Distributed Graph Hashing.

Area of Science:

• Computer Science
• Machine Learning
• Data Mining

Background:

• Hashing-based approximate nearest neighbors (ANN) search is crucial for big data due to its efficiency.
• Existing centralized hashing methods are inadequate for large-scale, distributed datasets.
• Distributed learning to hash is essential for modern data architectures.

Purpose of the Study:

• To develop a novel distributed graph hashing model for learning hash functions from distributed data.
• To address the limitations of centralized hashing in large-scale, decentralized applications.
• To enable efficient ANN search on data spread across multiple network agents.

Main Methods:

• Proposed a distributed graph hashing model utilizing a decomposable graph matrix.
• Formulated the hashing problem as a non-convex constrained distributed optimization problem.
• Transformed the problem into a bi-convex optimization problem and solved it using alternating direction method of multipliers (ADMM).

Main Results:

• The graph matrix can be decomposed into local matrices, constructible by individual agents.
• The proposed ADMM-based algorithms offer moderate communication and computational complexity.
• The methods are demonstrated to be scalable and effective on benchmark datasets.

Conclusions:

• The novel distributed graph hashing model effectively learns hash functions from distributed data.
• The proposed algorithms provide a scalable and efficient solution for distributed ANN search.
• This work advances hashing techniques for decentralized big data environments.