Neural Networks 101

1. Topic

Neural Networks 101 — a foundational course covering what neural networks are, how they learn, and the major modern architectures (CNNs, RNNs, Transformers).

2. Target Audience

Complete beginners. No prior coding experience. No prior linear algebra, calculus, or probability background assumed. Curious adults, students, or professionals from non-STEM backgrounds who want to genuinely understand neural networks rather than parrot buzzwords.

3. Prerequisites

Minimal. Comfort with:

• Basic high-school algebra (variables, equations, simple manipulation)
• Reading carefully and willing to do pencil-and-paper exercises
• Curiosity about how AI systems actually work

The course teaches all required math (vectors, dot products, derivatives intuition, probability basics) inline. Anything beyond high-school algebra is introduced with analogies and visual explanations before any equation appears.

4. Depth & Length

Comprehensive intro, ~30–40 hours total. Broad coverage with depth at key points: students should walk away genuinely understanding how a neural network learns — not just having seen the buzzwords. Roughly 10–12 modules.

5. Approach

Math/theory only — no code. Every concept is taught through:

• Visual & intuitive math: diagrams, animations, analogies first; equations introduced gradually with plain-English narration of every symbol.
• Pencil-and-paper exercises: worked examples on tiny networks (e.g., trace a forward pass through a 2-neuron network by hand).
• Interactive HTML labs: browser-based visualizations students can click and play with — gradient descent visualizers, neuron activation explorers, attention pattern viewers.
• No programming. No PyTorch, no NumPy, no code snippets. Math notation only where it adds clarity.

6. Required Topics

• Module 1–2: What is a neural network (perceptron, single neuron, biological inspiration vs. real story)
• Module 3–4: Feedforward networks, activation functions, the math of layers
• Module 5–6: How NNs learn — loss functions, gradient descent, backpropagation (with heavy visual scaffolding)
• Module 7: Regularization, generalization, why training is hard
• Module 8: CNNs (convolutional networks for vision)
• Module 9: RNNs and sequence models
• Module 10: Transformers and attention (conceptual)
• Module 11–12: Where NNs go wrong, ethics, what’s next, capstone

7. Format

Markdown course folder. Modules of .md lessons + .html interactive labs + per-lesson quizzes + per-module assignments + final capstone. Self-study or LMS-ready.

8. Constraints

• No code in lesson body. A lesson may reference what code would do conceptually, but never include actual code blocks.
• Math accessible from zero. Every symbol introduced needs a plain-English gloss the first time it appears.
• Hands-on without coding. Pencil-paper exercises and HTML labs are the engagement vehicle.
• Standalone. Every external link must work without an account or paywall, or be flagged as optional.

9. Success Criteria

A motivated student with no prior background should, after finishing this course, be able to:

• Explain in plain English how a neural network represents a function
• Trace, by hand, a forward pass and a backward pass through a tiny network
• Read a research paper’s architecture diagram and identify what each block does
• Reason about why a particular architecture (CNN vs. RNN vs. Transformer) fits a particular problem
• Spot common training pathologies (vanishing gradients, overfitting) from symptoms