# Ml Researcher

> Specialized for deep-dive academic research in Machine Learning (RL, Diffusion, VLM, VLA). Use when you need to search ArXiv, trace historical milestones, analyze benchmark datasets/metrics, compare competing solutions, and generate practical 'try-out' instructions. Results are saved to the workspace 'research' folder.

- Skill: `claio0128/ml-researcher` (Agent Skill)
- Install (CLI): `npx skillmds@latest add claio0128/ml-researcher`
- Raw SKILL.md: https://api.skillmd.com/api/skills/claio0128/ml-researcher/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: AI & ML
- Author: claio0128 (https://skillmd.com/u/claio0128)
- Updated: 2026-09-22
- Page: https://skillmd.com/skills/claio0128/ml-researcher

---


# ML Researcher

## Overview
This skill enables high-signal academic research and practical experimentation. It bridges the gap between theoretical understanding (papers) and hands-on application (code/data).

## Core Capabilities & Workflow

### 1. Problem Definition & Real-World Impact
Identify the core challenge and practical applications.
- **Problem Statement:** What fundamental gap/problem is this model or method trying to solve?
- **Real-Life Examples:** Provide 2-3 concrete scenarios where this technology is applied (e.g., "Automating complex medical document analysis" or "Providing spatial awareness for autonomous warehouse robots").

### 2. Trend Discovery & Initial Research
Identify the most recent or influential paper/trend on a specific ML topic.
- **Workflow:** Search topic (ArXiv/Scholar) → Rank by citation/impact/recency → Identify the current "state-of-the-art" (SOTA).

### 3. Historical Tracing & Milestones
Trace the development of the topic through its "Related Work" and key references.
- **Milestones:** Identify 3-5 key papers that represent significant shifts in the field.
- **Novelty Analysis:** For each milestone, explicitly state how it claimed novelty over previous work.

### 4. Benchmark & Evaluation Analysis
Identify the "Gold Standard" datasets and metrics used to judge performance in this field.
- **Datasets:** Catalog the most common training and evaluation datasets (e.g., MMMU for VLM, RT-X for VLA).
- **Metrics:** Identify the standard evaluation metrics (e.g., Accuracy, FID, Success Rate, Perplexity).

### 5. Competing & Alternative Solutions
Identify and compare different approaches to the same problem.
- **Search:** Find alternative architectures or algorithms.
- **Analysis:** Create a Pros/Cons comparison table for each solution.

### 6. Code & Data Discovery
Locate the "source of truth" for implementation and training data.
- **GitHub/HuggingFace:** Find official repositories, model weights, and datasets.

### 7. Practical "Try-out" Workflow
Generate actionable instructions for running the model or algorithm.
- **Environment Setup:** Requirements, dependencies, and hardware needs.
- **Execution:** Provide a minimal Python script or shell command for a "toy" demo.

### 8. Report Generation & Persistence
Synthesize findings into a structured report using the `DEEP_DIVE_TEMPLATE.md`.
- **Persistence:** ALWAYS save the final report as a Markdown file in the `./research/` directory.

## Resources

### references/
- **[DEEP_DIVE_TEMPLATE.md](references/DEEP_DIVE_TEMPLATE.md)**: Updated template with sections for Problem Definition and Real-World Examples.
- **[BENCHMARK_GUIDE.md](references/BENCHMARK_GUIDE.md)**: A quick-reference guide for ML benchmarks across domains.

---

## When to use this skill
- When the user asks to "deep dive" into a specific ML topic or paper.
- When you need to understand the "lineage" or "benchmarking standards" of a model.
- When comparing different architectures or training methods.
- When the user wants to "try out" a new model and needs setup instructions.
- When the user wants to document ML research results in the workspace.