In-Network ML

In-Network Machine Learning (ML) is the offloading of machine learning tasks to run within network devices, such as network switches or NICs.

Our research focuses on in-network ML on unmodified, off-the-shelf network devices. We have done significant work in this area, such as:

A Survey of in-network ML on programmable network devices - the best place to get introduced to the research area.

IIsy, our first work which introduced the mapping on classic ML algorithms to programmable pipelines. A later version of the work also introduced hybrid-deployment of in-network ML.

Planter, a modular framework for automated in-network ML deployment.

DINC, a modular framework for distributed in-network computing and in-network ML.

P4Pir, a solution for smart traffic analysis on IoT gateways.

FliP4, a distributed in-network attack detection framework based on federated tree models.

LOBIN and Linnet, building limit order books within switches for algorithmic trading.

QCMP, Load Balancing using In-Network Reinforcement Learning within the data plane.

INCS, In-network detection and mitigation of e-commerce bot traffic.

And more in the pipeline!

We acknowledge support from VMWare, Intel, NVIDIA, AMD (Xilinx) and EU Horizon/UKRI Guarantee SMARTEDGE project.