Currently only support approximate privacy. Here we assume we use Gaussian noise on randomly sampled batch so we get better composition: 1. the per batch privacy is computed using privacy amplication via sampling bound; 2. the composition is done using the composition with Gaussian noise. TODO(l...
gaussian_gray_noise_prob=0.4, poisson_scale=[0.05, 2.5], poisson_gray_noise_prob=0.4, gaussian_sigma_step=0.1, poisson_scale_step=0.005), keys=['lq'], ), dict( type='RandomJPEGCompression', params=dict(quality=[30, 95], quality_step=3), keys=['lq'], ), dict( type='Degradation...
"module": "$@discriminator_def.to(@device)", "device_ids": [ "@device" ] }, "train#sampler": { "_target_": "DistributedSampler", "dataset": "@train#dataset", "even_divisible": true, "shuffle": true }, "train#dataloader#sampler": "@train#sampler", "train#dataloader#shuffle": ...