Interpretable Machine Learning & White-Box Models
Autumn 2025
Why Interpretability matters?
Why Interpretability matters?
- Decision-making in high-stakes domains (healthcare, law, finance)
Why Interpretability matters?
- Decision-making in high-stakes domains (healthcare, law, finance)
- Trust, accountability, and ethical AI
Why Interpretability matters?
- Decision-making in high-stakes domains (healthcare, law, finance)
- Trust, accountability, and ethical AI
- Legal compliance (e.g., GDPR: “right to explanation”)
Why Interpretability matters?
- Decision-making in high-stakes domains (healthcare, law, finance)
- Trust, accountability, and ethical AI
- Legal compliance (e.g., GDPR: “right to explanation”)
- Quote or statistic about lack of trust in AI
What is Interpretability?
What is Interpretability?
- Definition: The degree to which a human can consistently predict the model’s output from its inputs.
What is Interpretability?
- Definition: The degree to which a human can consistently predict the model’s output from its inputs.
- Intuition: “Can I understand why this decision was made?”
What is Interpretability?
- Definition: The degree to which a human can consistently predict the model’s output from its inputs.
- Intuition: “Can I understand why this decision was made?”
- Example: Feature importance in linear models.
Interpretability
- Refers to how understandable a model is by itself.
Interpretability
- Refers to how understandable a model is by itself.
- It’s about the transparency of the model.
Interpretability
- Refers to how understandable a model is by itself.
- It’s about the transparency of the model.
- You can look at the model and directly understand how it makes decisions.
Interpretability
- Refers to how understandable a model is by itself.
- It’s about the transparency of the model.
- You can look at the model and directly understand how it makes decisions.
- Typically applies to simple models like: Linear regression, Decision Trees, etc.
Accuracy - Interpretability Tradeoff
Glass boxes
Interpretable vs Explainable
Interpretable vs Explainable
- Explainable: Refers to the ability to explain the behavior of a (possibly complex or opaque) model to humans.
Interpretable vs Explainable
- Explainable: Refers to the ability to explain the behavior of a (possibly complex or opaque) model to humans.
What Is a White-Box Model?
What Is a White-Box Model?
- Definition: A model whose internal workings are fully visible and understandable.
What Is a White-Box Model?
- Definition: A model whose internal workings are fully visible and understandable.
- Intuition: is a type of machine learning model whose internal logic, structure, and decision-making process are fully transparent and understandable to humans.
White-Box Model - characteristics
- You can see exactly how inputs are transformed into outputs.
White-Box Model - characteristics
- You can see exactly how inputs are transformed into outputs.
- They are typically interpretable by design.
White-Box Model - characteristics
- You can see exactly how inputs are transformed into outputs.
- They are typically interpretable by design.
- Easier to debug, audit, and trust.
When to Use White-Box Models?
When to Use White-Box Models?
- When interpretability is critical (transparent process)
When to Use White-Box Models?
- When interpretability is critical (transparent process)
- Small or medium-sized datasets (easy to analyze)
When to Use White-Box Models?
- When interpretability is critical (transparent process)
- Small or medium-sized datasets (easy to analyze)
- Proof-of-Concept test (understands the problem)
White-Box vs Black-Box Models
White-Box vs Black-Box Models
- Blax-Box Models: is a type of machine learning model whose internal workings are not easily understandable by humans, even if you have access to the code or parameters.
White-Box vs Black-Box Models
Summary
- Interpretable ML is essential for transparency and trust.
- White-box models are inherently interpretable and useful in critical settings.
