Revise — Compressed Notes¶
Compressed revision notes. Every lecture reduced to its formulas, tables, and flashcards. Not for first-time learners — use the Learn guides for that.
Prerequisite
These notes assume you already understand the concept. If a flashcard here confuses you, that concept hasn't stuck yet — go back to the corresponding Learn guide.
When to use these¶
- Day 8 of the study plan — after you've read the learn guides
- Between mock exams — quick refresh of a weak area
- Right before a specific practice question — narrow the topic first
The 12 revision decks¶
| # | Notes | Length | Focus |
|---|---|---|---|
| 1 | ML foundations | ~200 lines | Precision, recall, F1, confusion matrix |
| 2 | Neural networks | ~250 lines | MLP → CNN → RNN → LSTM → embeddings |
| 3 | Transformers | ~250 lines | Q/K/V, positional encoding |
| 4 | LLM decoding & APIs | ~300 lines | Decoding strategies, OpenAI API roles |
| 5 | Structured outputs & evaluation | ~250 lines | NLP metrics decision tree |
| 6 | AI safety | ~200 lines | CIA triad, attack taxonomy |
| 7 | Advanced prompting | ~250 lines | Prompting technique flashcards |
| 8 | Prompt security & APE | ~200 lines | Levenshtein, log-prob scoring |
| 9 | RAG fundamentals | ~200 lines | Retrieval mechanism, naive vs agentic |
| 10 | Dense retrieval | ~250 lines | Bi vs cross encoder, ColBERT |
| 11 | Vector indexing | ~300 lines | PQ formulas, index selection tree |
| 12 | Agentic RAG | ~250 lines | Observe-Think-Act, tool routing |
Format of each note deck¶
Every deck has these sections:
- TL;DR — 5 bullets summarising the whole lecture
- Exam-relevant concepts — the specific concepts that will be tested, each with a Definition / Formula / When-it-matters / Common-confusion block
- Likely MCQ angles — the top 10 question patterns the concept produces
- One-line flashcards — Q/A pairs for spaced repetition