Tampa Devs

When ML models extract facts, crucial metadata like confidence, provenance, and temporal validity are often discarded, forcing downstream systems to treat all AI assertions as equally true.

This hands-on workshop introduces jsonld-ex, an open-source Python library that extends the W3C JSON-LD standard with assertion-level metadata. Through three interactive Google Colab notebooks, we will explore two formal algebras built on this framework:

• The Confidence Algebra: Grounded in Jøsang's Subjective Logic, this replaces lossy scalar scores with rich opinion tuples. Learn principled methods for multi-source fusion, trust discounting, conflict detection, and temporal decay.
• The Compliance Algebra: Model regulatory obligations (like GDPR) as dynamic epistemic states. We will cover operators for jurisdictional composition, derivation chains, consent lifecycles, and erasure verification across data lineage graphs.
• Applied Practice: Build an end-to-end, confidence-aware knowledge extraction pipeline with round-trip interoperability to W3C standards.

Note: This will be a two part series

Popular explainability methods like SHAP, LIME, and Integrated Gradients often yield conflicting results for the same model. How do you decide which one to trust? This hands-on workshop moves beyond "visual intuition" to introduce quantitative evaluation of XAI, providing objective tools to measure explanation quality.
Working through three interactive Google Colab notebooks, participants will:

• Identify Disagreement: Generate multiple explanations on real datasets to see exactly where and why they diverge.
• Apply Metric Frameworks: Measure explanations across three critical dimensions: Faithfulness (accuracy to the model), Robustness (stability), and Complexity (interpretability).
• Build Evaluation Pipelines: Create comparison tables and learn a decision framework for choosing explainers based on use cases like regulatory audits or stakeholder communication.