Krum, the Library

A research framework for Byzantine-resilient distributed learning.

Krum provides a modular environment for comparing robust aggregation rules, simulating Byzantine attacks, running experiment campaigns, and analyzing results.

Reference

• Repository Architecture
• Aggregators
  • Average
  • Median
  • Krum / Multi-Krum
  • Bulyan
  • Brute
• Attacks
  • Identical
  • NaN
• Experiments
  • Configuration
  • Dataset
  • Model
  • Loss
  • Optimizer
  • Checkpoint
  • Models
  • Datasets
• Tools
  • PyTorch Utilities
  • Miscellaneous Utilities
  • Jobs Utilities
• Native Acceleration

Project

• Contributors

License

Krum is open-sourced under the MIT License. See the License file on GitHub for details. Source code and issue tracker live at https://github.com/calicarpa/krum.

Indices and tables

• Index

• Module Index

• Search Page