NVIDIA · NCA-GENM

The trust and safety thread

What a multimodal generative system owes its users. Opens in M2 with confounding factors hiding in an aggregate trend, and concentrates in M7's ethical principles, bias mitigation, data privacy vs. consent, and the multimodal-specific concern of content authenticity and deepfake detection.

NCAM-T4 · 11 lessons across 2 modules

  1. M2M2-01Data cleaning: missing values, outliers, scaling, and categorical encoding
  2. M2M2-02Exploratory data analysis: descriptive statistics and correlation
  3. M2M2-03Choosing the right chart and avoiding misleading visuals
  4. M2M2-04Attention maps as an explainability and debugging tool in multimodal settings
  5. M2M2-05Preparing multimodal data: augmentation and OCR for PDF extraction
  6. M2M2-06Identifying relationships, trends, and confounding factors in an analysis
  7. M7M7-01Ethical principles of trustworthy AI: privacy, safety, transparency, nondiscrimination
  8. M7M7-02Minimizing bias: disaggregated evaluation and mitigation
  9. M7M7-03Data privacy vs. data consent
  10. M7M7-04Content authenticity for multimodal generative AI: provenance, watermarking, disclosure, detection
  11. M7M7-05Hallucination, grounding, guardrails, and a trustworthy-AI checklist

Part of the throughlines running across the NCA-GENM prep course.