Search research articles

ABOUT JoVE

Overview Leadership Blog JoVE Help Center

AUTHORS

Publishing Process Editorial Board Scope & Policies Peer Review FAQ Submit

LIBRARIANS

Testimonials Subscriptions Access Resources Library Advisory Board FAQ

RESEARCH

JoVE Journal Methods Collections JoVE Encyclopedia of Experiments Archive

EDUCATION

JoVE Core JoVE Business JoVE Science Education JoVE Lab Manual Faculty Resource Center Faculty Site

Terms & Conditions of Use

Search research articles

Related Experiment Video

Updated: Sep 2, 2025

Author Spotlight: Enhancement of Salient Object Detection for Smart Grid Applications

Author Spotlight: Enhancement of Salient Object Detection for Smart Grid Applications

Published on: December 15, 2023

Benchmarking Perturbation-Based Saliency Maps for Explaining Atari Agents.

Tobias Huber¹, Benedikt Limmer¹, Elisabeth André¹

¹Chair for Human-Centered Artificial Intelligence, University of Augsburg, Augsburg, Germany.

Frontiers in Artificial Intelligence

|August 1, 2022

Summary

This summary is machine-generated.

Related Concept Videos

You might also read

Related Articles

Articles linked to this work by shared authors, journal, and citation graph.

Sort by

Same author

Closing the Carbon Loop: Continuous-Flow Synthesis of N-Aroyl Sulfoximines via Electrochemical CO<sub>2</sub> Reduction.

ChemSusChem·2026

Same author

A spatial in situ hybridization approach to T cell clonotype analysis using T cell receptor variable gene probes.

Cell reports·2026

Same author

A novel endpoint for liver surgery predicts surgery-related mortality: external validation in a surgical cohort with intrahepatic cholangiocarcinoma.

Frontiers in surgery·2026

Same author

Salvage Hepatectomy in Toxic Liver Syndrome as a Bridge to Orthotopic Liver Retransplantation-Case Series and Structured Narrative Review.

Journal of clinical medicine·2026

Same author

Incidence, patterns and management of complications (90-day-analysis) after liver surgery - a single center experience with 3177 consecutive hepatic resections.

Langenbeck's archives of surgery·2026

Same author

Couinaud segment-aware deep learning on point clouds for major liver resection planning.

International journal of computer assisted radiology and surgery·2026

Same journal

Structural impact of non-IID heterogeneity on federated behavioral anomaly detection in IoT and IoMT systems.

Frontiers in artificial intelligence·2026

Same journal

DiscoVerse: multi-agent pharmaceutical co-scientist for traceable drug discovery and reverse translation.

Frontiers in artificial intelligence·2026

Same journal

EEG-based cognition-aware task classification and scheduling using enhanced fuzzy transition modeling.

Frontiers in artificial intelligence·2026

Same journal

Autofluorescence and deep learning in early disease detection: biological foundations, clinical applications, and future directions.

Frontiers in artificial intelligence·2026

Same journal

Legal document summarization: a short review.

Frontiers in artificial intelligence·2026

Same journal

Generative AI adoption and its impact on teachers' self-efficacy and instructional confidence in Ghana.

Frontiers in artificial intelligence·2026

See all related articles

This study computationally evaluates saliency maps for Deep Reinforcement Learning (DRL) agents, finding that some methods need adjustments for accurate analysis of agent decisions and value estimation.

Area of Science:

Artificial Intelligence
Machine Learning
Deep Reinforcement Learning

Background:

Saliency maps explain Deep Reinforcement Learning (DRL) agent behavior by highlighting pixel importance.
Existing methods lack computational evaluation for DRL agents, especially considering their policy-based decision-making.

Purpose of the Study:

To computationally evaluate and compare perturbation-based saliency map approaches for DRL agents.
To assess the fidelity and parameter dependence of these saliency maps.

Main Methods:

Compared five perturbation-based saliency map approaches on DRL agents trained on Atari games.
Used sanity checks (parameter dependence) and input degradation (fidelity) as evaluation metrics.

Main Results:

Keywords:

deep reinforcement learning explainable artificial intelligence (XAI)explainable reinforcement learning feature attribution interpretable machine learning saliency maps

More Related Videos

Measuring Attention and Visual Processing Speed by Model-based Analysis of Temporal-order Judgments

Measuring Attention and Visual Processing Speed by Model-based Analysis of Temporal-order Judgments

Published on: January 23, 2017

Methods to Test Visual Attention Online

Methods to Test Visual Attention Online

Published on: February 19, 2015

Related Experiment Videos

Last Updated: Sep 2, 2025

Author Spotlight: Enhancement of Salient Object Detection for Smart Grid Applications

Author Spotlight: Enhancement of Salient Object Detection for Smart Grid Applications

Published on: December 15, 2023

Measuring Attention and Visual Processing Speed by Model-based Analysis of Temporal-order Judgments

Measuring Attention and Visual Processing Speed by Model-based Analysis of Temporal-order Judgments

Published on: January 23, 2017

Methods to Test Visual Attention Online

Methods to Test Visual Attention Online

Published on: February 19, 2015

A popular noise-based saliency method showed low dependence on output layer parameters, which was improved by focusing on specific actions.
Identified factors influencing saliency map choice based on whether analyzing action choice or state value estimation.

Conclusions:

Saliency map evaluation for DRL requires methods sensitive to policy and value functions.
Specific approaches are better suited for analyzing action selection versus state value in value-based DRL agents.