Study of FATES Properties in the MLOps field
Uloženo v:
| Název: | Study of FATES Properties in the MLOps field |
|---|---|
| Autoři: | Alkan, Emré, Ibazizene, Kaci |
| Přispěvatelé: | CESI : groupe d’Enseignement Supérieur et de Formation Professionnelle (CESI), HESAM Université - Communauté d'universités et d'établissements Hautes écoles Sorbonne Arts et métiers université (HESAM), IRIT : Institut de Recherche Informatique de Toulouse, UT2J : Université Toulouse 2 Jean Jaurès, ANR-24-IAS2-0002,FATES-MLOps,Intégration des propriétés FATES dans le développement continu des systèmes basés Machine Learning : application au MLOps(2024) |
| Zdroj: | https://hal.science/hal-05014141 ; IRIT : Institut de Recherche Informatique de Toulouse; UT2J : Université Toulouse 2 Jean Jaurès. 2025, pp.46. |
| Informace o vydavateli: | CCSD |
| Rok vydání: | 2025 |
| Témata: | Post-processing Algorithms 5.1.2. Fairness' Metrics 5.1.3. Fairness Libraries 5.2. Accountability 5.2.1. Technics to improve accountability 5.2.2. Adding a verification method 5.3. Transparency 5.3.1. Model Cards 5.3.2. Datasheet 5.4. Ethics 5.5. Safety &, amp, Security 5.5.1. Safety Technics 5.5.1.1. Robustness testing and validation 5.5.1.2. Explainable AI (XAI) 5.5.1.3. Human oversight 5.5.1.4. Security protocols 5.5.2. Security Techniques 5.5.2.1. Threat Modeling and Risk Assessment 5.5.2.2. Secure Development and Validation Practices 5.5.2.3. Monitoring Incident Response and Lifecycle Management 6. CONCLUSION, Security 5.5.1. Safety Technics 5.5.1.1. Robustness testing and validation 5.5.1.2. Explainable AI (XAI) 5.5.1.3. Human oversight 5.5.1.4. Security protocols 5.5.2. Security Techniques 5.5.2.1. Threat Modeling and Risk Assessment 5.5.2.2. Secure Development and Validation Practices 5.5.2.3. Monitoring, Incident Response, and Lifecycle Management 6. CONCLUSION, [INFO]Computer Science [cs], [INFO.INFO-AI]Computer Science [cs]/Artificial Intelligence [cs.AI] |
| Popis: | The MLOps movement builds upon the principles of DevOps (Kim et al., 2016), integrating the specific challenges of Machine Learning (ML) to enhance automation, integration, and monitoring throughout the model lifecycle (Testi et al., 2022). This systematic approach introduces new quality requirements for software systems, ensuring model performance and adaptability as data evolves. In this context, the shift from model-centric AI to data-centric AI highlights the necessity of formalizing and tracking fundamental principles throughout the ML system development process. The FAT/ML initiative, launched in 2014, initially introduced three key properties: Fairness, Accountability, and Transparency. These principles were later expanded with the inclusion of Ethics by Microsoft’s FATE research group and, more recently, Safety and Security, forming the FATES framework. The FATES-MLOps project aims to study these properties and propose a systematic approach to ensure their integration into ML systems developed using MLOps methodologies. While certain algorithms already address some of these properties (such as Fairness, Transparency, and Security), others, like Accountability and Ethics, rely more on organizational and regulatory commitments. However, there is still no unified framework or systematic indicators that guide ML scientists and engineers in ensuring adherence to the FATES principles.In this document, we will define the different FATES properties, analyze their interconnections, and explore existing tools and metrics that help integrate these principles into ML models. Ultimately, our goal is to ... |
| Druh dokumentu: | report |
| Jazyk: | English |
| Dostupnost: | https://hal.science/hal-05014141 https://hal.science/hal-05014141v1/document https://hal.science/hal-05014141v1/file/RapportCESI2025%20%281%29.pdf |
| Rights: | info:eu-repo/semantics/OpenAccess |
| Přístupové číslo: | edsbas.FD6BA9DF |
| Databáze: | BASE |
Nájsť tento článok vo Web of Science | Abstrakt: | The MLOps movement builds upon the principles of DevOps (Kim et al., 2016), integrating the specific challenges of Machine Learning (ML) to enhance automation, integration, and monitoring throughout the model lifecycle (Testi et al., 2022). This systematic approach introduces new quality requirements for software systems, ensuring model performance and adaptability as data evolves. In this context, the shift from model-centric AI to data-centric AI highlights the necessity of formalizing and tracking fundamental principles throughout the ML system development process. The FAT/ML initiative, launched in 2014, initially introduced three key properties: Fairness, Accountability, and Transparency. These principles were later expanded with the inclusion of Ethics by Microsoft’s FATE research group and, more recently, Safety and Security, forming the FATES framework. The FATES-MLOps project aims to study these properties and propose a systematic approach to ensure their integration into ML systems developed using MLOps methodologies. While certain algorithms already address some of these properties (such as Fairness, Transparency, and Security), others, like Accountability and Ethics, rely more on organizational and regulatory commitments. However, there is still no unified framework or systematic indicators that guide ML scientists and engineers in ensuring adherence to the FATES principles.In this document, we will define the different FATES properties, analyze their interconnections, and explore existing tools and metrics that help integrate these principles into ML models. Ultimately, our goal is to ... |
|---|