Thesis¶
Structure, outline, research questions, and comparison criteria for the thesis.
Overview¶
This thesis compares web frameworks for running ML models in the browser — evaluating TensorFlow.js, ONNX Runtime Web, Transformers.js, and MediaPipe across performance metrics and developer experience.
Thesis Structure¶
Part I: Introduction & Background¶
- Problem statement: Why browser-based ML matters
- Research questions
- Methodology (comparison criteria)
- Browser ML landscape overview
Part II: Technical Analysis¶
- Runtime architecture deep-dives
- Backend capabilities (CPU, WASM, WebGL, WebGPU, WebNN)
- Model distribution strategies
- Performance characteristics
Part III: Evaluation & Results¶
- Benchmark methodology
- Quantitative results (load time, inference, memory)
- Qualitative assessment (ease of integration, documentation, community)
- Cross-browser analysis
Part IV: Conclusion¶
- Key findings
- Recommendations
- Future work
Comparison Criteria¶
| Metric | Description | Measurement |
|---|---|---|
| Load time | Model download + initialization | performance.now() delta |
| Cold inference | First prediction (no warmup) | Mean of 5 runs after init |
| Warm mean | Subsequent predictions | Mean of 10 warm runs |
| Memory delta | Heap before vs after inference | Chromium performance.memory |
| Backend detection | Which backends available | Runtime capability queries |
| Prediction sanity | Output validity check | Top-5 class probability sum |