FeCA roadmap. 1st column: The centralized dataset distributed to clients. 2nd column: The k-means clustering results on different clients under non-IID data sample scenario, where black triangles and squares represent centroids. 3rd column: Eliminating one-fit-many centroids in Algorithm 2, indicated by hollow squares and triangles. 4th column: Centroids sent to the server. 5th column: Aggregation of received centroids on the server where red crosses represent recovered centroids.
Invention Summary:
FeCA, a one-shot federated clustering method, provides data privacy and enhanced performance by leveraging models trained locally on users’ devices. Traditional centralized machine learning approaches often face limitations due to siloed, private datasets that are limited in quantity and diversity.
Rutgers researchers have developed a method that developed a one-shot federated learning (provides data privacy by having models trained locally on a user’s device) for clustering (partitions a dataset into different groups based on the similarity of individual datapoints) by first performing standard clustering on the user’s device, refining the model, and then aggregating the refined model with that from other users’ devices. There is also a framework to expand this technique to other non-convex machine learning problems, including neural networks, enabling models to capture more consistent and nuanced patterns across user datasets.
Market Applications:
Advantages:
Publications:
Intellectual Property & Development Status: Provisional application filed. Patent pending. Available for licensing and/or research collaboration. For any business development and other collaborative partnerships, contact: marketingbd@research.rutgers.edu