Frank-Wolfe-based Multi-task Learning for Historical Document Restoration
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
During the last few years, research in historical document restoration and understanding (HDRU) has gained increasing popularity. One major problem facing HDRU is the presence of degradation, which renders historical documents unreadable. Although promising results have been obtained, these methods lack the ability to generalize across different datasets. Also, multiple pre-processing and post-processing steps are used, which add more computational complexity and make inference unpractical in real-life settings. Deep Learning (DL) has been successfully used to solve various supervised learning problems in computer vision, where labeled datasets are readily available. However, in HDRU, large annotated historical document datasets are not available. In this paper, we propose an efficient multitask learning (MTL) approach that is based on jointly training self-supervised and supervised learning modules. In the self-supervised learning module, we define two tasks that can be trained with unlabeled data. The first task consists of denoising, and the second task is to learn handwritten characteristics (text orientation). In the supervised learning module, a small subset of labeled data is used to perform text extraction or binarization. All the tasks are formulated as a multi-objective Frank-Wolfe-based optimization problem. We show that convergence to a Pareto optimal solution of jointly training multiple tasks together improves the overall invariance and accuracy of the model. DIBCO 2010-2017 datasets were used for training and DIBCO 2018 for testing. We achieved state-of-the-art results with an F-Score measure of 91.21.
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 | 0.001 | 0.000 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.001 | 0.000 |
| Science and technology studies | 0.001 | 0.000 |
| Scholarly communication | 0.000 | 0.001 |
| Open science | 0.001 | 0.000 |
| Research integrity | 0.000 | 0.001 |
| Insufficient payload (model declined to judge) | 0.003 | 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