Our decentralized sparse training technique, which requires that each local model in Dis-PFL only maintain a set number of active parameters during the whole local training and peer-to-peer communication process, substantially reduces communication and computation costs.
<u>What is DisPFL?</u>
It is suggested that personalized federated learning be used to address the issue of data heterogeneity among clients by developing unique, locally adapted models specifically for each user.
But because existing works are frequently constructed in a centralized manner, there is a great deal of communication pressure and vulnerability in the event that the central server is attacked or fails.
Using a decentralized (peer-to-peer) communication protocol called Dis-PFL, we offer a unique personalized federated learning framework in this study that uses personalized sparse masks to tailor sparse local models on the edge.
<h3><u>What is Communication?</u></h3>
Every communication has a sender, a message, and a recipient at a minimum. Although it might seem straightforward, communication is actually a very complicated topic.
There are a wide variety of factors that can influence how a message is transmitted from the sender to the recipient.
These include our feelings, the surrounding cultural context, the communication method, and even our geographic location.
Accurate, effective, and unambiguous communication is actually very difficult, which is why companies all over the world place such a high value on these abilities.
Detailed tests show that Dis-PFL dramatically reduces the communication bottleneck for the busiest node among all clients while also achieving improved model accuracy with lower computation costs and communication rounds.
Know more about communication with the help of the given link:
