cifar10/ - Code for cifar10 experiments (table 3)
cifar100/ - Code for cifar100 experiments (table 4)
mnist/ - Code for mnist experiments (table 2)
subsampled_cifar10/ - Code for subsampled cifar experiments (figure 3)
UCI/ - Code for UCI experiments (table 4)
- To run kernel code connect to docker by running:
bash docker_connect.sh
Notes:
-
Kernel code is tuned for tesla v100 GPUs, will be substanially slower on other gpus
-
All above mentioned code downloads a preprocessed cifar10/cifar100/cifar10.1 dataset from a public S3 bucket. But the raw pre-processing code used to generate those datasets can be inspected in
preprocess.py