Skip to content

Choose which data to make your model forget (Unlearn!), but watch out - every deletion ripples!

License

Notifications You must be signed in to change notification settings

AmanPriyanshu/The-Unlearning-Protocol

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

20 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

🧠 The Unlearning Protocol Game

demo-unlearning

An interactive game that demonstrates machine unlearning through a neural network trained on demographic data. Experience firsthand how making models forget affects their behavior across different population groups!

🎯 Educational Goals

This game helps players understand:

  • The concept and challenges of machine unlearning
  • How selective forgetting impacts model fairness
  • Ripple effects across demographic groups
  • The delicate balance between forgetting and maintaining performance

🎮 How It Works

1. The Forgetting Process

  • Select individual data points for the model to forget
  • Configure unlearning parameters (learning rate and epochs)
  • Watch how forgetting ripples through the model's behavior
  • Monitor performance changes across different demographics

2. Impact Visualization

  • Real-time Performance Tracking: See how unlearning affects model accuracy
  • Demographic Impact: Monitor changes across age, education, and work hours
  • Comparative Analysis: Compare unlearning vs retraining results
  • Global Statistics: Track overall model health

3. Strategic Elements

  • Choose which samples to forget wisely - not all forgetting is equal!
  • Balance aggressive vs gentle unlearning through parameter tuning
  • Monitor unintended consequences across different population groups
  • Aim for minimal collateral damage while achieving targeted forgetting

🎲 How to Play

  1. Sample Selection

    • Review candidate samples for unlearning
    • Each sample shows key demographic information
    • Consider potential ripple effects before choosing
  2. Configure Unlearning

    • Adjust the learning rate (0.001 to 0.1)
    • Set the number of unlearning epochs (1 to 50)
    • Higher values = more aggressive forgetting
  3. Monitor Impact

    • Watch performance changes across groups
    • Compare with reference retraining results
    • Look for unexpected demographic impacts

🎯 Challenge Goals

  1. Successfully make the model forget targeted samples
  2. Maintain balanced performance across demographics
  3. Minimize accuracy drop on unrelated groups
  4. Find optimal unlearning parameters for different scenarios