TimeInf: Time Series Data Contribution via Influence Functions

2024-07-21

International Conference on Learning Representations

AI Summary

Quantifying the contribution of individual time points in time-series data remains challenging, as existing attribution methods typically assume i.i.d. observations and thus fail to jointly preserve temporal dependencies and ensure interpretable, point-level attributions. Method: This paper introduces the first influence-function-based framework for non-i.i.d. time-series settings—Temporal-aware Influence Function (TIF)—which integrates temporal embedding constraints, sliding-window gradient propagation, and a time-sensitive Hessian-vector product for second-order approximation. Contribution/Results: TIF rigorously preserves temporal structure while enabling precise, time-point-level attribution. It achieves state-of-the-art performance on multi-source time-series forecasting tasks; accurately identifies harmful anomalies and critical supportive points; and generates intuitive, interpretable attribution heatmaps that facilitate visual identification of anomalous patterns.

Technology Category

• Machine Learning: Time-Series/Data Streams
• Knowledge Representation and Reasoning: Geometric, Spatial, and Temporal Reasoning
• Planning, Routing, and Scheduling: Temporal Planning

Application Category

• Graph Algorithms and Modeling for the Web: Algorithms and analysis for heterogeneous, signed, attributed, multi-relational, temporal, higher-order, and annotated Web-related graphs
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• User Modeling, Personalization and Recommendation: Attacks and countermeasures in recommendation systems

Abstract

Evaluating the contribution of individual data points to a model's prediction is critical for interpreting model predictions and improving model performance. Existing data contribution methods have been applied to various data types, including tabular data, images, and texts; however, their primary focus has been on i.i.d. settings. Despite the pressing need for principled approaches tailored to time series datasets, the problem of estimating data contribution in such settings remains unexplored, possibly due to challenges associated with handling inherent temporal dependencies. This paper introduces TimeInf, a data contribution estimation method for time-series datasets. TimeInf uses influence functions to attribute model predictions to individual time points while preserving temporal structures. Our extensive empirical results demonstrate that TimeInf outperforms state-of-the-art methods in identifying harmful anomalies and helpful time points for forecasting. Additionally, TimeInf offers intuitive and interpretable attributions of data values, allowing us to easily distinguish diverse anomaly patterns through visualizations.

Problem

• Estimating data contribution in time series with temporal dependencies
• Detecting anomalies in time series data effectively
• Providing interpretable attributions for diverse anomalous patterns

Innovation

• Leverages influence scores for time series
• Preserves temporal dependencies in data
• Detects anomalies with interpretable attributions