Document Ai: Research Notes
Status
Working note / experiment plan. No completed benchmark results are claimed here.
1. Scope and motivation
These notes organize a possible evaluation of layout-aware representations for documents and irregular visual text. The central question is whether the proposed change improves the target behavior under a matched training and evaluation budget. The note deliberately separates hypotheses from observations so that future results can be added without rewriting the rationale.
2. Context
Research on document ai often mixes improvements from architecture, data scale, preprocessing, and compute. A useful comparison therefore needs controlled baselines and explicit reporting of resource use. For this topic, the main confound is that results can be dominated by OCR quality, language coverage, and annotation conventions.
3. Working hypothesis
A focused change to the representation or interaction mechanism may improve entity-level F1 without increasing deployment cost disproportionately. The hypothesis should be rejected if gains disappear after matching parameter count, data exposure, or tuning budget.
4. Proposed approach
The first implementation should keep modality-specific preprocessing simple, project inputs into a shared representation space, and isolate the new component behind a small interface. Baselines should include a comparable model without the component and a stronger off-the-shelf reference. Any optimization should be applied to all systems, not only the proposed one.
5. Evaluation plan
| Dataset | Role | Primary measure |
|---|---|---|
| FUNSD | primary evaluation | entity-level F1 |
| SROIE | transfer / robustness | character error rate |
| CORD | transfer / robustness | exact match |
Planned comparisons include a matched-capacity baseline, an ablation that removes the proposed component, and an out-of-domain transfer check. Default training values for the first controlled run are learning rate 0.0002, batch size 48, and 5 independent seeds. These are planning values, not claims about a finished experiment.
6. Reproducibility checklist
- Separate model selection from final evaluation.
- Run at least one out-of-domain test.
- Track failed runs as well as successful runs.
- Document every exclusion rule.
7. Failure modes and responsible use
The analysis should report subgroup and category-level failures instead of relying only on a single aggregate score. Particular attention is needed because results can be dominated by OCR quality, language coverage, and annotation conventions. No production use is recommended without task-specific validation, data review, and an assessment of privacy and bias.
8. Open questions
- Which gain survives when the compute budget is matched?
- Does the proposed component improve calibration as well as the primary metric?
- How sensitive is the conclusion to preprocessing and random seed?
References
[1] Xu et al., LayoutLM, 2020. [2] Kim et al., Donut, 2022. [3] Jaume et al., FUNSD, 2019.