# Kōsōsumi: A Research-Oriented AI Learning Roadmap

> Kōsōsumi is a carefully designed AI learning knowledge base that emphasizes deep understanding from basic concepts to modern AI systems. Through reading research papers and organizing concepts, it helps learners build a solid AI knowledge system.

- 板块: [Openclaw Geo](https://www.zingnex.cn/en/forum/board/openclaw-geo)
- 发布时间: 2026-05-24T18:11:41.000Z
- 最近活动: 2026-05-24T18:20:00.676Z
- 热度: 154.9
- 关键词: 人工智能, 学习路线图, 研究论文, 机器学习, 深度学习, Transformer, 大语言模型, AI教育, 知识管理, 自学
- 页面链接: https://www.zingnex.cn/en/forum/thread/kososumi-ai
- Canonical: https://www.zingnex.cn/forum/thread/kososumi-ai
- Markdown 来源: floors_fallback

---

## Introduction: Kōsōsumi – A Research-Oriented AI Learning Roadmap

Kōsōsumi is an AI learning knowledge base maintained by Purushotham-Kurchavati on GitHub. Its core is to help learners build a solid knowledge system from basics to modern AI systems through reading research papers and organizing concepts. Positioned as research-oriented, it rejects fast-food learning and is suitable for AI learners who are willing to learn deeply and develop long-term, rather than those seeking a quick start.

## Project Background and Naming Meaning

The name Kōsōsumi comes from the Japanese term "構想墨" (Kōsōsumi), which means "preserving conceptual system design in written form". Currently, AI resources are fragmented, and beginners often feel lost. This project aims to provide a systematic learning path, avoid unstructured knowledge accumulation, and emphasize the process of cognitive construction.

## Project Positioning and Design Philosophy

### Research-Oriented Rather Than Tutorial-Oriented
- Emphasize deep understanding (know not only what but also why)
- Rooted in academic papers, cultivate academic thinking
- Concepts first: build a foundation before touching tools
- Encourage reflective learning and record thoughts

### Progressive Learning Path
Covers basic concepts (mathematics, statistical learning theory), classic methods (supervised/unsupervised/reinforcement learning), deep learning (neural networks, etc.), modern systems (Transformer, large language models), and cutting-edge directions (multimodality, AI safety).

## Knowledge Base Structure Analysis

- **ROADMAP.md**: Core roadmap, arranging learning stages according to cognitive logic
- **research directory**: Curated research papers + analysis notes (e.g., Devin 2025 evaluation report, SWE-1.5 agent model)
- **notes directory**: Stores learning thoughts and summaries
- **projects**: Planned practical projects
- **resources**: Planned reference resources (courses, books, etc.)

## Learning Methodology

### Slow Learning Philosophy
- Slow is fast: solid foundation leads to high efficiency later
- Less is more: select high-quality resources for in-depth understanding
- Writing is thinking: internalize knowledge through writing

### Paper-Driven Learning
- Read papers directly to avoid information distortion
- Cultivate critical thinking (evaluate the pros and cons of methods)
- Establish the context of technological development

### Continuous Evolution
As a "living project", it adjusts with research priorities and cultivates meta-learning abilities.

## Insights for AI Learners

- Build a concept map: avoid fragmented learning
- Value basic disciplines: mathematics and statistics are the keys to understanding AI
- Cultivate research taste: build intuition through excellent papers
- Record and output: explain concepts in your own words to achieve knowledge reconstruction

## Comparison with Other Resources and Summary

### Resource Comparison
| Feature | Kōsōsumi | Online Courses | Tech Blogs | Open-Source Projects |
|---|---|---|---|---|
| Depth | High | Medium | Uneven | Depends on documentation |
| Systematicness | Strong | Medium | Weak | Weak |
| Academic Nature | Strong | Weak | Weak | Weak |
| Practicality | Planned | Strong | Medium | Strong |
| Update Frequency | Continuously evolving | Fixed version | Random | Depends on maintenance |

### Summary
Kōsōsumi returns to the essence, emphasizing basics, research ability, and systematic thinking. It is suitable for AI learners who aim for long-term development and provides a deep learning path worth exploring.