A comprehensive analysis of cross-platform machine learning runtimes on mobile applications

High level description

Cross-platform frameworks like React-Native and Flutter has enabled developers to use a single language and code base to develop apps for both iOS and Android. With the growing popularity for machine-learning based applications, cross-platform options for machine-learning runtimes has emerged. An example of this is the ONNX runtime which enables machine-learning models to run on a large verity of devices and operating systems including mobile devices. Today Apple and Google have their own runtimes for iOS (Core ML) and Android (ML Kit). This thesis aims to investigate and compare the native mobile runtimes for Android and iOS and see how they compare to a cross platform solution like ONNX.

Who are we looking for?

Bachelor/Master of Science in Computer Science/Engineering

Project description

In this Thesis, design and implement three different mobile apps. A naive Android app, a native iOS app and lastly a cross platform app. The apps should have the same functionality and implement the same ML model and compare how they perform and if there are any drawbacks to using the cross-platform solution.

Purpose and Scope

In this thesis, investigate these questions:

• Are there any significant drawbacks to use the cross-platform solution?
• How does the development experience compare between the different ML runtimes?
• How does the different ecosystems compare in terms of availability of finished ML models?
• Is it possible to convert an ML model from one runtime to the other?

References:

ML Kit, https://developer.android.com/ml#ml-kit-sdks-ready-to-use-for-common-user-flows

Core ML, https://developer.apple.com/machine-learning/core-ml/

ONNX runtime for React Native, https://onnxruntime.ai/docs/get-started/with-javascript.html#onnx-runtime-for-react-native

TensorFlow Lite for Flutter, https://pub.dev/packages/tflite_flutter