10-15% cost of interpreation seems like a non-starter to me when they are trying to squeeze in every level of performance. I'm seeing some convergence in the industry regarding "inference models" file formats / representations
I would challenge that assertion. Your going to get a larger drop on an android phone when thermal throttling kicks in.
- https://github.com/tensorflow/tfjs-node
- https://developer.android.com/ndk/guides/neuralnetworks/index.html
- https://developer.apple.com/documentation/coreml
- https://discourse.wicg.io/t/api-set-for-machine-learning-on-the-web/2491
- https://github.com/AngeloKai/ml-for-web
- https://github.com/WICG/shape-detection-api
- https://www.tensorflow.org/api_docs/swift/
- https://github.com/intel/webml-polyfill/blob/master/docs/api.md