Interpretability Evaluation

Interpretability evaluation focuses on developing and applying methods to assess the understandability and trustworthiness of machine learning models, particularly deep neural networks. Current research emphasizes developing more robust and reliable metrics, including those that consider model modularity, conceptual similarity, and the reduction of explanation noise in techniques like Integrated Gradients. This work is crucial for building trust in AI systems and ensuring responsible deployment across various applications, from copyright infringement detection to active learning strategies, by providing rigorous methods for evaluating the quality and accuracy of explanations generated by these models.

Papers

November 7, 2024

October 30, 2024

MIXAD: Memory-Induced Explainable Time Series Anomaly Detection
Minha Kim, Kishor Kumar Bhaumik, Amin Ahsan Ali, Simon S. Woo
Anomaly Detection Inherent Interpretability Multivariate Time Series Data Interpretability Evaluation

October 20, 2024

A Novel Interpretability Metric for Explaining Bias in Language Models: Applications on Multilingual Models from Southeast Asia
Lance Calvin Lim Gamboa, Mark Lee
Language Model Multilingual Model Bias Evaluation LLM Bias Bias Challenge Interpretability Evaluation

September 24, 2024

Training Neural Networks for Modularity aids Interpretability
Satvik Golechha, Dylan Cope, Nandi Schoots
Neural Network Interpretable Network Interpretability Evaluation

February 14, 2024

Interpretable Measures of Conceptual Similarity by Complexity-Constrained Descriptive Auto-Encoding
Alessandro Achille, Greg Ver Steeg, Tian Yu Liu, Matthew Trager, Carson Klingenberg, Stefano Soatto
High Similarity Vision Based Visual Data Auto Encoders Semantic Affinity Interpretability Evaluation

April 11, 2023

OpenAL: Evaluation and Interpretation of Active Learning Strategies
W. Jonas, A. Abraham, L. Dreyfus-Schmidt
Global Evaluation Active Learning Open Source Abstract Interpretation Active Sampling Interpretability Evaluation

March 24, 2023

IDGI: A Framework to Eliminate Explanation Noise from Integrated Gradients
Ruo Yang, Binghui Wang, Mustafa Bilgic
New Framework Saliency Map Gradient Computation Integrated Gradient Interpretability Evaluation

June 11, 2022

Improving Accuracy of Interpretability Measures in Hyperparameter Optimization via Bayesian Algorithm Execution
Julia Moosbauer, Giuseppe Casalicchio, Marius Lindauer, Bernd Bischl
Bayesian Optimization Hyperparameter Optimization Related Hyperparameters Enhancing Accuracy Bayesian Algorithm Interpretability Evaluation

May 18, 2022

The Solvability of Interpretability Evaluation Metrics
Yilun Zhou, Julie Shah
Concept Based Explanation Quality Interpretability Evaluation

February 11, 2022

InterpretTime: a new approach for the systematic evaluation of neural-network interpretability in time series classification
Hugues Turbé, Mina Bjelogrlic, Christian Lovis, Gianmarco Mengaldo
Inherent Interpretability Time Matter Novel Approach Time Series Classification Interpretability Method Interpretability Evaluation

Interpretability Evaluation

Papers

Interplay between Federated Learning and Explainable Artificial Intelligence: a Scoping Review

Interpretable Measurement of CNN Deep Feature Density using Copula and the Generalized Characteristic Function

MIXAD: Memory-Induced Explainable Time Series Anomaly Detection

A Novel Interpretability Metric for Explaining Bias in Language Models: Applications on Multilingual Models from Southeast Asia

Training Neural Networks for Modularity aids Interpretability

Interpretable Measures of Conceptual Similarity by Complexity-Constrained Descriptive Auto-Encoding

OpenAL: Evaluation and Interpretation of Active Learning Strategies

IDGI: A Framework to Eliminate Explanation Noise from Integrated Gradients

Improving Accuracy of Interpretability Measures in Hyperparameter Optimization via Bayesian Algorithm Execution

The Solvability of Interpretability Evaluation Metrics

InterpretTime: a new approach for the systematic evaluation of neural-network interpretability in time series classification