Spectroscopic sensing · 2025–present
PRISM-SERS and few-shot microbial recognition
Developing spectral mapping for analytes unseen during training and studying few-shot recognition across instruments and sample conditions.
Qatar University · Ongoing research · Manuscript in preparation
Approach
Can SERS spectra be mapped across instruments for analytes unseen in training, and recognised from only a few examples?
PRISM-SERS studies how portable-device spectra can be mapped to a laboratory reference when target analytes are held out of training, using unpaired data and no class labels at inference. A second workstream benchmarks time-series foundation-model representations for few-shot classification with separated support and query samples.
- Map portable spectra to a reference
- Hold target analytes out of training
- Benchmark few-shot classification
My contribution
I develop the spectral preprocessing, transfer and classification workflows and design comparisons that account for sample identity and differences between acquisition domains.
Study scope
This work is ongoing, and no results have been published yet. The two workstreams address different questions: spectral mapping tests recovery of comparable spectra, while few-shot experiments assess classification from limited examples.