{"id":49153269,"url":"https://github.com/klauseduard/deep-learning-knowledge-base-demo","last_synced_at":"2026-04-22T07:07:04.638Z","repository":{"id":351943236,"uuid":"1213169458","full_name":"klauseduard/deep-learning-knowledge-base-demo","owner":"klauseduard","description":"Demo: structured deep learning knowledge base with interactive teaching materials — built with Claude Code","archived":false,"fork":false,"pushed_at":"2026-04-17T07:00:28.000Z","size":293,"stargazers_count":0,"open_issues_count":0,"forks_count":0,"subscribers_count":0,"default_branch":"master","last_synced_at":"2026-04-17T07:31:52.576Z","etag":null,"topics":["backpropagation","deep-learning","interactive-learning","knowledge-base","machine-learning","neural-networks","obsidian","teaching-materials"],"latest_commit_sha":null,"homepage":null,"language":"HTML","has_issues":true,"has_wiki":null,"has_pages":null,"mirror_url":null,"source_name":null,"license":null,"status":null,"scm":"git","pull_requests_enabled":true,"icon_url":"https://github.com/klauseduard.png","metadata":{"files":{"readme":"README.md","changelog":null,"contributing":null,"funding":null,"license":null,"code_of_conduct":null,"threat_model":null,"audit":null,"citation":null,"codeowners":null,"security":null,"support":null,"governance":null,"roadmap":null,"authors":null,"dei":null,"publiccode":null,"codemeta":null,"zenodo":null,"notice":null,"maintainers":null,"copyright":null,"agents":null,"dco":null,"cla":null}},"created_at":"2026-04-17T05:37:16.000Z","updated_at":"2026-04-17T07:00:32.000Z","dependencies_parsed_at":null,"dependency_job_id":null,"html_url":"https://github.com/klauseduard/deep-learning-knowledge-base-demo","commit_stats":null,"previous_names":["klauseduard/deep-learning-knowledge-base-demo"],"tags_count":null,"template":false,"template_full_name":null,"purl":"pkg:github/klauseduard/deep-learning-knowledge-base-demo","repository_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/klauseduard%2Fdeep-learning-knowledge-base-demo","tags_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/klauseduard%2Fdeep-learning-knowledge-base-demo/tags","releases_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/klauseduard%2Fdeep-learning-knowledge-base-demo/releases","manifests_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/klauseduard%2Fdeep-learning-knowledge-base-demo/manifests","owner_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners/klauseduard","download_url":"https://codeload.github.com/klauseduard/deep-learning-knowledge-base-demo/tar.gz/refs/heads/master","sbom_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/klauseduard%2Fdeep-learning-knowledge-base-demo/sbom","scorecard":null,"host":{"name":"GitHub","url":"https://github.com","kind":"github","repositories_count":286080680,"owners_count":32125101,"icon_url":"https://github.com/github.png","version":null,"created_at":"2022-05-30T11:31:42.601Z","updated_at":"2026-04-22T00:31:26.853Z","status":"online","status_checked_at":"2026-04-22T02:00:05.693Z","response_time":58,"last_error":null,"robots_txt_status":"success","robots_txt_updated_at":"2025-07-24T06:49:26.215Z","robots_txt_url":"https://github.com/robots.txt","online":true,"can_crawl_api":true,"host_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub","repositories_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories","repository_names_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repository_names","owners_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners"}},"keywords":["backpropagation","deep-learning","interactive-learning","knowledge-base","machine-learning","neural-networks","obsidian","teaching-materials"],"created_at":"2026-04-22T07:06:45.296Z","updated_at":"2026-04-22T07:07:04.624Z","avatar_url":"https://github.com/klauseduard.png","language":"HTML","funding_links":[],"categories":[],"sub_categories":[],"readme":"# Deep Learning Knowledge Base — Demo\n\nA demonstration of building a structured knowledge base and interactive teaching materials using [Claude Code](https://claude.ai/code), with deep learning as the subject matter.\n\n## What this is\n\nThis repository demonstrates a workflow for learning a complex technical domain with an LLM as a collaborator. The knowledge base started as AI-generated drafts seeded from François Fleuret's *The Little Book of Deep Learning* (2023), then improved through iterative review — the learner reads the articles, asks questions about what's unclear or missing, and the material gets refined or expanded in response. Interactive teaching materials emerged from this process: each one was prompted by a genuine question (\"I forgot the math — show me with real numbers\", \"what happens when I stack 20 sigmoid layers?\") rather than planned upfront.\n\nMost content has been removed from this public repository. A knowledge base built this way is inherently personal — shaped by one learner's questions, gaps, and interests — and publishing it as a reference would misrepresent its purpose. Additionally, the LLM can generate draft articles far faster than a person can engage with them, so much of the initial material hasn't been reviewed or refined yet. What remains here is enough to demonstrate the *approach*: the article format, the teaching materials, and how they connect.