Privacidad y seguridad en sistemas de aprendizaje federado: revisión de métodos y desafíos computacionales
Federated learning enables model training on distributed information without necessarily centralizing raw records. Nevertheless, parameter exchanges, participant coordination, and malicious behavior create threats that do not disappear when data remain at their source. This paper examines privacy and security safeguards in federated learning systems and their computational trade-offs. A structured narrative literature review was conducted through a targeted search of verifiable academic sources and thematic analysis of foundational studies, methodological contributions, and surveys. Findings distinguish inference and reconstruction attacks, model poisoning, backdoors, and server-related threats, and compare differential privacy, secure aggregation, robust optimization, and verification mechanisms. The reviewed literature suggests that no single method simultaneously guarantees confidentiality, integrity, and availability. Appropriate protection depends on the threat model, architecture, accuracy requirements, communication overhead, and computational constraints. Trustworthy deployment, therefore, requires complementary safeguards, explicit evaluation metrics, and reproducible tests tailored to each use case.
Authors
- Francisco Ricardo Baños Solis
- Maricarmen Montserrat Velázquez Hernández
- Adelina Martínez Nieto
Institutions
- Instituto Tecnológico de Oaxaca (MX)
- Tecnológico Nacional de México (MX)
Publication Details
- Journal
- Revista Hambatu Science
- Published
- 2026-11-21
- DOI
- https://doi.org/10.63862/rhs-v1n4-356-381-2026
- Primary Topic
- Privacy-Preserving Technologies in Data
- Type
- article
- Field-Weighted Citation Impact
- 0.00