Federated learning is a technique that allows an algorithm to be trained across multiple servers or edge devices that store local data samples without having to share or exchange that data.
Fremont, CA: Edge computing would be critical to the Internet of Things’ progress (IoT). IoT devices produce massive quantities of data, ranging from smartphones and smartwatches to tiny computers embedded in machines and infrastructure. This is done in the cloud, and the relevant data is sent back to the unit, instructing it on how to react. However, this setup is inefficient for time-critical applications like medical devices, which must process and respond to sensor data almost immediately due to latency. The use of the cloud often restricts the ability to install IoT devices in areas with limited or no internet access. This may be in space, deep under the sea, or regions of the Earth’s surface where the internet is still underserved.
Edge computing will be used in 2021 to resolve these issues by relocating data processing from a central server to a location closer to where it was collected. This is possible thanks to specialized accelerator chipsets built by companies like Nvidia and Intel that can run relatively advanced machine-learning algorithms far from the conventional cloud’s powerful centralized servers and sometimes even on the system itself.
This will virtually eliminate latency. Also, there are developing software and hardware optimization strategies like model quantization to speed up cloud-based processing and reduce latency even more. These advancements would aid in the global adoption of sophisticated IoT products.
Edge computing would also be able to provide more localized control over our data. Federated learning is a technique that allows an algorithm to be trained across multiple servers or edge devices that store local data samples without having to share or exchange that data. This ensures that confidential data, such as medical or proprietary information, can be stored on the computer and held safe.