English
97 tutorials · 10 blog posts
850,368 English words
This is a bilingual (English / Traditional Chinese) tutorial site covering digital image processing, natural language processing, pattern recognition, machine learning and deep learning, computer vision, optimization and information theory, from the mathematical foundations to recent research.
Every tutorial starts with a plain-language explanation, then gives the rigorous math, runnable Python code, figures, a "Modern view" built from survey papers, exercises, and verified references. Some tutorials also include interactive demos that run in your browser, or a Jupyter notebook you can open in Google Colab.
Start with the getting-started guide, or pick a track on the Learn page.
A learning map from foundations to advanced topics. These totals cover published tutorials and blog posts and update with the site’s content.
Tutorial topics count each topic once across languages. Page totals count each language edition separately and exclude landing pages, indexes, and drafts.
97 tutorials · 10 blog posts
850,368 English words
97 tutorials · 10 blog posts
1,136,614 Chinese characters
Plus 140,691 non-Chinese words and numbers
Approximate prose counts include headings, exercises, and reference text. Code, equations, metadata, link destinations, and component attributes are excluded. English words and Chinese characters are different units and are not added together.
| Track | EN | 繁中 |
|---|---|---|
| Foundations | 2 | 2 |
| Image Processing | 13 | 13 |
| Natural Language Processing | 12 | 12 |
| Pattern Recognition | 10 | 10 |
| Machine Learning | 15 | 15 |
| Deep Learning for Computer Vision | 29 | 29 |
| Optimization | 8 | 8 |
| Information Theory | 8 | 8 |
| Total | 97 | 97 |
Beginner: 16 · Intermediate: 26 · Advanced: 55
This site is created and maintained by Shyang-En Weng (翁祥恩), a PhD candidate in Computer Science at National Yang Ming Chiao Tung University (NYCU), Taiwan, working on image processing, pattern recognition and deep learning, with a focus on image restoration, low-light enhancement and object detection. More at shyandram.dev.
Most of the content and code on this site was constructed by AI models: Anthropic's Claude models (through Claude Code, mainly Claude Opus 5.5) and OpenAI's GPT models (through Codex, mainly GPT-6.1 Sol and GPT-6 Astra). The author chose the topics and course structure, supplied the original articles and research material, and reviewed the results.
This site is still under review. References were checked link by link and code was run before publishing, but the tutorials still need further checking and revision by the author, and they may contain mistakes or omissions. Please read them alongside the original sources; if you find a problem, report it on GitHub.
A planning estimate for developing the course and website, bilingual revision, code verification, and ongoing upkeep at public API rates. These are usage scenarios, not paid invoices: complete historical token usage is unavailable.
| Item | USD | Approx. TWD |
|---|---|---|
| One-time AI/API development | US$ 425–1,275 | NT$ 13,600–40,800 |
| Annual updates and operation | US$ 127–635 | NT$ 4,064–20,320 |
| Development + first year | US$ 552–1,910 | NT$ 17,664–61,120 |
Prices checked: 2026-11-05. Default conversion: 1 USD = 32 TWD, a planning assumption rather than a live exchange rate. The range represents usage scenarios, not a statistical confidence interval.
Development allows 50–150 million input tokens and 10–30 million billed output tokens, covering drafting, revision, retries, and reasoning usage. Annual maintenance assumes another 1–5 million input and 0.2–1 million output tokens per month. These are planning allowances, not measurements inferred from published word counts.
| Model | Share | Input / output |
|---|---|---|
| GPT-6.1 Sol | 60% | 2 / 10 |
| Claude Opus 5.5 | 30% | 4 / 20 |
| GPT-6 Astra | 10% | 10 / 50 |
Weighted rates are US$3.40 per million input tokens and US$17.00 per million output tokens. Development = token cost × 1.25. Annual operation = (12 months of update tokens + a US$20–100 allowance for shared domain costs and extra tools) × 1.25. The first-year total adds development once; 25% is contingency. The multiplier changes token allowances only.
The site serves static pages; reading does not call a model API. Within free-plan limits, Cloudflare Pages static-asset requests cost US$0. The domain/tool allowance is not a vendor quote. Model costs use standard short-context text API rates without caching or batch discounts. Long context, faster service tiers, extra tool calls, and image/video generation may add costs.
Human review, equipment, and taxes are excluded. If development uses ChatGPT/Claude subscription allowances, use subscription and overage invoices instead of adding the API-equivalent estimate for the same usage again. Replace this estimate with actual billing when complete records are available.
OpenAI API pricing · Claude API pricing · Cloudflare Pages pricing · Cloudflare Pages limits