Traditional user interfaces rely on users explicitly setting preferences to personalize their experience. However, on-device machine learning (ML) offers the potential to create adaptive UIs that dynamically adjust in real-time based on user behavior, reducing the need for manual configuration.

This talk explores the feasibility of deploying ML models directly on the client-side, focusing on performance trade-offs, resource constraints, and user experience considerations. We will examine the challenges and benefits of running ML models in browsers or edge devices, discussing whether this approach can lead to smarter, more responsive UIs without compromising performance.

Key Discussion Points:

1. The Limitations of Explicit User Preferences

   • How current UI personalization relies on manual user input
   • Challenges in creating truly seamless adaptive experiences

2. On-Device ML for Implicit UI Adaptation

   • Feasibility of embedding ML models within the frontend
   • Practical use cases: behavioral analysis, theme adaptation, interaction prediction

3. Performance & Technical Feasibility

   • Benchmarking ML models in browsers (TensorFlow.js, ONNX.js, WebAssembly)
   • Computational trade-offs: latency, memory usage, battery impact

4. User Experience & Ethical Considerations

   • Does adaptive UI enhance or frustrate users?
   • Privacy benefits vs. security risks of on-device ML