ML论文复现实验代理工作流设计器

You are an expert ML research engineer. I want to reproduce the results from a research paper. Help me create a complete reproduction workflow.

Paper Details

• Paper title: [PAPER_TITLE]
• Key contribution: [BRIEF_DESCRIPTION]
• Original repo (if any): [REPO_URL_OR_NONE]

Your Tasks

1. Paper Analysis

• Identify the core algorithm/method
• List all hyperparameters mentioned
• Note the datasets used and evaluation metrics
• Flag any ambiguities or missing details

2. Implementation Plan

• Break down the implementation into modules
• Suggest the tech stack (PyTorch/JAX/etc.)
• Identify which parts can use existing libraries vs. custom code
• Estimate implementation complexity for each module

3. Data Pipeline

• How to obtain/prepare each dataset
• Preprocessing steps
• Data loading and augmentation strategy

4. Training Recipe

• Exact training configuration
• Learning rate schedule
• Hardware requirements estimate
• Expected training time

5. Evaluation Checklist

• Metrics to compute
• Baseline comparisons
• Statistical significance tests
• Ablation studies to run

6. Common Pitfalls

• Known issues with this type of method
• Numerical stability concerns
• Reproducibility gotchas

Provide the workflow as a structured markdown document I can use as a project plan.