Machine Learning / Evaluated demonstration
Separate spectral components and highlight signals the model cannot explain.
Actual recording · Explanatory English captions · Original synthetic data
Recorded demonstration of a functional local application. A pilot defines data, access and deployment; this website serves the video and screenshots.
THE PROBLEM
THE WORKFLOW
NMF learns three components from 240 synthetic training spectra. Change mixture proportions, noise and an unfamiliar peak; compare the observed spectrum with its reconstruction and inspect the residual. A compatible wavelength/intensity CSV can also be imported.
RECORDED EVIDENCE
| Check | Cases |
|---|---|
| Documented checks; synthetic data | 99 of 100 |
100 heldout examples: 60 known mixtures and 40 with a controlled new peak. 99 decisions match expectations, retaining and reporting the false positive. Alignment with simulator components is computed only after training. Timings measure the analysis function on CPU, excluding HTTP and training.
WHAT WE COULD BUILD WITH YOU
Instrument, wavelength range, preprocessing, reference data and independent validation.
Synthetic spectra from 400 to 800 nm. Contributions are relative to the model, not chemical concentrations or substance identification. One false positive occurred among 60 normal mixtures; all 40 with the test peak were flagged.
Scope a Machine Learning pilot ↗
Explore demo
Sensors → prediction → detected change
Separate changing inputs from declining model performance.
30 of 30 checks · Original synthetic dataDocumented checks; synthetic data
Explore demo
Trials → uncertainty → next design
Recommend trials using Bayesian optimisation and resource limits.
10 of 10 checks · Original synthetic dataDocumented checks; synthetic data