{"id":18216,"date":"2026-06-05T08:36:03","date_gmt":"2026-06-05T06:36:03","guid":{"rendered":"https:\/\/www.uni.lu\/research-en\/?post_type=events&p=18216"},"modified":"2026-06-05T08:36:04","modified_gmt":"2026-06-05T06:36:04","slug":"doctoral-defence-arno-michel-denis-geimer","status":"publish","type":"events","link":"https:\/\/www.uni.lu\/research-en\/events\/doctoral-defence-arno-michel-denis-geimer\/","title":{"rendered":"Doctoral Defence: Arno Michel Denis GEIMER"},"content":{"rendered":"\n
<\/div>\n\n\n

The Doctoral School in Science and Engineering<\/a> is happy to invite you to Arno Michel Denis GEIMER\u2019s defence entitled<\/p>\n\n\n\n

\nHeterogeneity in Federated Learning: Rethinking Evaluation Methods<\/h2>\n\n\n\n

Supervisor<\/strong>: Prof Radu STATE<\/a><\/p>\n\n\n\n

The thesis investigates Federated Learning (FL), a decentralized machine learning paradigm that enables multiple data holders to collaboratively train models without sharing raw data. Since its introduction, FL has gained significant attention due to its potential to address privacy constraints in increasingly regulated data environments.<\/p>\n\n\n\n

During the defense, we provide a comprehensive overview of Federated Learning, including its core workflow and key algorithmic components. The primary contributions of the thesis address participant heterogeneity, a central challenge in federated systems. Data heterogeneity will be studied by analyzing contribution invariance under different aggregation mechanisms, characterizing how individual client updates influence global model dynamics.  On the other side, system-level heterogeneity will be addressed by introducing a decentralized hyperparameter optimization algorithm that adapts to variations in hardware capabilities across participants. To further support this line of research, we have developed a novel simulation layer for device-limited workload heterogeneity. The proposed simulation framework addresses a gap in existing benchmarking practices, where such sources of heterogeneity are often underrepresented. We conclude with perspectives on future research directions and the evolving role of Federated Learning in privacy-preserving artificial intelligence.<\/p>\n\n\n\n