Prophet: Speeding up Distributed DNN Training with Predictable Communication Scheduling.
Zhenwei ZhangQiang QiRuitao ShangLi ChenFei XuPublished in: ICPP (2021)
Keyphrases
- communication overhead
- training process
- communication cost
- distributed control
- meeting scheduling
- distributed systems
- spatially distributed
- computer networks
- cooperative
- distributed network
- flow control
- training phase
- dynamic scheduling
- distributed computation
- communication systems
- computational grids
- distributed environment
- fully distributed
- supervised learning
- communication networks
- scheduling algorithm
- lightweight
- online learning
- scheduling problem
- training set
- multi agent
- global knowledge
- group communication
- flexible manufacturing systems
- single point of failure
- information dissemination
- resource allocation
- test set
- open systems
- geographically distributed
- multi party
- communication channels
- mobile agents
- training examples
- training samples
- artificial neural networks
- data sets
- hearing impaired