You can download and install the latest version of our simple Android demo
to participate in collaborative benchmarking, optimization, autotuning
and co-design of AI-based applications: APK Version 2.12.
Just make sure that your Android device is configured to allow
installation from sources other than the Play Store before
downloading and installing our CK-based Android apps.
You can check it as follows:
You can see all aggregated results from these experiments (speed, accuracy, costs) across diverse mobile phones in the public repository (note that we gradually move all CK SOTA scoreboards here). You can also check the collective training set with image misclassifications.
Our applicaiton downloads slightly modified AI pipelines assembled from cKnowledge components (frameworks, models, datasets) and runs them across diverse platforms and environments provided by volunteers similar to SETI@home to collaboratively find the most efficient AI solutions in terms of speed, accuracy, costs and other metrics or automatically detect software and hardware bugs (see paper 1, paper 2 and paper 3).
We use this open-source technology is used in the open and reproducible ACM ReQuEST tournaments to co-design Pareto-efficient deep learning stacks (speed, accuracy, costs) and in the open project with Raspberry Pi foundation to crowdsource compiler autotuning and machine learning!
You can check out another Android application to help crowdsource benchmarking and autotuning of popular compilers (LLVM, GCC): You can now download, install and run two CK-based Android applications to