AgentSkillsCN

Search Papers

检索与归一化流、密度估计及生成模型相关的学术论文

SKILL.md
--- frontmatter
description: Search for academic papers related to normalizing flows, density estimation, and generative models
allowed-tools: WebSearch, WebFetch, Write, Read, Glob
argument-hint: <keywords or topic, e.g. "autoregressive normalizing flows" or "Jacobian determinant computation">

Search Papers

Search for academic papers relevant to BreezeForest and normalizing flow research.

Search query: $ARGUMENTS

Task: Find and Organize Relevant Papers

1. Construct Search Queries

Based on "$ARGUMENTS", construct multiple search queries targeting:

  • arXiv (site:arxiv.org)
  • Google Scholar
  • Semantic Scholar

Use combinations of the user's keywords with domain-specific terms:

  • normalizing flows, autoregressive flows
  • density estimation, universal density estimator
  • Jacobian determinant, triangular Jacobian
  • conditional CDF, cumulative distribution function
  • generative models, flow-based models
  • mixture models, Gaussian mixture

2. Execute Searches

Use WebSearch to search across academic sources. Run at least 3 different query variations to maximize coverage. For each search:

  • Include "arxiv" or "paper" in queries to bias toward academic results
  • Search for both recent papers (2023-2026) and foundational works

3. Fetch Paper Details

For the most relevant results (top 5-10), use WebFetch on the arxiv abstract page to extract:

  • Title
  • Authors
  • Abstract summary
  • Publication date
  • arXiv ID

4. Assess Relevance to BreezeForest

For each paper, assess relevance to BreezeForest's key techniques:

  • Autoregressive architecture: Does it use autoregressive decomposition of joint density?
  • Jacobian computation: How does it handle the Jacobian determinant? (triangular, numerical, exact)
  • Universality: Does it claim or prove universal approximation for densities?
  • Inversion/sampling: How does it handle inverse mapping for generation? (bisection, analytical)
  • Regularization: Does it address overfitting / sample memorization?
  • Mixture models: Does it use mixture components?

5. Output Structured Results

Output results in this format:

markdown
# Paper Search: [query topic]
**Date**: YYYY-MM-DD

## Summary
[1-2 sentence overview of findings]

## Papers Found

### [Relevance: High/Medium/Low] Paper Title
- **Authors**: ...
- **Year**: ...
- **arXiv**: https://arxiv.org/abs/XXXX.XXXXX
- **Key contribution**: [1 sentence]
- **Relation to BreezeForest**: [How this connects to BreezeForest's approach]
- **Techniques**: [e.g., autoregressive flow, coupling layers, continuous normalizing flow]

### [Next paper...]

## Key Takeaways
1. ...
2. ...

## Suggested Follow-up
- [Further search directions or papers to read in depth]

6. Save Results

Save the results to notes/papers/search_YYYY_MM_DD_[topic_slug].md. Create the directory if it doesn't exist.