.png){kind=icon}
返回
Expected Integral Discrete Gradient: Diagnosing Autonomous Driving Model
.png){kind=icon}
PRE
AI
DOI：10.1109/TVT.2024.3436614.png)
.png){kind=icon}
原文求助.png){kind=icon}
分享.png){kind=icon}
收藏
摘要
En 中文
The demand for explainability in complex machine learning (ML) models is ever more pressing, particularly within safety-critical domains like autonomous driving. The Integrated Gradient (IG), a prominent attribution-based explainable artificial intelligence method, offers an effective solution for explaining and diagnosing ML models. However, IG is primarily designed for deep neural network models, hindering its broader applicability to various structured machine-learning models. Moreover, the issue of selecting a suitable single baseline point in IG methods still needs to be solved. In response to these challenges, this paper introduces a model-agnostic explainable technique called the Expected Integral Discrete Gradient (EIDG). This approach extends the capabilities of IG to encompass a wide range of machine-learning models by leveraging numerical differentiation. It replaces the previous single baseline point scheme with a distributed multi-baseline method to reveal how varying baselines affect the output. Our method is thoroughly evaluated on standard machine learning models, targeting scenarios in autonomous driving scenarios, thereby validating its effectiveness in explaining and diagnosing models. Our work will inspire and equip developers and users with the necessary tools to promote the adoption of attribution explanations across various machine-learning domains.
Keyword:
Numerical models
Machine learning
Autonomous vehicles
Predictive models
Data models
Sun
Artificial neural networks
Explainable artificial intelligence
integrated gradient
machine learning
autonomous driving
期刊
IF:
7.1
论文数:
1.8W
被引数:
6.6W
机构
引用论文
SVCE: Shapley Value Guided Counterfactual Explanation for Machine Learning-Based Autonomous Driving
PREAI
分享收藏Intercellular Adhesion Molecule 1 (ICAM-1) Gene Variant is Associated with Coronary Artery Calcification Independent of Soluble ICAM-1 Levels细胞间粘附分子1 (ICAM-1) 基因变异与冠状动脉钙化相关，与可溶性ICAM-1水平无关
PREAI
分享收藏
Functional MRI of the cortical sensorimotor system in patients with hereditary spastic paraplegia.png)
Spinal Cord
OAAI
Spinal CordIF0
OAAI分享收藏A Survey on Explainable Artificial Intelligence (XAI): Toward Medical XAI可解释人工智能 (XAI) 调查: 走向医学XAIOAAI
OAAI分享收藏
查看更多内容