Deploying your model to mobile app: Offline or Online?
summary
Shows a real fraud detection model at about 84% accuracy deployed as an online API using Flask-style tooling and cloud hosting, and contrasts that with on-device examples like a plant disease classifier where a ~200–250 MB model is impractical to bundle. Covers concrete tools: FastAPI, lightweight servers, and one‑line exposure with ngrok or hosting on PythonAnywhere, plus AWS and model servers for production. Lists tradeoffs: easy scaling and instant updates online versus offline privacy, zero per‑request cost, but higher device CPU/RAM needs and update friction.
AI-written summary · about
chapters (6)
- 0:00Introduction to model deployment
- 4:29Free tools for deployment
- 11:55Advantages of online deployment
- 15:10Advantages of offline deployment
- 23:16Choosing between online and offline
- 28:35Summary and examples
related talks
15mEmbed Your Machine Learning Model in Your App?
50mBuilding Intelligent Mobile Apps using Machine Learning
36mMachine Learning for Mobile Developers
35mA simple way to validate and monitor the performance of your ML Applications.
46mProduction Model Deployment
47mMachine Learning on Android Demystified
29mRunning Deep Learning Models on Mobile with optimised speed
40mMLSecOps with Automated Online and Offline ML Model Evaluations on Kubernetes