Deep learning black box and pattern recognition analysis using Guided Grad-CAM for phytolith identification
Bibliographic record
Abstract
BACKGROUND AND AIMS: In this article, visual explainers are applied to give transparency to the black box of a trained VGG19 model for the identification of multi-cell phytoliths of the Avena, Hordeum and Triticum genera. The aim is to demonstrate its proper learning by visually highlighting the phytolith characteristics that the deep learning model uses to classify these phytoliths; we then compare the model's methods with those employed manually by archaeobotanists. METHODS: The visual explainers used for this purpose are Grad-CAM, Guided Backpropagation and Guided Grad-CAM, the last being a combination of the previous two. This combined tool not only highlights the most relevant regions when classifying phytoliths on microscope images, but also emphasizes every detail within those areas. KEY RESULTS: The importance of the wave pattern as a decision-maker (key identifying characteristic) when classifying phytoliths has been demonstrated for 91 % of the microscope images. Similarly, the papillae have been a key in 86 % of Avena images, in 94 % when images included papillae, and the dendritic long-cell shape in 38 % of Triticum images. CONCLUSIONS: The analysis of the microscope images using Guided Grad-CAM has validated the established patterns in phytolith identification, such as highlighting the significance of the wave pattern. Additionally, it revealed that varying phytolith characteristics might be prominent for different genera and led to the discovery that dendritic long-cell shape, as an independent category, is also distinctive. This research is part of an effort to establish a set of computer vision best practices in computational archaeology.
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 | 0.001 |
| 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".