Dell Products Inc.

Maira Beatriz Hernandez Moran

Paulo Abelha Ferreira

Pablo Nascimento Da Silva

One method includes receiving, at a central node, a respective gradient from each node of a cluster, and each of the gradients comprises a respective update to a global machine learning model maintained at the central node, aggregating, by the central node, the gradients to obtain an aggregated gradient for the cluster, and the aggregated gradient is part of a list of aggregated gradients, running, by the central node, a robust aggregation operation on the aggregated gradients in the list to obtain an outlier score for the cluster, and when the outlier score equals or exceeds a specified value, or falls within a specified range of values, identifying the cluster as an outlier, and identifying nodes within the cluster as suspicious.

Lightweight attacker identification method for federated learning with secure byzantine-robust aggregation via clustering

One example method includes performing various operations in a federated learning environment that includes nodes configured and operable to communicate with a central server. The operations include gathering aggregate gradient information from clusters that each include a respective group of some of the nodes, examining the respective gradient information for each of the clusters, and identifying outlier clusters based on the respective aggregate gradient information of those clusters, for each of the clusters identified as an outlier cluster, setting all the nodes in those clusters as suspect nodes, and performing a clustering operation that comprises reassigning each of the suspect nodes to one of the clusters, and a decision to reassign any given one of the suspect nodes is based on an amount of information expected to be obtained as a result of that decision.

Efficient entropy-guided byzantine attacker identification in federated learning

One example method includes receiving at a client node of a federation a global machine-learning model that is to be trained by the client node using a training dataset that is local to the client node. In response to receiving the global machine-learning model, determining at the client node if the global machine-learning model is trending toward an overfitted state using a validation dataset. The overfitted state indicates that the global machine-learning model has not been received from a server that is part of the federation because of a client isolation attack. In response to determining that the global machine-learning model is trending towards the overfitting state, causing the client node to leave the federation. In response to determining that the global machine-learning model is not trending towards the overfitted state, training the global machine-learning model using the training dataset to thereby update the global machine-learning model.

Detecting client isolation attacks in federated learning through overfitting monitoring

Identifying malicious clients in federated learning is disclosed while enhancing privacy. The clients are clustered such that cluster updates in the federated learning are generated. When a suspect cluster is identified, clients in the suspect clusters are labeled as suspect and clients in clusters that are not suspect are labeled as fair. The clients are reclustered and the clusters and clients are relabeled without changing the labels of clients that were previously deemed fair. After one or more iterations, the malicious clients are identified, and corrective actions can be performed.

Growing community of inventors

Rio de Janeiro, Brazil

Maira Beatriz Hernandez Moran