Expert lithological labels do not improve learning when used to predict marine sediment properties
Abstract
Scientific drilling has produced large archives of physical, chemical and magnetic measurements linked to descriptions of the rock recovered. These archives are mostly incomplete: instruments sample at different intervals, recovery is partial, and destructive analyses preclude later measurement of the same material. We train an self-supervised embedding model on the LILY ocean drilling database using six routinely logged properties (bulk density, magnetic susceptibility, natural gamma radiation, and red, green and blue reflectance) across 236,963 observations from 296 boreholes, with a masked variational autoencoder that has a heteroscedastic Gaussian decoder and learnable mask tokens. The embeddings separate geologist-assigned lithologies as well as a variant trained with those labels, and increasing the weight on the labels degrades reconstruction while leaving classification unchanged. The labels are recoverable from the measurements geologists use to assign them, so training on them adds nothing. Lithological categories compress continuous measurements into classes, and compressing them discards what the measurements record about the processes that produced the rock. The embeddings also predict physical properties that were absent from training, on boreholes the model has never seen. While depth should account for this since sediments compact with burial, it does not. On held-out boreholes depth alone reaches a median R2 of -0.02, against 0.20 for the embedding.
Related
- Response to comment on “Using genomic data and machine learning to predict antibiotic resistance: A tutorial paper”
- Lithological and mesoclimatic controls on denudation rates in transform margin crystalline massifs: Insights from cosmogenic 10Be and 26Al in semi-arid Brazil
- Fe Isotope Fractionation during Early Diagenetic Pyrite Formation: Constraints from Reaction-Transport Modeling of the Santa Barbara Basin Sediment
- Late Holocene paleoenvironmental and paleoclimatic changes recorded in a sediment core from Sidi Ali El Mekki lagoon (Northeast Tunisia)
- Anthropogenically induced shifts in sediment vegetation impact coastal benthic microbiomes
BibTeX
@misc{aiken2026expert,
title = {{Expert lithological labels do not improve learning when used to predict marine sediment properties}},
author = {John M. Aiken and William Gilpin and Thorsten W. Becker},
howpublished = {arXiv},
year = {2026},
month = oct,
doi = {10.48550/arxiv.2610.36410},
eprint = {2610.36410},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2610.36410},
note = {CC BY 4.0},
}
RIS
TY - UNPB
TI - Expert lithological labels do not improve learning when used to predict marine sediment properties
AU - John M. Aiken
AU - William Gilpin
AU - Thorsten W. Becker
PY - 2026
DA - 2026/10/28
PB - arXiv
DO - 10.48550/arxiv.2610.36410
UR - https://arxiv.org/abs/2610.36410
AB - Scientific drilling has produced large archives of physical, chemical and magnetic measurements linked to descriptions of the rock recovered. These archives are mostly incomplete: instruments sample at different intervals, recovery is partial, and destructive analyses preclude later measurement of the same material. We train an self-supervised embedding model on the LILY ocean drilling database using six routinely logged properties (bulk density, magnetic susceptibility, natural gamma radiation, and red, green and blue reflectance) across 236,963 observations from 296 boreholes, with a masked variational autoencoder that has a heteroscedastic Gaussian decoder and learnable mask tokens. The embeddings separate geologist-assigned lithologies as well as a variant trained with those labels, and increasing the weight on the labels degrades reconstruction while leaving classification unchanged. The labels are recoverable from the measurements geologists use to assign them, so training on them adds nothing. Lithological categories compress continuous measurements into classes, and compressing them discards what the measurements record about the processes that produced the rock. The embeddings also predict physical properties that were absent from training, on boreholes the model has never seen. While depth should account for this since sediments compact with burial, it does not. On held-out boreholes depth alone reaches a median R^2 of -0.02, against 0.20 for the embedding.
ER -