W3C
published a public working draft of the Web Neural Network API (referred to as “WebNN”). The JavaScript API was developed by W3C and some stakeholders as a means to allow hardware acceleration of neural network inference through the browser.
W3C established a web machine learning working group earlier this year.
Google and Microsoft are among the companies participating in the working group, and they hope to use dedicated machine learning hardware accelerators and native instructions in web browsers.
The design of WebNN considers use cases such as person detection, face recognition, super-resolution, image captioning, sentiment analysis, noise suppression, etc.
Regarding concerns about any website being able to use machine learning hardware, according to the official introduction, the security considerations of this new API include: disabling it by default in all cross-domain frameworks and is associated with the security of WebGPU and WebGL.