Domain Adaptation in LIBS Streaming Using Transfer Learning and Self-Learning for Soil Classification
Bibliographic record
Abstract
Laser-induced breakdown spectroscopy (LIBS) has become a promising technology for determining the chemical composition of soil samples. The application of machine learning (ML) has accelerated the development of LIBS in soil analysis. However, ML for LIBS is challenging because 1) building robust ML models requires large calibration datasets while LIBS experimental data is typically limited, further limiting data streaming, 2) matrix effects deteriorate LIBS performance, and LIBS data is sensitive to changes in the apparatus, causing emission lines distribution highly variable. These issues may cause concept drift in LIBS streaming and make the relation between the input and the target spectra variable over time, leading to an ML model constructed for one LIBS system becomes less applicable to a different LIBS system. We propose transfer learning to conquer the challenges of limited data. We then use domain adaptation with self-learning to self-adapt to the domain shift in LIBS streaming to alleviate the matrix effects and improve the model generalization. To test the efficiency of the proposed method, we conduct experiments on the same soil samples but with different experiment parameters, such as wavelength and laser energy. The collected spectra are fed into our model in chunks. The EMSLIBS dataset used in the 2019 EMSLIBS competition is utilized to construct the transfer-learning model, which serves as the foundation for the model developed using our experimental data. Following this, self-learning is undertaken for each chunk by repeatedly predicting the current chunk using the model trained by previous chunks and then taking the confident predictions as pseudo-labels for co-training the model. It is shown that the average accuracy is improved by 9% with transfer learning and up to 15% better with transfer learning and self-learning during data streaming compared to an ML model that does not implement transfer learning and support adaption.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame distilled prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.
Codex and Gemma teacher scores by category
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.000 | 0.000 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.000 | 0.000 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.000 | 0.000 |
| Open science | 0.000 | 0.000 |
| Research integrity | 0.000 | 0.000 |
| Insufficient payload (model declined to judge) | 0.000 | 0.000 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one teacher head, not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".