/research¶
Conduct background research on a topic relevant to your ML project. This command generates a structured research artifact and saves it to the research directory.
Usage¶
What It Does¶
- Analyzes project context -- Reads your codebase, recent experiments, and current challenges to determine what research would be most valuable.
- Gathers information -- Searches for relevant techniques, papers, best practices, and implementation strategies.
- Synthesizes findings -- Produces a structured Markdown document with a clear summary, key references, and actionable recommendations.
- Saves the artifact -- Writes the research note to
.researchpad/experiments/research/with a date-prefixed filename.
Example¶
This might produce a file like:
With contents:
# Learning Rate Scheduling Strategies
## Summary
Explored cosine annealing, warm restarts, and one-cycle policies
for improving convergence on the current dataset.
## Key Findings
- Cosine annealing with warm restarts showed 12% faster convergence
in similar problem domains
- One-cycle policy is particularly effective for fine-tuning
## References
- Smith & Topin, 2019: Super-Convergence
- Loshchilov & Hutter, 2017: SGDR
## Recommendations
1. Try cosine annealing with T_0=10, T_mult=2
2. Compare against current StepLR schedule
Artifact Format¶
Each research artifact is a Markdown file saved to:
.researchpad/experiments/research/YYYY-MM-DD-topic.md
The file follows a consistent structure: title, summary, key findings, references, and recommendations. The dashboard renders these files in the Research panel.