Explanation Methods for Sequential Data Model - From Post-hoc to Interpretable-by-design Approaches for Time Series Classification

The increasing availability of high-dimensional time series data, such as electrocardiograms, stock indices, and motion sensors, has led to the widespread use of time series classifiers in critical fields like healthcare, finance, and transportation. However, the complexity of these models often makes them black boxes, hindering interpretability. In high-stakes domains, explaining a model’s decisions is vital for trust and accountability. Effective eXplainable AI (XAI) methods for sequential data are essential for providing insights and reinforcing expert decision-making. This presentation addresses the challenge of explaining sequential data models, focusing on time series classification. We begin by reviewing the current literature on XAI for time series classification. Then, we present a series of works that illustrate the transition from general-purpose post-hoc explanation approaches to interpretable-by-design methods. First, we introduce a local post-hoc agnostic subsequence-based time series explainer that can be used to elucidate the predictions of any time series classifier. Next, we demonstrate, through a real case study on car crash prediction, how insights from a post-hoc explainer were crucial in developing an effective interpretable-by-design method. Additionally, we showcase an interpretable subsequence-based classifier by enhancing SAX with dilation and stride to capture temporal patterns effectively. Finally, we explore the use of subsequence-based approaches in other sequential domains like mobility trajectories and text.

Advancing Learning with Temporal Knowledge Graphs

Relational machine learning has become a pivotal area of study, focusing on the analysis of relational data represented in graph structures, such as knowledge graphs (KGs). These graphs capture complex relationships between entities and are crucial in domains like social networks, recommender systems, and computational finance. Recently, the focus has shifted towards temporal knowledge graphs (tKGs), which integrate the temporal aspect of data, allowing for the modeling of evolving relationships over time. In this talk, I will introduce our innovative approaches to learning with tKGs, emphasizing the importance of interpretability and dynamic modeling. One of our key contributions is a framework that combines graph representation learning with temporal reasoning, enabling accurate predictions and providing clear and understandable explanations for these predictions. This is achieved through mechanisms that focus on relevant subgraphs and dynamic relationships, offering insights into the underlying processes. Another contribution TLogic provides a novel framework that leverages temporal logical rules to enhance the explainability and robustness of predictions on tKGs. TLogic provides a structured way to interpret the temporal dynamics within the data, ensuring that the predictions are consistent with the temporal context. This approach highlights the potential for generalizing learned knowledge across different datasets, showcasing the flexibility and power of temporal knowledge graphs in advancing the field of relational machine learning.