Recent advancements in machine learning have improved performance while also increasing computational demands. While federated and distributed setups address these issues, their structures remain vulnerable to malicious influences. In this paper, we address a specific threat: Byzantine attacks, wherein compromised clients inject adversarial updates to derail global convergence. We combine the concept of trust scores with trial function methodology to dynamically filter outliers. Our methods address the critical limitations of previous approaches, allowing operation even when Byzantine nodes are in the majority. Moreover, our algorithms adapt to widely used scaled methods such as Adam and RMSProp, as well as practical scenarios, including local training and partial participation. We validate the robustness of our methods by conducting extensive experiments on both public datasets and private ECG data collected from medical institutions. Furthermore, we provide a broad theoretical analysis of our algorithms and their extensions to the aforementioned practical setups. The convergence guaranties of our methods are comparable to those of classical algorithms developed without Byzantine interference.

The increasing adoption of smartwatches has concurrently expanded the potential for ongoing health monitoring via single-lead ECG recordings. This advancement has led to new possibilities for automatically identifying heart conditions. Neural networks have become one of the most popular methods for handling the classification of ECGs, achieving notable successes in this field. However, there are significant challenges that need to be addressed. One such challenge is the lack of labeled single-lead ECG data. Another one is that single-lead ECGs contain less diagnostic information compared to traditional 12-lead ECGs, which limits the effectiveness of neural networks in ECG classification. Addressing these challenges, in this paper we introduce a novel approach that leverages the multi-view structure of ECG records in a contrastive learning paradigm. Our method not only involves creating positive pairs through augmentations of the original signal but also through diverse combinations of leads from the same ECG. After pretraining models on 12-lead ECGs, we fine-tune them using single-lead ECGs to improve classification performance. We compare the proposed method with existing contrastive learning frameworks and show its superior performance in the classification of cardiac abnormalities.