ChameleonAPI: Automatic and Efficient Customization of Neural Networks for ML...
Talkat 18th USENIX Symposium on Operating Systems Design and Implementation (OSDI '24) · 2024-07-10..07-12·
summary
Paper builds ChameleonAPI that parses app code to extract a decision-making summary, then uses it to construct an application-specific loss function and resample training data to fine-tune lightweight vision and NLP models. Evaluation on 57 apps and a 77-app empirical study shows per-app customization boosts application accuracy from 69% to 84% and reduces incorrect application decisions by 43%, with low customization latency and deployment via existing ML API interfaces.
AI-written summary · about
chapters (5)
- 0:00ChameleonAPI intro
- 1:27Generic models vs real apps
- 9:15ChameleonAPI solution
- 12:48Evaluation results
- 15:42Conclusion & Q&A
related talks
39mCustomize Your App with MLKit
10mMachine Learning for mobile
35mA simple way to validate and monitor the performance of your ML Applications.
15mEmbed Your Machine Learning Model in Your App?
50mBuilding Intelligent Mobile Apps using Machine Learning
47mMyths and Reality of Android Machine Learning
36mKeynote! The opportunity for Android devs with AI and ML
36mLeveraging Machine Learning APIs to Write Better Software