Estrategias de explicación para sistemas inteligentes en IoT
The combination of Internet of Things (IoT) solutions with Artificial Intelligence (AI) techniques has facilitated automation and improved the efficiency of multiple processes across various domains. These emerging solutions are being adopted in vital areas, such as healthcare, transportation, and p...
| Autor: | |
|---|---|
| Tipo de recurso: | tesis de maestría |
| Fecha de publicación: | 2023 |
| País: | España |
| Institución: | Universidad Complutense de Madrid (UCM) |
| Repositorio: | Docta Complutense |
| Idioma: | inglés |
| OAI Identifier: | oai:docta.ucm.es:20.500.14352/88406 |
| Acceso en línea: | https://hdl.handle.net/20.500.14352/88406 |
| Access Level: | acceso abierto |
| Palabra clave: | 004(043.3) XAI IoT Artificial intelligence Machine learning Interpretability Internet of the medical things Industrial internet of things Inteligencia artificial Aprendizaje automático Interpretabilidad Internet de las cosas médicas Internet industrial de las cosas Informática (Informática) 33 Ciencias Tecnológicas |
| Sumario: | The combination of Internet of Things (IoT) solutions with Artificial Intelligence (AI) techniques has facilitated automation and improved the efficiency of multiple processes across various domains. These emerging solutions are being adopted in vital areas, such as healthcare, transportation, and production. However, understanding complex AI models is often difficult for humans, as these models are generally limited to making decisions without explaining the reasoning behind them. The research on Explainable AI (XAI) arises from the need to enhance the interpretability of AI "black box" models, with the primary goal of improving users’ trust in the decisions made by these systems. In this work, I review the literature on XAI methods in IoT environments to identify the most important limitations in explaining this kind of system, as well as identifying the specific characteristics of IoT applications from the explicability standpoint. Then, I explain two AI models in IoT use cases through a selection of the so-called model-agnostic XAI methods. The first scenario focuses on a heart attack prediction model based on time series, which is developed in the context of the Medical Internet of Things (IoMT). The AI model in the second scenario aims to detect manufacturing defects in parking sensors and is applied within the context of the Industrial Internet of Things (IIoT). For each of the applied methods, I identify their strengths and weaknesses with special consideration for the usual requirements in IoT deployments. Based on the approach proposed in the iSee project, I design explanation strategies by combining multiple methods that intend to fulfill the goals of the different types of users receiving the explanations. Designing explanation strategies is a dual-purpose approach. On one hand, it allows combining XAI methods to enhance their capabilities and broaden their scope. On the other hand, it offers the possibility of reusing the solution in other scenarios with similar characteristics. Finally, I discuss the identified issues and lessons learned based on the results of the conducted tests. |
|---|