Tag
The paper proposes XACT, a framework for learning sparse attribution masks over time-frequency transforms to explain time-series classifiers, demonstrating improved precision and interpretability over baselines.
This paper presents a validation protocol for knowledge tracing models, evaluating predictive performance and the stability and faithfulness of explanations using XGBoost and deep learning baselines on ASSISTments datasets.
This paper proposes an explainable hate speech detection framework integrating DistilBERT embeddings, BiLSTM, and an attention mechanism, achieving high F1-scores on benchmark datasets for both binary and multi-class classification.
This paper introduces a dual-model masking metric to benchmark ten explainable methods for temporal attribution in sequential recommendation systems, finding that gradient-based methods like GradientSHAP and Integrated Gradients yield the most faithful and robust attributions.
This paper presents Signal2Symbol, a neuro-symbolic framework for explainable anomaly detection in physiological time-series data like ECG and EEG, leveraging symbolic tokenization and temporal reasoning to produce interpretable explanations of anomalous patterns.
The paper proposes a new evaluation method for machine learning explanations by converting explanations into predictors and testing their ability to reproduce model predictions. It demonstrates that the effectiveness of explanation methods like SHAP and PDP depends on the independence of features in the data.
This paper proposes ProKDA, a progressive knowledge-to-decision alignment method for explainable hateful meme detection that achieves state-of-the-art performance by decoupling explanation and detection tasks.
This paper presents an interpretable machine learning approach using a naive Bayes classifier on a small clinical dataset to predict cognitive impairment from inflammatory biomarkers, identifying I-309 (CCL1) as a key predictive feature.
FakeSpotter is a content- and strategy-agnostic tool designed to estimate the viral misinformation risk of textual content by measuring structural fingerprints, using repeated LLM assessments and logistic regression classifiers, with reported macro F1 scores of 0.788 for short texts and 0.793 for long texts on a labeled corpus.
This paper proposes Causal Sufficient and Necessary Regional Explanations (SNRE), a framework that uses causal inference to learn input and output regions where membership is both sufficient and necessary for model predictions, improving explainability.
This paper introduces NObSP, a framework for decomposing neural network predictions into per-feature contribution functions and interaction residuals using oblique subspace projections, improving interpretability and reducing attribution errors compared to existing methods.
SPICE introduces a generalizable framework for analyzing polysemanticity in neural networks, enabling systematic comparison across CNNs and Transformers and automatically determining concept clusters per neuron.
The paper proposes an integrated framework for multiclass gait disorder classification using bilateral GRF and COP signals, with 3D visualization and explainable AI, achieving high accuracy.
The paper proposes an Explanation-Driven Feature Acquisition (EDFA) method that unifies counterfactual, semifactual, and alterfactual explanations to jointly address algorithmic recourse and feature acquisition, aiming for lower-cost and more actionable recourse with validity guarantees.
This paper introduces the Concept-Integrated Transformer (CIT), which leverages large language models for concept supervision to achieve accurate and interpretable predictions from mobile sensing data in small-cohort health studies.
This paper introduces XAI-Arena, an LLM-as-a-judge framework for scalable and reproducible evaluation of explainable AI explanation quality, showing strong correlation with human judgments.
This paper introduces a post-hoc explainable AI framework that converts AI agent execution traces into structured reports and natural-language explanations for transparency and auditability.
This paper proposes a multi-objective neural basis model (MONBM) framework to simultaneously optimize accuracy, interpretability, and fairness in generalized additive neural networks, revealing complex trade-offs between these trustworthiness dimensions.
The paper presents a solution to the IJCAI 2025 Counterfactual Routing Competition, using integer programming with constraint generation to find minimal changes to road networks for counterfactual explanations, achieving fast runtime and competitive solution quality.
The paper proposes EPA-CarbonNet, a six-layer transformer architecture for carbon credit price prediction that integrates market data and policy text, but tests on S&P carbon index data show mixed results with a random walk outperforming on some metrics while directional accuracy is promising.