Skip to content

IBMFL 1.0.7
Compare
Choose a tag to compare
@chalianwar chalianwar released this 16 Mar 20:24
· 25 commits to main since this release
v1.0.7
688a586
This commit was created on GitHub.com and signed with GitHub’s verified signature. The key has expired.
GPG key ID: 4AEE18F83AFDEB23
Expired
Learn about vigilant mode.

Release 1.0.7

IBMFL 1.0.7 includes new functionality, improvements, and bug fixes.

Changes

  • Re-organization of examples to help end-user
  • New tutorial on how to add a new fusion algorithm
  • Improved README files
  • Internal change on how PyTorch models are specified. The change enables new optimizers, see this tutorial for more details

New Functionality

  • New fusion algorithm to train (Doc2Vec) models.
  • New robust fusion algorithm (Comparative Elimination) to train robust neural network models.
  • New connection type PubSub. It replaces RabbitMQ based connection type supported by previous releases.
  • New tutorial on how to add a new fusion algorithm to IBM FL
  • New Jupyter notebook to illustrate quorum and rejoin support
  • New Jupyter notebook to illustrate how to train Pytorch models

Bug fixes

  • Minor fix for iterative averaging fusion algorithm.
  • Fix memory leakage issue in the Aggregator side. Now memory usage stays stable as the number of global rounds increases.
Assets 2
Loading