Research system
Drift-Robust Explainable TinyML for Electronic-Nose Sensing
Chronological Evaluation, Resource-Aware Explanations, and Reproducible Edge Deployment
This project studies three linked ideas: (1) electronic-nose sensors drift over time, degrading naive machine-learning performance; (2) explanation methods for those models must themselves be evaluated for cost and stability, not assumed reliable; and (3) any claim about TinyML deployment must be backed by real embedded measurements, not software proxies. The site below distinguishes executed evidence from planned work throughout — see the reproducibility policy.
DatasetChronological MLDrift evaluationExplainabilityQuantizationnRF52840PPK2Reproducible artifactsPublication
Explore this as an interactive 3D process view →
See How It Works3-Minute ThesisExplore Digital TwinExplore MethodologyView ExperimentsReproduce ResearchView SourceOpen PaperHugging Face
Dataset, at a glance
Observations
13910
Features
128
Chronological batches
10
Gas classes
6
Full dataset documentation and verified archive hash →
Pipeline status
EXECUTED
16/ 27 stagesPROTOCOL FROZEN
2/ 27 stagesFAILED
2/ 27 stagesBLOCKED
6/ 27 stagesRUNNING
1/ 27 stages