This is not an officially supported Google product.
chowdhery@, mcrouse@, ryansturm@, mehrdadh@, Last updated: July 1, 2020
The competitiveness of browsers in future will be determined by empowering the user to control their data’s privacy while being optimized for lightening fast user-interactions. AI has enabled web applications to offer several new features, for example text suggestions or recommending the next video to watch. ML models require user data to train accurate models which directly conflicts with empowering users’ privacy. On-device ML enables new paradigm empowering users to control their privacy because the user data never leaves their device and the deployed models may be personalized if the user would like to. So far Google has seen successful deployment of on-device ML models in a wide-range of applications, including Nest, Assistant, Gmail, and Youtube. In this intern project, we investigate the regime in which on-device ML models can be successfully deployed in Chrome to enable new capabilities & features. Mobile browsers introduce additional systems challenges related to memory and/or latency constraints that require investigation for on-device ML models to perform well.