What programming language and framework does Knowledge-Distillation-Zoo use?
Knowledge-Distillation-Zoo is implemented using Python 3.7 and relies on the PyTorch deep learning framework (version 1.3.1) along with torchvision (version 0.4.2). This makes it accessible for researchers and developers familiar with the PyTorch ecosystem.
Which Knowledge Distillation methods are implemented in this repository?
The repository includes implementations for a variety of KD methods such as Logits mimic learning, ST soft target, AT attention transfer, Fitnet, NST neural selective transfer, PKT, FSP, FT, RKD, AB, SP, Sobolev, BSS, CC, LwM, IRG, VID, OFD, AFD, CRD, and DML. It focuses on basic methods for ease of reference.
What datasets and network architectures are supported for experiments?
Knowledge-Distillation-Zoo supports experiments on the CIFAR10 and CIFAR100 datasets. For network architectures, it primarily uses Resnet-20 and Resnet-110, allowing for comparative studies of KD methods across these common models and datasets.