20240028911. EFFICIENT SAMPLING OF EDGE-WEIGHTED QUANTIZATION FOR FEDERATED LEARNING simplified abstract (Dell Products L.P.)

EFFICIENT SAMPLING OF EDGE-WEIGHTED QUANTIZATION FOR FEDERATED LEARNING

Organization Name

Dell Products L.P.

Inventor(s)

Pablo Nascimento Da Silva of Niteroi (BR)

Vinicius Michel Gottin of Rio de Janeiro (BR)

Paulo Abelha Ferreira of Rio de Janeiro (BR)

EFFICIENT SAMPLING OF EDGE-WEIGHTED QUANTIZATION FOR FEDERATED LEARNING - A simplified explanation of the abstract

This abstract first appeared for US patent application 20240028911 titled 'EFFICIENT SAMPLING OF EDGE-WEIGHTED QUANTIZATION FOR FEDERATED LEARNING

Simplified Explanation

The abstract of this patent application describes a method that involves sampling a specified number of edge nodes using historical statistics, calculating a composite time for each node, and selecting nodes for sampling based on a cutoff threshold. The composite time is determined by adding the federated learning time and the execution time of a quantization selection procedure.

• The method involves running an edge node sampling algorithm with a specified parameter 's' to determine the number of edge nodes to be sampled.
• Historical statistics from the edge nodes are used to calculate a composite time for each node.
• The composite time is the sum of the federated learning time and the execution time of a quantization selection procedure.
• An outlier boundary is identified, and a cutoff threshold is defined based on this boundary.
• Edge nodes that are at or below the cutoff threshold are selected for sampling.

Potential applications of this technology:

• This method can be applied in edge computing systems where sampling of edge nodes is required for analysis or optimization purposes.
• It can be used in federated learning systems to select a subset of edge nodes for participation in the learning process.
• The method can be utilized in distributed systems to improve efficiency by selecting only the most relevant edge nodes for certain tasks.

Problems solved by this technology:

• The method solves the problem of selecting a representative subset of edge nodes for analysis or participation in distributed tasks.
• It addresses the challenge of determining the composite time of edge nodes based on historical statistics.
• The method solves the problem of setting a cutoff threshold to identify edge nodes that are outliers and should not be included in the sampling process.

Benefits of this technology:

• The method allows for efficient sampling of edge nodes, reducing computational and communication overhead.
• It improves the accuracy of analysis or learning processes by selecting edge nodes that are most relevant or representative.
• The method provides a systematic approach to selecting edge nodes based on composite time, ensuring fairness and consistency in the sampling process.

Original Abstract Submitted

one example method includes running an edge node sampling algorithm using a parameter ‘s’ that specifies a number of edge nodes to be sampled, using historical statistics from the edge nodes, calculating a composite time for each of the edge nodes, and the composite time comprises a sum of a federated learning time and an execution time of a quantization selection procedure, identifying an outlier boundary, defining a cutoff threshold based on the outlier boundary, and selecting, for sampling, the edge nodes that are at or below the cutoff threshold.