Noa Zilberman Seminar

The rat race between user-generated data and data processing systems is currently won by data.

As we apply machine learning to more and more use cases, we increase processing requirements, while data volume keeps growing. Surprisingly, network devices may turn out to be our saviours.

This talk will provide background on programmable network devices, and the emergence of in-network computing - the offloading of applications to run within network devices. It will discuss the current role of network devices in training and inference, including stories of success and failure. Last, I will introduce IIsy, our in-network classification framework, and demonstrate through several use cases why network devices are the machine learning accelerators of tomorrow.