ExplainabilityTransparencyAccountabilitycs.AI
Interactive dense pixel visualizations for time series and model attribution explanations
Udo Schlegel, Daniel A. Keim
August 27, 2024
MLVis@EuroVis
3 citations
Key Takeaways
-
Interactive visualization tools can make AI explanations more accessible: DAVOTS demonstrates that dense-pixel visualizations allow non-experts to explore how neural networks make decisions on time series data by showing raw data, network activations, and attribution explanations side-by-side, making the "black box" more transparent.
-
Evaluating AI explanations requires specialized approaches: The paper highlights that standard metrics for assessing explanation quality are often inadequate, particularly for complex data types like time series, suggesting that responsible AI practices need domain-specific evaluation methods beyond generic metrics.
-
Clustering and organization strategies help identify model patterns: By applying clustering to visualized data and presenting multiple ordering strategies, the approach helps users discover whether models are learning meaningful patterns or relying on spurious correlations—a key concern for trustworthy AI deployment.
-
Visual analytics can democratize model understanding: The interactive nature of DAVOTS allows stakeholders without deep technical expertise to investigate model behavior themselves, supporting more informed decision-making about whether and how to deploy AI systems responsibly.
Abstract
The field of Explainable Artificial Intelligence (XAI) for Deep Neural Network models has developed significantly, offering numerous techniques to extract explanations from models. However, evaluating explanations is often not trivial, and differences in applied metrics can be subtle, especially with non-intelligible data. Thus, there is a need for visualizations tailored to explore explanations for domains with such data, e.g., time series. We propose DAVOTS, an interactive visual analytics approach to explore raw time series data, activations of neural networks, and attributions in a dense-pixel visualization to gain insights into the data, models' decisions, and explanations. To further support users in exploring large datasets, we apply clustering approaches to the visualized data domains to highlight groups and present ordering strategies for individual and combined data exploration to facilitate finding patterns. We visualize a CNN trained on the FordA dataset to demonstrate the approach.