Machine Learning and AI with Simple Python and Matlab Scripts. Courseware for Non-computing Majors (M. Ümit Uyar)（Z-Library）

# Machine Learning and AI with Simple Python and Matlab Scripts: Courseware for Non-computing Majors ## 【One-Line Pitch】 A hands-on, math-light introduction to machine learning and AI for engineering and science students, built around simple Python and Matlab scripts that tackle real-world problems from stock trading to Alzheimer's prognosis. If you've been intimidated by AI's math prerequisites, this courseware-style book gets you coding neural networks from day one. ## 【Book Arc】 - **Opening (~0%–9%)**: Foundations of artificial neural networks (ANNs) — what neurons, weights, and layers are, how forward propagation computes outputs, and why training means minimizing error/loss by adjusting weights. Includes a fully worked numerical example showing the math step-by-step. - **Early (~9%–25%)**: Training mechanics and first applications — back propagation, learning rates, bias inputs, and weight update rules. Two complete case studies (sleep-study grade prediction and London bike rental demand) with full Python and Matlab scripts you can run and modify. - **Early (~25%–34%)**: Financial applications — moving averages (SMA/EMA), momentum, and Bollinger Bands as ANN inputs for stock trading predictions. Shows how to select inputs and interpret non-deterministic results across runs. - **Middle (~34%–47%)**: Advanced ANN architectures and validation — a four-hidden-layer network (120-60-15-5-1) for Alzheimer's disease prognosis using gene expression data, with leave-one-out cross-validation (LOOCV) to prevent over-fitting. Introduces natural language processing with keyword-frequency inputs for market prediction from tweets. - **Late (~47%–100%)**: The book continues into convolutional neural networks (speech and OCR), recurrent neural networks (chatbots, translators), and advanced optimization techniques (particle swarm, differential and grammatical evolution), plus current AI tools. Excerpts do not cover these later chapters in detail. ## 【Key Takeaways】 - **ANNs are just weight-adjustment machines** (Early): The entire training process reduces to minimizing error by updating weights via back propagation — no advanced math required beyond basic derivatives. A worked example shows error dropping from 5,974 to 5,273 in one iteration. - **Bias is non-negotiable for real problems** (Early): Every layer needs a bias input, randomly initialized and trained alongside weights, to make predictions accurate for real-world data. Skipping bias leads to poor convergence. - **Input selection matters more than architecture** (Early): Choosing the right inputs (e.g., study hours and sleep hours for grade prediction; wind speed and temperature for bike rental) is the single most important design decision. Outputs can be direct measurements or derived values like growth rates. - **Learning rate controls training stability** (Early): The learning rate (α) determines how much weights change per iteration — too large causes oscillation, too small makes training painfully slow. The book shows the exact update equations. - **Cross-validation prevents over-fitting** (Middle): Leave-one-out cross-validation trains on N−1 samples and tests on the held-out one, repeated N times. This catches ANNs that memorize noise instead of learning patterns — a common failure with too many hidden layers. - **Real applications need domain expertise** (Middle): The Alzheimer's example shows architecture selection (120-60-15-5-1) came from experimentation, and financial trading requires understanding SMA periods, momentum, and Bollinger Bands — AI alone doesn't make domain knowledge obsolete. - **ANNs are non-deterministic** (Early): Running the same stock-trading script twice can give different predictions due to random weight initialization. Consistent results require careful selection of technical indicators and ANN parameters. ## 【Reading Tips】 - **Skim the math, focus on the scripts**: The book's strength is its runnable code. If equations feel heavy, jump to the Python/Matlab listings and trace how they implement the concepts — the code comments explain the logic. - **Deep-read Chapter 2–3 for fundamentals**: The worked numerical example (inputs 3 and 5, output 75) is worth understanding fully — it demystifies forward/back propagation better than any formula alone. - **Run the case studies before moving on**: The sleep-study and bike-rental examples are your training wheels. Modify inputs, change learning rates, and observe how predictions shift — this builds intuition no amount of reading can provide. - **Treat exercises as mini-projects**: Each chapter ends with exercises asking you to repeat experiments across Python and Matlab versions. Doing even a few builds cross-language confidence. - **Skip ahead if you're here for CNNs/RNNs**: The early chapters are ANN-heavy. If your interest is chatbots or speech recognition, the later chapters cover those, but you'll need the ANN foundation first. ## 【Coverage Limits】 This guide covers the ANN fundamentals and early applications (Chapters 2–6) in detail. The later chapters on CNNs, RNNs, and evolutionary algorithms are mentioned but not analyzed — excerpts do not include their content. ##