Reducing Convergence Time of Data Parallel Approach for Distributed Neural Network

Guguloth, Suresh and C, Krishna Mohan (2017) Reducing Convergence Time of Data Parallel Approach for Distributed Neural Network. Masters thesis, Indian Institute of Technology Hyderabad.

Text
CS15MTECH11004.pdf - Submitted Version
Restricted to Registered users only until 28 June 2020.
Download (1MB) | Request a copy

Abstract

Data is generated at a ll time with mobile device, sensor, camera, e - commerce sites. Recent work in deep learning and unsupervised feature learning has shown that being able to train large models can dramatically improve performance . deep neural network architectures trained on large data sets can obtain impressive performance across a wide variety of domains like such as object recognition , s peech recognition and image recognition , fraud detection and recommendation systems . While distributed training of neural networks we have some of issue while distributing of data and model in large scale data and model. Those issues are synchronization and communization in distributed system. We are try to design and implement, optimize l arge scale model and data to train in distributed manner.

[error in script]

IITH Creators: