Properties

Properties are characteristics of the trained model, the procedure used to train it (including training data), or its ability to perform inference.

MLTE Properties

• Functionality
  • Task Efficacy (e.g., Prediction Accuracy, Error Rate, etc)
  • Fairness
  • Interpretability
• Robustness
  • Robustness to Naturally Occurring Data Challenges
  • Robustness to Adversarial Attack
  • Robustness to Device-Generated Perturbations
  • Security
• Costs
  • Model Size
  • Training Costs
  • Inference Performance

To read more about these properties and their implementation, see the Resources page.