Cedar Works
Cedar Works Explainable AI Lectures
Abstract visualization of neural network decision pathways
Development Roadmap

Learning path through explainable AI

Three structured phases move learners from interpretability theory to applied model auditing — each phase builds directly on the previous, with no assumed background in machine learning research.

Phase structure

Each phase has a defined scope, a measurable outcome, and a set of practical exercises. Progress is sequential — the curriculum does not branch into optional tracks.

1
Phase One

Foundations of Interpretability

Most practitioners encounter XAI after already working with predictive models. This phase starts from that position — examining what a model actually learns versus what its outputs suggest, and where those two things diverge.

Concepts covered include the distinction between interpretable-by-design models and post-hoc explanation methods, the role of feature attribution, and the conditions under which a model explanation can be trusted.

Linear models & decision trees Attribution fundamentals Trust calibration
Active
Duration 4 weeks
Format Video + reading
Exercises 6 structured
2
Phase Two

Methods and Tooling in Practice

SHAP and LIME are the two most commonly cited XAI tools, but they answer different questions and fail in different ways. This phase works through both in depth, alongside attention visualization for transformer-based models and counterfactual generation.

Learners apply each method to a real dataset — tabular credit data, image classification outputs, and a text classification pipeline — comparing what each explanation reveals and where it misleads.

SHAP values LIME approximation Attention maps Counterfactuals Faithfulness testing
3
Phase Three

Auditing Models in Regulated Contexts

Explanation methods only matter when they connect to a decision. This phase focuses on applying XAI in settings where model outputs carry legal or operational weight — loan approvals, clinical triage support, and hiring screening.

Learners produce explanation artifacts designed for non-technical reviewers: structured audit reports, visual summaries of feature influence, and documented edge cases where the model behaves unexpectedly.

Regulatory alignment Audit documentation Stakeholder reporting Edge case analysis

What the roadmap covers in numbers

These figures reflect the current curriculum scope. Cedar Works has been refining this structure since 2015, adjusting phase boundaries as the XAI field itself has matured.

12
Weeks total Across all three phases
18
Practical exercises Applied to real datasets
4
Core XAI methods SHAP, LIME, attention, counterfactuals