Thursday October 28 9:00 PM – Thursday October 28 9:30 PM in Talks I

Deploying a Mobile App on Tensorflow: Lessons Learned

Reshama Shaikh, Nidhin Pattaniyil

Prior knowledge:
Previous knowledge expected
Training a deep learning model


As a data scientist, have you wanted to learn more about how deep learning models are deployed? This talk covers considerations for deploying a model: model size, quantization, framework options, production considerations and optimizations, platform options and performance metrics.


Our Medium article

See our Medium article for a background: Deploying A Deep Learning Model on Mobile Using TensorFlow and React

Intro (5 mins)

  • Introduce deep learning model, dataset
  • Show web app
  • Show mobile app

Review computer vision concepts (5 mins)

  • Top frameworks for training a model (TensorFlow, PyTorch)
  • Popular models for web/mobile deployment (ResNet, MobileNet)

Deploying as a service (web app) (5 mins)

  • Benefits of deploying a model as service
  • Overview of different deployment options: TensorFlow-serving vs simple flask
  • Production considerations: latency, size of model
  • Production optimizations: How to apply post training quantization to improve the latency

Deploying on mobile (10 mins)

  • Benefits of running a model natively on a users phone
  • Overview of different deployment options:
  • Platform specific options: Core ML (Apple), vs TensorFlow Lite (Google)
  • Cross platform options: React Native with TensorFlow.js
  • Sample Code for inference
  • Performance metrics on different devices

Q&A (5 mins)