Libra: In-network Gradient Aggregation for Speeding up Distributed Sparse Deep Training.
Heng PanPenglai CuiZhenyu LiRu JiaPenghao ZhangLeilei ZhangYe YangJiahao WuJianbo DongZheng CaoQiang LiHongqiang Harry LiuLaurent MathyGaogang XiePublished in: CoRR (2022)
Keyphrases
- distributed network
- peer to peer
- computer networks
- communication overhead
- network traffic
- communication cost
- data transfer
- recurrent networks
- communication networks
- radial basis function network
- heterogeneous networks
- network nodes
- distributed systems
- supervised learning
- high dimensional
- cooperative
- peer to peer networks
- distributed environment
- single point of failure
- network model
- wide area network
- camera network
- central server
- network management
- mobile sensor
- load balance
- sparse data
- social networks
- training process
- recurrent neural networks
- network structure
- multi agent