This diagram illustrates the progression from human judgment to rules-based systems and finally to machine learning (ML) as tasks become increasingly complex.

• Human Judgment: At the core, humans rely on experience and intuition to make decisions, which works well for simple, low-complexity tasks.
• Rules-Based Systems: As complexity grows, predefined rules encoded in software handle structured data. These systems are limited by their inability to adapt to new, unforeseen situations.
• Machine Learning: For tasks with high data complexity (e.g., large datasets) and rules complexity (e.g., intricate relationships in data), ML is the optimal solution. ML adapts and learns patterns directly from data, surpassing the limitations of static rules. 

This hierarchical view highlights ML’s unique capability to handle the nuanced challenges often encountered in fields like gastroenterology, where vast, complex data can inform diagnoses and decision-making.

2. Data: The Foundation of ML

Types of Data

• Structured Data: Organized data in tables (e.g., patient demographics, lab results).
• Unstructured Data: Data like images, videos, and text (e.g., endoscopy images).
• Semi-structured Data: A mix of structured and unstructured data, such as JSON files.

Role of Data 

• Data serves as the machine’s “experience.”
• Quality and volume of data significantly impact the accuracy of ML models.

As we progress more in our modules, we will discuss the data setup process. As the subtitle suggests, its ‘the foundation of ML’, so adequate care is needed at this stage. If this process is not done right, it may lead to biases in the algorithm. More about that in our future modules.

3. Algorithms: The Heart of ML

Types of ML Algorithms

• Supervised Learning: The machine learns from labeled data where outcomes are known.
  • Example: Predicting patient outcomes based on medical history.
• Unsupervised Learning: The machine identifies patterns in data without pre-existing labels.
  • Example: Clustering patients based on symptom similarity.
• Reinforcement Learning: The machine learns by trial and error to maximize rewards.
  • Example: Navigating surgical tools in simulations.

Common Algorithms

• Linear Regression: Identifies linear relationships between variables.
• Decision Trees: Splits data into branches based on decision rules.
• Neural Networks: Mimics the human brain’s neurons to identify complex patterns.

4. Training and Testing: The Learning Process

Steps of Model Development

1. Training:
   • The model learns from a training dataset.
   • Adjusts its parameters (weights and biases) to minimize errors.
2. Validation:
   • Ensures the model generalizes well to new data by using a validation dataset.
3. Testing:
   • Evaluates the model’s accuracy and performance on unseen data.

Metrics for Evaluation

• Accuracy, precision, recall, and mean squared error (MSE) are commonly used metrics to assess the model’s effectiveness.

You can read about the importance of different metrics to a specific question in this paper.

5. Key ML Concepts

Feedback Loops

• ML models operate on a cycle: Predict → Measure → Adjust.
• Algorithms fine-tune predictions by analyzing errors and making corrections.

Gradient Descent

• A mathematical technique used to optimize the model by finding the best set of parameters.
• It adjusts weights based on the gradient of the loss function, ensuring the model learns effectively.

Activation Functions

• Functions like ReLU (Rectified Linear Unit) introduce non-linearity, enabling models to handle complex data relationships.