Machine Learning Fundamentals

What you'll learn

• 1. What is Machine Learning?

  Machine learning is a subset of artificial intelligence that gives systems the ability to learn and improve from data without being explicitly programmed for every scenario. ---

• 2. Types of Machine Learning

• Supervised Learning

  Every training example has a corresponding label. The model learns a mapping from inputs → outputs. | Task Type | Output | Examples | |-----------|--------|----------| | Classification | Discrete category | Spam detection, image recognit…

• Unsupervised Learning

  No labels for training data. The model finds hidden patterns or structure in the data. | Task Type | What It Finds | Examples | |-----------|--------------|----------| | Clustering | Groups of similar data points | Customer segmentation,…

• Reinforcement Learning

  The model learns through consequences of actions in a specific environment. It receives rewards for good actions and penalties for bad ones. | Component | Role | |-----------|------| | Agent | Makes decisions (the learner) | | Environmen…

• 3. The ML Workflow

  ---

• 4. Key ML Concepts

• Bias-Variance Trade-off

  | Concept | Meaning | Symptom | |---------|---------|---------| | High Bias | Model is too simple (underfitting) | Low training accuracy, low test accuracy | | High Variance | Model is too complex (overfitting) | High training accuracy,…

• Train/Test/Validation Split

  Cross-validation (k-fold): Split data into k folds, train on k-1, validate on 1. Rotate k times. Averages out variance in evaluation.

• Evaluation Metrics

  | Metric | Formula | Use When | |--------|---------|----------| | Accuracy | Correct / Total | Balanced classes | | Precision | TP / (TP + FP) | Cost of false positive is high (spam filter) | | Recall | TP / (TP + FN) | Cost of missing p…

• Feature Engineering

  The process of using domain knowledge to create/select/transform input features that make ML algorithms work better. Common techniques: - One-hot encoding for categorical features - Normalization/standardization for numerical features -…

• 5. Handling Imbalanced Data

  Imbalanced data = unequal instances across classes (e.g., 10 malignant vs 90 benign samples in disease detection). Most classification algorithms are sensitive to class imbalance — a model predicting "benign" for everything would get 90%…

• Strategies

  | Strategy | How It Works | Trade-off | |----------|-------------|-----------| | Under-sampling | Randomly remove majority class samples to match minority | Simple, but loses potentially relevant information | | Over-sampling | Duplicate…

• SMOTE (Synthetic Minority Over-sampling Technique)

  Instead of simply duplicating minority samples, SMOTE creates new synthetic samples by interpolating between existing minority examples: Key insight: Synthetic examples are plausible because they're created close to existing minority exa…

• Critical Rule for Cross-Validation

  Never apply over/under-sampling before the train-test split. Apply sampling only to the training fold within each cross-validation iteration. The test fold must remain untouched (original distribution) for honest evaluation. Why this mat…

• 6. Generative AI

  Generative AI creates new content (images, text, music, code) by learning patterns from existing data. It's one of the most significant recent advances in AI.

• How It Works

  Generative AI pits two neural networks against each other to produce new and original digital works based on sample inputs.

• Generative Adversarial Networks (GANs)

  | Component | Role | |-----------|------| | Generator | Creates fake samples trying to fool the discriminator | | Discriminator | Tries to distinguish real data from generator's fakes | | Training | Both improve adversarially until gener…

• 6. Cloud ML Stack (Service Layers)

  Modern cloud platforms organize ML services into three layers: | Layer | Expertise Needed | Customization | Use Case | |-------|-----------------|---------------|----------| | AI Services | None (API call) | Low (pre-trained) | Quick int…

• 7. ML in System Design Interviews

  When ML comes up in system design: | System | ML Application | Approach | |--------|---------------|----------| | News Feed | Content ranking | Supervised (engagement prediction) | | Search Engine | Query understanding, result ranking |…

• 8. Data Versioning for ML

  ML projects need to version both code AND data (training datasets, model weights). Standard git doesn't handle large binary files well. Data Version Control (DVC): - Git extension for managing code + data together - Stores large files in…

• 9. Code Quality in ML Projects

• Code Review Checklist

  Architecture/Design: - Single Responsibility Principle (one class/method = one job) - Open/Closed Principle (extend, don't modify) - Code duplication ("three strikes" rule — refactor after 3rd copy) - Potential off-by-one errors and loop…

• Summary

  | Concept | Key Takeaway | |---------|-------------| | Supervised | Labeled data → predict outcomes (classification/regression) | | Unsupervised | No labels → discover structure (clustering/reduction) | | Reinforcement | Trial and error…