Aprendizado Federado Incremental e Sensível ao Risco para Modelos de Ranqueamento em Cenários com Distribuições Heterogêneas de Dados

• Gestefane Rabbi UFMG
• Daniel Xavier de Sousa IFG
• Marcos André Gonçalves UFMG

DOI: https://doi.org/10.5753/webmedia_estendido.2026.34910

Resumo

This dissertation proposes FedRisk, a novel Federated Learning to Rank (FLTR) method designed to address the challenge of building a reliable global model under non-independent and non-identically distributed (non-IID) data. The method combines a risk-sensitive aggregation strategy, which adjusts each client’s contribution based on the variability of its prediction errors, with the incorporation of explicit temporal memory of historical global parameters. Evaluated on the MSLR-WEB10K benchmark with 100 clients under Dirichlet sampling, FedRisk achieves a 15.6% improvement in nDCG@5 over FedProx and attains performance comparable to centralized training in nDCG@10, while reducing inter-round variability by more than 90%. Recognized with the Best Full Paper Honorable Mention at SBBD 2025, this research advances the development of effective and stable federated ranking models for decentralized Web environments.

Palavras-chave: Federated Learning, Learning to Rank, Non-IID Data, Risk Sensitivity, Temporal Memory, Web Systems

Referências

Daniel J Beutel, Taner Topal, Akhil Mathur, Xinchi Qiu, Javier Fernandez-Marques, Yan Gao, Liam Sani, Hang Li Kwing, Titouan Parcollet, Pedro PP Gusmão, et al. 2020. Flower: A Friendly Federated Learning Research Framework. In arXiv preprint arXiv:2007.14390.

Peter Kairouz, H Brendan McMahan, Brendan Avent, Aurélien Bellet, Mehdi Bennis, Arjun Nitin Bhagoji, Kallista Bonawitz, Zachary Charles, Graham Cormode, Rachel Cummings, et al. 2021. Advances and open problems in federated learning. Foundations and Trendsæ in Machine Learning 14, 1–2 (2021), 1–210.

Sai Praneeth Karimireddy, Satyen Kale, Mehryar Mohri, Sashank Reddi, Sebastian Stich, and Ananda Theertha Suresh. 2020. SCAFFOLD: Stochastic controlled averaging for federated learning. In ICML.

Tian Li, Anit Kumar Sahu, Ameet Talwalkar, and Virginia Smith. 2020. Federated optimization in heterogeneous networks. In Proceedings of MLSys. 429–450.

Ji Liu, Jianshu Huang, et al. 2021. From distributed machine learning to federated learning: A survey. arXiv preprint arXiv:2104.14362 (2021).

Tie-Yan Liu. 2010. Learning to rank for information retrieval. In Proceedings of SIGIR. 904.

H Brendan McMahan, Eider Moore, Daniel Ramage, Seth Hampson, and Blaise Agüera y Arcas. 2017. Communication-efficient learning of deep networks from decentralized data. In AISTATS.

Jose Neto et al. 2020. Privacidade de dados em aprendizado de máquina distribuído: conceitos, desafios e perspetivas. Revista Brasileira de Computação (2020).

Tao Qin and Tie-Yan Liu. 2013. Introducing LETOR 4.0 Datasets. CoRR abs/1306.2597 (2013).

Paulo Henrique Rodrigues, Daniel Xavier Sousa, Celso França, Gestefane Rabbi, Thierson Couto Rosa, and Marcos André Gonçalves. 2025. Risk-sensitive optimization of neural deep learning ranking models with applications in ad-hoc retrieval and recommender systems. Information Processing & Management 62, 4 (2025), 104126.

Paulo Henrique Rodrigues, Daniel Xavier Sousa, Thierson Couto Rosa, and Marcos André Gonçalves. 2022. Risk-sensitive deep neural learning to rank. In Proceedings of SIGIR. 803–813.

Jianyu Wang, Qinghua Liu, Huan Liang, Gauri Joshi, and H Vincent Poor. 2020. Tackling ecological fallacy in federated learning. In ICML.

Shuguang Wang and Guido Zuccon. 2022. Is non-IID data a threat in federated online learning to rank?. In Proceedings of SIGIR. 2801–2813.

Tao Yu, Eugene Bagdasaryan, and Vitaly Shmatikov. 2022. Salvaging federated learning by local adaptation. arXiv preprint arXiv:2202.04752 (2022).

Hangyu Zhu, Junyuan Jin, Huazhuo Li, and Xiao Liang. 2021. Federated learning on non-IID data: A survey. Neurocomputing 465 (2021), 371–390.