当前位置:首页 > 报告详情

将 DEEL 工具箱应用于 LARD.pdf

上传人: 哆哆 编号:631234 2025-04-19 48页 4.82MB

1、 DEEL-All rights reserved to IVADO,IRT Saint Exupry,CRIAQ and ANITI.Confidential and proprietary documentDEEL Certification Mission DEEL Core TeamApplying the DEEL Toolboxon LARDMlanie Ducoffe,Maxime Carrere,Lo Fliers,Adrien Gauffriau,Christophe Gabreau,Jean-Brice Ginestet,Franck Mamalet,Claire Page

2、tti,Thierry SammourThierry Daubos,Corentin Friedrich,Agustin Martin Picard,Vincent Mussot,Paul Novello,Yannick Prudent,David Vigouroux DEEL-All rights reserved to IVADO,IRT Saint Exupry,CRIAQ and ANITI.Confidential and proprietary document323/04/2025Norms®ulation How to integrate IA in critical s

3、ystems?AI Act EASA Concept Paper ARP 6983EN 50716 SAE J3061UEUELearning assurance from EASA DEEL-All rights reserved to IVADO,IRT Saint Exupry,CRIAQ and ANITI.Confidential and proprietary documentTransparency?Robustness?Training setTrain modelNeural NetworkUnseen dataMake predictionsA typical ML pro

4、cessGeneralisation?Performance guarantees?Representativeness?Biases?Relevance?DEEL-All rights reserved to IVADO,IRT Saint Exupry,CRIAQ and ANITI.Confidential and proprietary documentLARDLanding Approach Runway Detection DEEL-All rights reserved to IVADO,IRT Saint Exupry,CRIAQ and ANITI.Confidential

5、and proprietary documentLARD6TASK:Autonomous vision-based landing23/04/2025 DEEL-All rights reserved to IVADO,IRT Saint Exupry,CRIAQ and ANITI.Confidential and proprietary documentLARD:System-level ODD7TASK:Autonomous vision-based landingDefining a Generic landing approach cone:23/04/2025Operational

6、 Design Domain DEEL-All rights reserved to IVADO,IRT Saint Exupry,CRIAQ and ANITI.Confidential and proprietary documentLARD:ML Constituent8Intended function 1(VBL intended function):Pose estimation of the aircraft with respect to the runway when the aircraft flies within the generic landing approach

word格式文档无特别注明外均可编辑修改,预览文件经过压缩,下载原文更清晰!
三个皮匠报告文库所有资源均是客户上传分享,仅供网友学习交流,未经上传用户书面授权,请勿作商用。
本文主要介绍了DEEL认证任务的核心团队如何应用DEEL工具箱在LARD(基于视觉的自主着陆)任务上,以及他们在AI合规性、规范和性能方面的研究成果。文章提到了关键的数据集特征、模型架构和训练方法,以及如何通过规范和性能保证来确保着陆安全。研究团队还探讨了如何检测和缓解数据集和模型中的偏差,提高模型的鲁棒性,并通过 conformal prediction 和 XAI(解释性机器学习)技术为预测提供概率保证和解释。此外,文章还简要介绍了用于影响函数和异常检测的Tensorflow工具箱Influenciae,以及如何通过颜色聚类和边界框过滤来改进检测效果。最后,研究团队提出了一个用于获取检测框概率保证的OOD(异常外推)检测方法,并使用XPlique工具为模型决策提供解释。
"如何确保AI在关键系统中的集成安全?" "AI在自动驾驶着陆系统中的应用挑战有哪些?" "如何通过机器学习确保飞机着陆的安全性?"
客服
商务合作
小程序
服务号
折叠