In-network aggregation has been proposed to accelerate gradient aggregation in the distributed training of neural networks. My current research focus is on loss-tolerant transmission protocol for federated learning.

Some background can be found in the following papers.

https://www.usenix.org/system/files/nsdi21-lao.pdf

dsa-icnp23.pdf

2023_INFOCOM_INAlloc.pdf

https://arxiv.org/pdf/2305.04279