Deep Learning for the Detection of Tabular Information from Electronic Component Datasheets
Why this work is in the frame
A frame that forgets how it found something cannot be audited. These are the routes that admitted this work.
Bibliographic record
Abstract
The global electronic components supply chain consists of tens of thousands of e-component manufacturers who fabricate over a billion distinct components. These are described in datasheets that differ in style, layout and content, and frequently publish the salient product information in tables. Keeping up-to-date on this information consumes a great deal of human effort and corporate resources. Based on the motivation that AI-based techniques are strong candidates to minimize human intervention in many applications, in this paper, we aim at the first stage of this problem and conduct a comparison of deep learning methods in detecting tabular elements in these documents. Deep learning-based object detectors are shown to be state of the art in detection tasks in different domains therefore we chose two cutting-edge models to adapt to this field, namely Faster-RCNN and RetinaNet. We use backbone networks which are pre-trained on visually salient datasets then employ transfer learning techniques to adapt to our domain. We compare the two networks under two different datasets, namely a dataset that is widely used in academic studies and a private dataset that is used by the suppliers in real supply chains. Our numerical results show that the two networks adapt well to the domain with Faster-RCNN exhibiting marginally better precision with more than 1% difference. However, RetinaNet stands out with promising recall values indicating Feature Pyramid Network architecture can potentially detect technical documents better.
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.
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 it