ExplainabilityTransparencyAccountabilitycs.LG
Sparse, Efficient and Explainable Data Attribution with DualXDA
Galip Ümit Yolcu, Moritz Weckbecker, Thomas Wiegand, Wojciech Samek, Sebastian Lapuschkin
February 19, 2024
5 citations
Key Takeaways
-
Data attribution can now scale to real-world AI systems: DualXDA achieves computational speedups of up to 4,100,000x compared to existing methods, making it practical to identify which training data influenced a model's decisions on medium-scale datasets and models—a capability previously limited to small-scale applications.
-
Sparse explanations reveal what actually matters: By producing "sparse" attributions (focusing on fewer, more important training examples rather than spreading scores across many), DualXDA helps practitioners identify the decisive patterns in data that drive model predictions, making it easier to spot problematic training data or debug model failures.
-
Better transparency for model accountability: The framework enables organizations to provide transparency about individual predictions and model behavior, supporting critical use cases like identifying data-related causes of poor performance—essential for responsible deployment in high-stakes domains.
-
Efficiency gains enable practical debugging: The dramatic improvement in explanation speed means practitioners can now feasibly use data attribution for model debugging and quality assurance on realistic datasets, rather than relying on approximations that might miss important aspects of how models actually work.
Abstract
Data Attribution (DA) is an emerging approach in the field of eXplainable Artificial Intelligence (XAI), aiming to identify influential training datapoints which determine model outputs. It seeks to provide transparency about the model and individual predictions, e.g. for model debugging, identifying data-related causes of suboptimal performance. However, existing DA approaches suffer from prohibitively high computational costs and memory demands when applied to even medium-scale datasets and models, forcing practitioners to resort to approximations that may fail to capture the true inference process of the underlying model. Additionally, current attribution methods exhibit low sparsity, resulting in non-negligible attribution scores across a high number of training examples, hindering the discovery of decisive patterns in the data. In this work, we introduce DualXDA, a framework for sparse, efficient and explainable DA, comprised of two interlinked approaches, Dual Data Attribution (DualDA) and eXplainable Data Attribution (XDA): With DualDA, we propose a novel approach for efficient and effective DA, leveraging Support Vector Machine theory to provide fast and naturally sparse data attributions for AI predictions. In extensive quantitative analyses, we demonstrate that DualDA achieves high attribution quality, excels at solving a series of evaluated downstream tasks, while at the same time improving explanation time by a factor of up to 4,100,000x compared to the original Influence Functions method, and up to 11,000x compared to the method's most efficient approximation from literature to date. We further introduce XDA, a method for enhancing Data Attribution with capabilities from feature attribution methods to explain why training samples are relevant for the prediction of a test sample in terms of impactful features, which we showcase and verify qualitatively in detail.