\n\n```mermaid\nflowchart LR\n    subgraph seed [\"1. Seed\"]\n        S[Source material\u003cbr\u003e\u003ci\u003ebook, paper, topic\u003c/i\u003e]\n    end\n\n    subgraph generate [\"2. Generate\"]\n        G[LLM drafts\u003cbr\u003econcept articles]\n    end\n\n    subgraph learn [\"3. Learn\"]\n        L[Learner reads\u003cbr\u003eand questions]\n    end\n\n    subgraph refine [\"4. Refine\"]\n        direction TB\n        R1[Clarify jargon\u003cbr\u003eadd links]\n        R2[Add worked\u003cbr\u003eexamples]\n        R3[Generate teaching\u003cbr\u003ematerial]\n        R4[Create reference\u003cbr\u003eentries]\n    end\n\n    S --\u003e G --\u003e L\n    L -- \"unclear term\" --\u003e R1\n    L -- \"forgot the math\" --\u003e R2\n    L -- \"show me interactively\" --\u003e R3\n    L -- \"what does this symbol mean?\" --\u003e R4\n    R1 \u0026 R2 \u0026 R3 \u0026 R4 -- improved article --\u003e L\n\n    style seed fill:#f5f3ef,stroke:#e2ddd5\n    style generate fill:#e8f0f8,stroke:#3d6b9e\n    style learn fill:#fef6e8,stroke:#b87a1a\n    style refine fill:#e8f5ec,stroke:#27864a\n```\n\nThe result combines:\n\n- **Concept articles** — structured markdown files covering deep learning topics, interlinked with wiki-style `[[references]]`\n- **Interactive teaching materials** — standalone HTML pages with step-by-step math walkthroughs, interactive simulations, and explorable visualizations\n- **Reference materials** — glossary, notation guide, timeline, and key people for quick lookup\n\n## Demo scope\n\n**5 fully developed concept articles** demonstrate the article format and editorial approach:\n\n| Article | Description |\n|---------|-------------|\n| [backpropagation](concepts/backpropagation.md) | The algorithm that makes training feasible — chain rule, BPTT, adjoint method, memory requirements |\n| [gradient-descent](concepts/gradient-descent.md) | SGD, learning rates, momentum, Adam, mini-batch trade-offs |\n| [loss-functions](concepts/loss-functions.md) | MSE, cross-entropy (with softmax explained), contrastive loss, autoregressive loss |\n| [activation-functions](concepts/activation-functions.md) | ReLU through GELU, desirable properties, initialization interaction |\n| [multilayer-perceptrons](concepts/multilayer-perceptrons.md) | The simplest deep architecture, universal approximation theorem |\n\nThe remaining ~90 concept articles are **stubs** — they retain their title, tagline, and connections section to preserve the linking structure, but their full content is not included in this demo.\n\n## Interactive teaching materials\n\nEach material is a self-contained HTML file — no build step, no server. Try them live via GitHub Pages or clone the repo and open locally.\n\n### Backpropagation\n- **[The Math of Backpropagation](https://klauseduard.github.io/deep-learning-knowledge-base-demo/teach/backpropagation/index.html)** — partial derivatives refresher, chain rule with numbers, full forward/backward pass by hand, how autograd works. Includes synced network diagrams that highlight the active computation at each step.\n- **[Vanishing \u0026 Exploding Gradients](https://klauseduard.github.io/deep-learning-knowledge-base-demo/teach/backpropagation/vanishing-gradients.html)** — simulation with controls for depth (2-50 layers), activation function, initialization scheme, skip connections, and layer normalization. Watch gradient magnitudes respond in real time.\n\n### Gradient Descent\n- **[SGD by the Numbers](https://klauseduard.github.io/deep-learning-knowledge-base-demo/teach/gradient-descent/sgd-math.html)** — worked mini-batch example: per-example gradients, averaging, parameter update, plus a multi-epoch training runner with adjustable learning rate and batch size.\n- **[Gradient Descent — Interactive Learning](https://klauseduard.github.io/deep-learning-knowledge-base-demo/teach/gradient-descent/index.html)** — 3D loss landscape exploration, learning rate comparison, momentum visualization, optimizer race (SGD vs Adam).\n\n### Loss Functions\n- **[Cross-Entropy \u0026 Softmax — Worked Math](https://klauseduard.github.io/deep-learning-knowledge-base-demo/teach/loss-functions/cross-entropy-math.html)** — step-by-step: softmax computation, cross-entropy loss, gradients, parameter update, autoregressive cross-entropy with perplexity, and contrastive loss (CLIP-style).\n- **[Cross-Entropy \u0026 Softmax Explorer](https://klauseduard.github.io/deep-learning-knowledge-base-demo/teach/loss-functions/index.html)** — adjust logits with sliders, watch softmax probabilities and the -log(p) loss curve respond. Temperature control, gradient visualization, presets for edge cases.\n\n### Activation Functions\n- **[Activation Function Explorer](https://klauseduard.github.io/deep-learning-knowledge-base-demo/teach/activation-functions/index.html)** — drag along curves to probe function values and derivatives, compare any two functions side by side, stack 1-30 layers to see how repeated activation crushes or preserves signals.