The Speakers

 Tree Network Design for Faster Distributed Machine Learning Process with Distributed Dual Coordinate Ascent

Abstract: This project focuses on optimizing the efficiency of distributed machine learning (ML) by designing an optimal tree-based network structure. In the distributed machine learning process, coordinating across multiple nodes can become a bottleneck due to communication delays and network latency. To address this challenge, the project proposes a hierarchical, tree-shaped architecture that minimizes communication costs between nodes, making the distributed learning process faster and more scalable. The core of the methodology involves using Distributed Dual Coordinate Ascent on a Tree network (DDCA-Tree), an algorithm well-suited for large-scale machine learning tasks. DDCA-Tree breaks down complex optimization problems into smaller, manageable subproblems that can be solved independently by nodes in a distributed system on any tree network. The proposed tree network design complements this approach by ensuring that efficient communication flow between nodes, further accelerating convergence rates. This approach is particularly useful for large-scale applications such as big data analytics, neural networks, and other machine learning algorithms where data is distributed across multiple machines. By leveraging the tree structure and DDCA, this project aims to enhance the scalability, speed, and overall performance of distributed machine learning processes.