Paper Note Format

A paper note should preserve what the paper claims, what the method changes, how it is evaluated, and how it connects to reusable concepts. It should not become a copied abstract or a raw reading log.

For public paper reviews intended to read like a beginner-friendly blog post, use Longform paper review guide before drafting.

For papers that combine AI methods, computational biology objects, and mathematical objectives, start from AI Computational Biology Math paper template.

Suggested Shape

• Citation: verified title, authors, venue or preprint status, and link.
• Status: one of the values in Reading status.
• Question: what problem the paper tries to solve.
• Method: core model, objective, data, and evaluation protocol.
• Results: what is claimed, with metric and benchmark context.
• Claims: extracted statements following Claim extraction.
• Evidence: claim-to-result mapping using Evidence table.
• Benchmark: task, split, metric, and scope using Benchmark card when needed.
• Baselines: what the method is compared against.
• Ablations: which components explain the result.
• Limits: assumptions, failure modes, missing comparisons, or leakage risks, using Limitation taxonomy when useful.
• Artifacts: public code, data, splits, configs, weights, logs, predictions, and environment using Artifact availability.
• Reproducibility: public code, data, config, and run details following Reproducibility checklist.
• Reproduction plan: smallest follow-up experiment, if the paper is worth checking directly.
• Connections: links to concepts, research maps, and projects.

Depth Levels

Not every paper note needs the same length.

Depth	Target	Required Content
Seed note	5-10 min read	citation, question, claim, method, one equation if relevant, evidence table, limits, links
Full note	15-25 min read	seed note plus architecture walkthrough, tensor/object contracts, ablations, benchmark details, later variants, implementation risks
Longform review	25-40 min read	explanatory Korean article or synthesis post; keep the paper note as the canonical reference

Architecture papers should usually start as seed notes and be promoted to full notes if they anchor a reusable model family.

Formula Slot

When a method depends on a mathematical object, include the canonical equation and define symbols. For example, a supervised objective is:

$$\hat{\theta }=\arg \underset{\theta }{\min }\frac{1}{n}\sum _{i=1}^n\mathcal{L}(f_{\theta }(x_i),y_i)$$

Checks

• Are metadata and links verified?
• Are claims separated from personal interpretation?
• Are metrics tied to a split and evaluation protocol?
• Are baselines and ablations strong enough to support the claim?
• For chem-bio papers, are assay harmonization, negative sets, activity cliffs, and applicability domain addressed?
• Are figures or tables summarized rather than copied?
• Are uncertain details marked as unresolved?

Related

• Papers
• Paper review workflow
• Longform paper review guide
• Reading status
• AI Computational Biology Math contract
• AI Computational Biology Math paper template
• Claim extraction
• Evidence table
• Benchmark card
• Ablation map
• Limitation taxonomy
• Artifact availability
• Reproducibility checklist
• Reproduction plan
• Paper comparison matrix
• Paper brief workflow
• Metric
• Baseline
• Ablation study
• Applicability domain
• test split