Build Saliency-Map Explanations for Dermatology Triage
Overview
What this challenge is about.
You receive a trained CNN (ResNet-50 backbone, 7-class lesion classifier) and a 1,000-image held-out test set with dermatologist labels. Implement Integrated Gradients, GradCAM, and SmoothGrad. Compute quantitative faithfulness metrics (insertion/deletion curves) and a small qualitative review where you have a non-dermatologist reviewer (you) tag each saliency map's plausibility on a 1-5 scale for 50 random cases. Recommend one method per use-case (triage explanation vs. model-debug) and write a 3-page memo for medical affairs.
The Brief
What you'll do, and what you'll demonstrate.
Recommend and justify a saliency method for a clinical dermatology classifier, balancing faithfulness and plausibility for clinician review.
Earning criteria — what you'll demonstrate
- Implement gradient-based saliency methods for a real CNN classifier
- Quantify saliency faithfulness with insertion/deletion curves
- Reason about the faithfulness/plausibility trade-off in medical settings
- Communicate XAI choices to a non-ML clinical audience
Program Fit
Where this fits in your program.
Sharpens the same skills your degree expects you to demonstrate.
Skills
Skills you'll demonstrate.
Each one shows up on your verified credential.
Careers
Roles this prepares you for.
Real titles. Real skill bridges. Pick the one closest to your trajectory.
AI Safety Researcher
Quantifying saliency faithfulness vs. plausibility on a clinical classifier and writing a clinician-facing memo is exactly the day-one work of an AI safety researcher in healthtech.
This challenge sharpens
- interpretability
- faithfulness-evaluation
- saliency-maps
ML Researcher
Comparing interpretability methods with proper faithfulness metrics is the kind of methodology work ML researchers do at applied research labs.
This challenge sharpens
- integrated-gradients
- gradcam
- interpretability
Computer Vision Engineer
Building per-image saliency pipelines for a production CNN transfers directly to CV-engineer work at any imaging-AI product team.
This challenge sharpens
- saliency-maps
- gradcam
- pytorch