Klaus Mueller

We are researchers from six different countries with diverse research backgrounds. We gathered at Schloss Dagstuhl to reflect on data visualizations for public health communication during the COVID-19 pandemic. There, we shared what we had learned and discussed our experiences visualizing COVID-19 data to gain useful insights into the role of data visualization in future pandemics. In this article, we highlight missed opportunities and considerations and provide recommendations. We learned that visualization practices should include cross-disciplinary collaborators with socio-technical expertise, consider personalization and audience more extensively, and use coordinated communication pathways to ensure effective, timely communication in future pandemics.

Blind and low-vision (BLV) users face significant barriers when encountering charts in digital content. While recent advances in large language models (LLMs) show promise for chart interpretation, existing accessible systems rely on pre-structured data formats, limiting applicability to real-world charts. We present GraphWhisper, a conversational system enabling BLV users to explore charts through natural language without requiring pre-structured data. GraphWhisper employs enhanced chart-specific prompting strategies that guide LLMs to systematically extract and analyze data directly from chart images, achieving 98.1% accuracy on visualization literacy benchmarks. A user study with 15 BLV participants demonstrated 94% question-answering accuracy across 185 questions and exceptional satisfaction scores (4.87-5.0 on 5-point scales), validating the system’s practical effectiveness for everyday chart accessibility.

Multivariate data contain an abundance of information and many techniques have been proposed to allow humans to navigate this information in an ordered fashion. For this work, we focus on methods that seek to convey multivariate data as a collection of bivariate scatterplots or parallel coordinates plots. Presenting multivariate data in this way requires a regime that determines in what order the bivariate scatterplots are presented or in what order the parallel coordinate axes are arranged. We refer to this order as a visualization sequence. Common techniques utilize standard statistical metrics like correlation, similarity or consistency. We expand on the family of statistical metrics by incorporating the rigidity of causal relationships. To capture these relationships, we first derive a causal graph from the data and then allow users to select from several semantic traversal schemes to derive the respective chart sequence. We tested the sequences with a crowd-sourced user study and a user interview to confirm that the causality-informed visualization sequences help viewers to better grasp the causal relationships that exist in the data, as opposed to sequences derived from correlations or randomization alone.

Misleading visualizations pose a significant challenge to accurate data interpretation. While recent research has explored the use of Large Language Models (LLMs) for detecting such misinformation, practical tools that also support explanation and correction remain limited. We present MisVisFix, an interactive dashboard that leverages both Claude and GPT models to support the full workflow of detecting, explaining, and correcting misleading visualizations. MisVisFix correctly identifies 96% of visualization issues and addresses all 74 known visualization misinformation types, classifying them as major, minor, or potential concerns. It provides detailed explanations, actionable suggestions, and automatically generates corrected charts. An interactive chat interface allows users to ask about specific chart elements or request modifications. The dashboard adapts to newly emerging misinformation strategies through targeted user interactions. User studies with visualization experts and developers of fact-checking tools show that MisVisFix accurately identifies issues and offers useful suggestions for improvement. By transforming LLM-based detection into an accessible, interactive platform, MisVisFix advances visualization literacy and supports more trustworthy data communication

Humans inherently connect certain colors with particular concepts in semantically meaningful ways that facilitate visual communication. These colors are known as semantically resonant colors. For instance, we associate with shades of blue, with red. In this paper, we investigate how language models, including Word2Vec, RoBERTa, GPT-4o mini and the vision language model CLIP generate and represent nuanced semantically resonant colors for diverse concepts. To achieve this, we utilized a large dataset of color names and concepts, tailored models for the structure of each language model, and developed an interactive web interface, CONCEPT2COLOR, as a use case. Additionally, we conducted experiments and a detailed analysis to assess the ability of these models to generate meaningful colors. Through these experiments, we examined how factors such as model design, training data and context affect the color output. Our findings reveal the capabilities and limitations of language models in processing and generating semantically resonant colors for concepts, thus contributing insights into how they depict semantic color-concept connections. These insights have implications for data visualization, design, and human-computer interaction, where leveraging effective semantic color generation can enhance communication and user experience.

This paper evaluates the visualization literacy of modern Large Language Models (LLMs) and introduces a novel prompting technique called Charts-of-Thought. We tested three state-of-the-art LLMs (Claude-3.7-sonnet, GPT-4.5-preview, and Gemini-2.0-pro) on the Visualization Literacy Assessment Test (VLAT) using standard prompts and our structured approach. The Charts-of-Thought method guides LLMs through a systematic data extraction, verification, and analysis process before answering visualization questions. Our results show Claude-3.7-sonnet achieved a score of 50.17 using this method, far exceeding the human baseline of 28.82. This approach improved performance across all models, with score increases of 21.8% for GPT-4.5, 9.4% for Gemini-2.0, and 13.5% for Claude-3.7 compared to standard prompting. The performance gains were consistent across original and modified VLAT charts, with Claude correctly answering 100% of questions for several chart types that previously challenged LLMs. Our study reveals that modern multimodal LLMs can surpass human performance on visualization literacy tasks when given the proper analytical framework. These findings establish a new benchmark for LLM visualization literacy and demonstrate the importance of structured prompting strategies for complex visual interpretation tasks. Beyond improving LLM visualization literacy, Charts-of-Thought could also enhance the accessibility of visualizations, potentially benefiting individuals with visual impairments or lower visualization literacy.

Policy-gradient methods such as Proximal Policy Optimization (PPO) are typically updated along a single stochastic gradient direction, leaving the rich local structure of the parameter space unexplored. Previous work has shown that the surrogate gradient is often poorly correlated with the true reward landscape. Building on this insight, we visualize the parameter space spanned by policy checkpoints within an iteration and reveal that higher performing solutions often lie in nearby unexplored regions. To exploit this opportunity, we introduce ExploRLer, a pluggable pipeline that seamlessly integrates with on-policy algorithms such as PPO and TRPO, systematically probing the unexplored neighborhoods of surrogate on-policy gradient updates. Without increasing the number of gradient updates, ExploRLer achieves significant improvements over baselines in complex continuous control environments. Our results demonstrate that iteration-level exploration provides a practical and effective way to strengthen on-policy reinforcement learning and offer a fresh perspective on the limitations of the surrogate objective.

Causal networks are widely used in many fields to model the complex relationships between variables. A recent approach has sought to construct causal networks by leveraging the wisdom of crowds through the collective participation of humans. While this can yield detailed causal networks that model the underlying phenomena quite well, it requires a large number of individuals with domain understanding. We adopt a different approach: leveraging the causal knowledge that large language models, such as OpenAI’s GPT-4, have learned by ingesting massive amounts of literature. Within a dedicated visual analytics interface, called CausalChat, users explore single variables or variable pairs recursively to identify causal relations, latent variables, confounders, and mediators, constructing detailed causal networks through conversation. Each probing interaction is translated into a tailored GPT-4 prompt and the response is conveyed through visual representations which are linked to the generated text for explanations. We demonstrate the functionality of CausalChat across diverse data contexts and conduct user studies involving both domain experts and laypersons.