Global Outlier Detection in a Federated Learning Setting with Isolation Forest

Across several domains, it is common to find examples of data points that are local outliers but not Federated learning (FL) is a machine learning paradigm global outliers. For example, in the medical field, a given where multiple parties collaborate to train a shared machine medical condition may be common in one region and rare in learning model without centralizing data at a single location another [8]. Therefore, in a study conducted at a center located [1]. During model training, data holders refrain from directly in a low-prevalence region, individuals suffering from that exchanging raw data; instead, they share model parameters condition may appear as local outliers. However, if the center such as gradients, weights, or other forms of processed participates in a FL multicenter study including centers in information. This distributed learning paradigm is typically areas where the condition is more common, those individuals facilitated by a coordinating server, often referred to as the would not appear as global outliers. In most cases, for the aggregator, which collects local contributions from data holders, training of FL models, a consortium would be interested in commonly known as clients, and aggregates them to create discarding global outliers and retaining local ones.