for several use cases across the three domains. The human-AI interaction scenarios in these use cases range from a digital assistant (e.g., providing advice to air traffic control operators on deviations to avoid an activated military area) to co-learning and to full AI-based control (e.g., fully automated rescheduling of railway operations). As opposed to purely technical performance evaluation of the AI solutions, validation goals of the project include sociotechnical aspects focusing on AI-human collaboration performance as well as attitudes of human operators towards such solutions.

In this presentation, we introduce the challenges and approaches of the AI4REALNET project with a particular focus on transport infrastructure applications (railways and air traffic management). We summarize the technical findings contributed by project partners, including AI models and digital environments for simulation and testing. Furthermore, we discuss in detail the ongoing work on identification and operationalization of key performance indicators relevant to sociotechnical aspects of human-AI interaction for AI-supported decision making in such critical infrastructures, including: social-technical decision quality; AI acceptability and trustworthiness; user experience; AI and human learning curves; task allocation balance; and long-term consequences of AI assistants.

Validation of human-centered concerns alongside technical, economical, and regulatory challenges will allow the contributions of the AI4REALNET project to not only advance the state of the art in the respective academic fields, but also to make an impact on operators and stakeholders in critical infrastructures, including railway and air traffic management.

Network data analyses addressing both the underlying graph structure and multivariate data attached to nodes and/or edges of such networks are important for a variety of research and application areas. Centrality analyses focus specifically on identification of important nodes in the network, with different and sometimes contradicting definitions of importance. To facilitate network analyses with multiple centrality algorithms, especially when applied to multivariate and temporal network data, interactive graph/network visualization approaches are key. The existing approaches typically do not address the requirement of supporting multiple centralities at once. In this paper, we report on extensions of our previous visual network analysis approach titled ViNCent, including a modern web-based implementation, support for multivariate and temporal network data, community analyses, and multiple interactive functions.

Research in ML4VIS investigates how to use machine learning (ML) techniques to generate visualizations, and the field is rapidly growing with high societal impact. However, as with any computational pipeline that employs ML processes, ML4VIS approaches are susceptible to a range of ML-specific adversarial attacks. These attacks can manipulate visualization generations, causing analysts to be tricked and their judgments to be impaired. Due to a lack of synthesis from both visualization and ML perspectives, this security aspect is largely overlooked by the current ML4VIS literature. To bridge this gap, we investigate the potential vulnerabilities of ML-aided visualizations from adversarial attacks using a holistic lens of both visualization and ML perspectives. We first identify the attack surface (i.e., attack entry points) that is unique in ML-aided visualizations. We then exemplify five different adversarial attacks. These examples highlight the range of possible attacks when considering the attack surface and multiple different adversary capabilities. Our results show that adversaries can induce various attacks, such as creating arbitrary and deceptive visualizations, by systematically identifying input attributes that are influential in ML inferences. Based on our observations of the attack surface characteristics and the attack examples, we underline the importance of comprehensive studies of security issues and defense mechanisms as a call of urgency for the ML4VIS community.

Industrial applications of Artificial Intelligence (AI) can be hindered by the issues of explainability and trust from end users. Human-computer interaction and eXplainable AI (XAI) concerns become imperative in such scenarios. However, the prior evidence of applying more general principles and techniques in specialized industrial scenarios is often limited. In this case study, we focus on designing interactive interfaces of XAI solutions for operators in the pulp and paper industry. The explanation techniques supported and compared include counterfactual and feature importance explanations. We applied the user-centered design methodology, including the analysis of requirements elicited from operators during site visits and interactive interface prototype evaluation eventually conducted on site with five operators. Our results indicate that the operators preferred the combination of counterfactual and feature importance explanations. The study also provides lessons learned for researchers and practitioners.

Electric power is the foundation of modern society, yet Europe is currently facing an energy crisis, increasing interest in power generation, energy infrastructure, and grid resilience. However, power plant data are complex and multidimensional, making it difficult to gain an overview or understanding. Visualization methods can help to reduce cognitive load and facilitate exploration of such data. In this paper, we propose EuroEnergyVis, a web-based visualization approach designed for the interactive exploration of power plant data across European countries. The design requirements were motivated by gaps identified in prior work. We conducted interviews with six domain experts in power systems and energy, which indicate that our tool enhances the user experience when exploring European power plants. Their reflections also suggest directions for future work.

Similarity-based analysis is a common and intuitive tool for exploring large data sets. For instance, grouping data items by their level of similarity, regarding one or several chosen aspects, can reveal patterns and relations from the intrinsic structure of the data and thus provide important insights in the sense-making process. Existing analytical methods (such as clustering and dimensionality reduction) tend to target questions such as "Which objects are similar?"; but since they are not necessarily well-suited to answer questions such as "How does the result change if we change the similarity criteria?" or "How are the items linked together by the similarity relations?" they do not unlock the full potential of similarity-based analysis—and here we see a gap to fill. In this paper, we propose that the concept of similarity could be regarded as both: (1) a relation between items, and (2) a property in its own, with a specific distribution over the data set. Based on this approach, we developed an embedding-based computational pipeline together with a prototype visual analytics tool which allows the user to perform similarity-based exploration of a large set of scientific publications. To demonstrate the potential of our method, we present two different use cases, and we also discuss the strengths and limitations of our approach.

As extreme weather events such as floods, storms, and heatwaves proliferate, local and regional authorities face challenges in predicting, monitoring, and assessing these events and their impacts. The introduction of impact-based warning services requires detailed, location-specific information on local vulnerability and impacts. This necessitates complementing conventional data with insights from local actors, and to explore novel methods for relevant public data monitoring through social media and news outlets. This paper presents a visual analytics pipeline that was co-developed with practitioners, aiming to detect impacts of extreme weather events, particularly floods, using Volunteered Geographic Information (VGI). The pipeline steps include: collecting VGI from social media, classifying and analysing the data, and visualizing it through an interactive interface. An empirical evaluation study was performed with meteorological and hydrological experts to assess the developed visual interface. The study collected and analysed feedback on the usability of the interface and identified interaction patterns from the experiment’s screen recordings.