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.

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.

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).

• 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.)

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.

• 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

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.