A Visual Empirical Region of Influence Pattern Recognition Tool for Leave-One-Out Data Analysis

Martinez, Rubel F.

doi:10.2172/793409

Cited by 1 publication

(3 citation statements)

References 1 publication

(1 reference statement)

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Influence functions compute the sway of individual training samples, facilitating pointed insights [19], whereas TracIn quantifies both positive and negative training influences [81]. Techniques like LOO display influence by showing prediction shifts [264], while Jackknife and Shapley Valuation assign values signifying influence magnitude [170], [91]. On the contrary, the tools [98], [106] intensify visual clarity in output representation, with PaLM showcasing iterative model responses [120].…”

Section: Results Reporting and Visualizationmentioning

confidence: 99%

“…Within the domains of data valuation and anomaly detection, XAI methodologies proffer a comprehensive framework to discover the influence of training data on predictive modeling, exhibiting adaptability across various data types such as tabular, text, and image [19], [170]. Sample valuationbased explainers such as influence functions, TracIn, LOO, Jackknife, and Shapley Valuation are engineered to accommodate a wide array of data, with TracIn showing substantial applicability in image and text scenarios [19], [81], [264], [170], [91]. Conversely, sample anomaly-based explainers are typically optimized for specific data modalities, with approaches like O2u-Net, TAPUDD, and Smirnov et al centering on image data [106], [117], [98], while Jia et al offer a versatile methodology that is effective across data formats [15], and PaLM focuses on tabular datasets [120].…”

Section: Data Acquisition and Collectionmentioning

confidence: 99%

“…In data modeling and analysis, XAI is used to unravel the impact of training data on predictive models. Sample valuation-based explainers such as influence functions [19] and TracIn [81] utilize gradients in their assessments, while methods like LOO [264] measure changes in model predictions after sample exclusion. On the anomaly detection front, Smirnov et al pair challenging samples during training [98], O2u-Net ranks samples by loss curves highlighting potential mislabels [106], and Jia et al employ a time series classifier to spot mislabeling [15].…”

Section: Data Modeling and Analysismentioning

confidence: 99%

See 2 more Smart Citations

Classification of Kinematic Data Using Explainable Artificial Intelligence (XAI) for Smart Motion

Yong¹

2021

Explainable Artificial Intelligence for Smart Cities

View full text Add to dashboard Cite

Given the complexity and lack of transparency in deep neural networks (DNNs), extensive efforts have been made to make these systems more interpretable or explain their behaviors in accessible terms. Unlike most reviews, which focus on algorithmic and model-centric perspectives, this work takes a "data-centric" view, examining how data collection, processing, and analysis contribute to explainable AI (XAI). We categorize existing work into three categories subject to their purposes: interpretations of deep models, referring to feature attributions and reasoning processes that correlate data points with model outputs; influences of training data, examining the impact of training data nuances, such as data valuation and sample anomalies, on decision-making processes; and insights of domain knowledge, discovering latent patterns and fostering new knowledge from data and models to advance social values and scientific discovery. Specifically, we distill XAI methodologies into data mining operations on training and testing data across modalities, such as images, text, and tabular data, as well as on training logs, checkpoints, models and other DNN behavior descriptors. In this way, our study offers a comprehensive, data-centric examination of XAI from a lens of data mining methods and applications.

show abstract