ARMORY v0.9.0

Changelog

Infrastructure

• Enable multiple external repos to be pulled (#617)

Configuration Files

• Add ART default parameters to configs for clarity (#623)

Scenarios

• Update sincnet baseline scenario to utilize full audio during inference (#625 #607)
• Allow separate models for training on poisoned and cleaned datasets (#611)
• Make poisoned splits deterministic (#624)

Examples

• Different training and inference mode examples in armory-example (twosixlabs/armory-example#44) and in librispeech (#625)

Datasets

• Experimental subset of PyTorch native DataLoaders (#618)
• Enable targeted attack labels in preloaded datasets (#622)