Open Mandy-77 opened 3 years ago
The code for those experiments was more experimental and not part of the official release. It is still available to view: for some pointers, see https://github.com/HazyResearch/butterfly/blob/cnn/cnn/permuted_experiment.py for the entrypoint, https://github.com/HazyResearch/butterfly/blob/432c8d1d5ae8008f9c13bc13035d1c4d7bf7a0c4/cnn/models/presnet.py#L347 for the main permutation model, and https://github.com/HazyResearch/butterfly/blob/cnn/cnn/permutation_utils.py for other utilities described in the paper (e.g. the auxiliary unsupervised objective).
Hi, thanks for sharing your codes. I was wondering how to cover the results shown in Fig. 3 in the paper 'Kaleidoscope: An Efficient, Learnable Representation For All Structured Linear Maps', which used K-matrix and ResNet18 to learn permutation.