\n\n### Tensors\n- **[Tensor Fundamentals](https://klauseduard.github.io/deep-learning-knowledge-base-demo/teach/tensors/index.html)** — interactive grid visualization of tensor shapes, indexing, reshaping, and broadcasting.\n\n## Reference materials\n\n| File | Contents |\n|------|----------|\n| [reference/notation.md](reference/notation.md) | Mathematical symbols and conventions ($\\hat{y}$ = predicted, $\\nabla$ = gradient, etc.) |\n| [reference/glossary.md](reference/glossary.md) | 55+ term definitions (logits, softmax etymology, cross-entropy, epoch, backbone, etc.) |\n| [reference/timeline.md](reference/timeline.md) | 42 milestones from McCulloch-Pitts (1943) to frontier models (2025) |\n| [reference/key-people.md](reference/key-people.md) | ~40 researchers organized by contribution area |\n\n## Structure\n\n```\nconcepts/           95 topic articles (5 full, 90 stubs in this demo)\nteach/              Interactive HTML teaching materials\n  backpropagation/    3 pages (math walkthrough, vanishing gradients)\n  gradient-descent/   2 pages (SGD math, interactive landscapes)\n  loss-functions/     2 pages (cross-entropy math, interactive explorer)\n  activation-functions/ 1 page (function explorer)\n  tensors/            1 page (tensor fundamentals)\ntools/              5 framework/library articles (stubs)\nreference/          4 reference documents (notation, glossary, timeline, people)\n_index.md           Master index of all articles\n_connections.md     Cross-cutting themes and relationships\n_sources.md         Bibliography\n_open_questions.md  Unresolved questions across the field\n```\n\n## Design principles\n\n**Concept articles** follow a consistent structure: tagline → overview → key details (intuition, etymology, technical subsections) → learn (links to teaching materials) → connections → sources → open questions. Inline `[[wiki-links]]` connect articles where concepts are naturally referenced.\n\n**Teaching materials** use vanilla HTML/CSS/JS with CDN-hosted libraries (KaTeX for math, Plotly for charts). They follow a consistent design system (Source Serif for headings, Inter for body, JetBrains Mono for code/values). Two complementary formats:\n- *Math walkthroughs* — step-through panels with concrete numbers at every step, synced diagrams\n- *Interactive explorers* — parameter controls with immediate visual feedback, \"what to notice\" hints\n\n**Reference materials** are look-up resources, not study materials. They support the concept articles without duplicating them.\n\n## How to use\n\n### Concept articles (Markdown)\n\nThe concept articles use `[[wiki-links]]` for cross-referencing and embedded Plotly charts for visualizations. To get the full experience:\n\n1. **[Obsidian](https://obsidian.md/)** (free) — open this repository as an Obsidian vault. Wiki-links become clickable, the graph view shows connections between articles, and LaTeX math renders inline.\n2. **[Obsidian Plotly plugin](https://github.com/Dmytro-Shulha/obsidian-plotly)** — install via Obsidian's community plugins to render the embedded charts in the loss-functions and activation-functions articles.\n\nWithout Obsidian, the articles are still readable as plain Markdown — `[[wiki-links]]` won't be clickable but the text is clear. The Plotly chart data is visible as YAML blocks.\n\n### Teaching materials (HTML)\n\nThe `teach/` directory contains standalone HTML files. Just open them directly in a browser (Firefox, Chrome, etc.) — no server, no build step, no installation. They load fonts and math rendering from CDNs, so an internet connection is needed on first open.\n\n### Reference materials (Markdown)\n\nPlain Markdown with some LaTeX math notation. Readable in any Markdown viewer or text editor; best in Obsidian where the math renders.\n\n## Built with\n\nThe knowledge base was built iteratively using [Claude Code](https://claude.ai/code) — article drafting, review, teaching material generation, and cross-linking were done in conversation.\n\nMuch of the repetitive structure (article templates, teaching material scaffolding, linting, visualization helpers) was encapsulated in a set of personal [Claude Code skills](https://docs.claude.com/en/docs/claude-code/skills) — for example `kb-seed` (generate article drafts from a topic list), `kb-teach` (scaffold interactive teaching materials), `kb-lint` (check article structure), and `kb-viz` (generate diagrams). These skills are not included in this repository — they're personal tooling tailored to one learner's workflow, and they'd need to be rewritten to match someone else's article conventions and design preferences. But the concept of per-domain skills is worth knowing about: a small set of invokable prompts (like `/kb-teach \u003cconcept\u003e`) reduces the friction of consistent knowledge base work significantly.\n\n## License\n\nContent is provided for demonstration purposes. The teaching material HTML files can be opened directly in any modern browser.\n","project_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fklauseduard%2Fdeep-learning-knowledge-base-demo","html_url":"https://awesome.ecosyste.ms/projects/github.com%2Fklauseduard%2Fdeep-learning-knowledge-base-demo","lists_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fklauseduard%2Fdeep-learning-knowledge-base-demo/lists"}