def __init__(self, i, o):
super().__init__()
self.alpha = 0.1
b = torch.sign(torch.rand(o) - 0.5) # bit information to store random
self.register_buffer('b', b)
self.sign_loss_private = SignLoss(self.alpha, self.b)
self.linear = nn.Linear(i, o, bias=True)
self.weight = self.linear.weight
self.gn = nn.GroupNorm(o // 16, o, affine=False)
self.dropout = nn.Dropout(p=0.4)
self.fc = nn.Sequential(
nn.Linear(o, o // 4, bias=False),
nn.LeakyReLU(inplace=True),
nn.Linear(o // 4, o, bias=False),
)
self.register_buffer('key_private', None)
self.register_buffer('skey_private', None)
self.init_scale(True) #相比v1 此处是true
self.init_bias(True) #相比v1 此处是true
self.reset_parameters()
self.requires_reset_key = False
self.key_type = 'random'
def __init__(self, i, o):
super().__init__()
self.alpha = 0.1
b = torch.sign(torch.rand(o) - 0.5) # bit information to store random
self.register_buffer('b', b)
self.sign_loss_private = SignLoss(self.alpha, self.b)
self.linear = nn.Linear(i, o, bias=True)
self.weight = self.linear.weight
self.in_features = i
self.register_buffer('key_private', None)
self.register_buffer('skey_private', None)
self.init_scale(True) #相比v1 此处是true
self.init_bias(True) #相比v1 此处是true
self.reset_parameters()
self.requires_reset_key = False
self.key_type = 'random'
def __init__(self, i, o, ks=1):
super().__init__()
self.alpha = 0.1
b = torch.sign(torch.rand(o) - 0.5) # bit information to store random
self.register_buffer('b', b)
self.sign_loss_private = SignLoss(self.alpha, self.b)
self.conv = nn.Conv1d(i, o, ks, bias=False)
self.gn = nn.GroupNorm(o // 16, o, affine=False)
self.weight = self.conv.weight
self.register_buffer('key_private', None)
self.register_buffer('skey_private', None)
self.init_scale(True)
self.init_bias(True)
self.reset_parameters()
self.requires_reset_key = False
self.key_type = 'random'
def __init__(self, i, o, ks=1):
super().__init__()
self.alpha = 0.1
b = torch.sign(torch.rand(o) - 0.5) # bit information to store random
self.register_buffer('b', b)
self.sign_loss_private = SignLoss(self.alpha, self.b)
self.conv = nn.Conv1d(i, o, ks, bias=False)
self.bn0 = nn.BatchNorm1d(o, affine=False)
self.bn1 = nn.BatchNorm1d(o, affine=False)
# self.fc = nn.Sequential( )
self.fc = nn.Sequential(
nn.Linear(o, o // 4, bias=False),
nn.LeakyReLU(inplace=True),
nn.Linear(o // 4, o, bias=False),
)
# self.fc = nn.Sequential(
# nn.Linear(o, o // 4, bias=False),
# nn.LeakyReLU(inplace=True),
# nn.Linear(o // 4, o, bias=False),
# nn.Sigmoid()
# )
self.weight = self.conv.weight
self.register_buffer('key_private', None)
self.register_buffer('skey_private', None)
self.init_scale(True)
self.init_bias(True)
self.reset_parameters()
self.requires_reset_key = False
self.key_type = 'random'