secretflow.security.privacy.mechanism.tensorflow#

secretflow.security.privacy.mechanism.tensorflow.layers#

Classes:

EmbeddingDP(*args, **kwargs)

GaussianEmbeddingDP(*args, **kwargs)

Embedding differential privacy perturbation using gaussian noise

LabelDP(eps)

Label differential privacy perturbation
class secretflow.security.privacy.mechanism.tensorflow.layers.EmbeddingDP(*args, **kwargs)[源代码]#

基类：Layer, ABC

Methods:

__init__()

call(inputs)

This is where the layer's logic lives.
__init__() → None[源代码]#
abstract call(inputs)[源代码]#

This is where the layer’s logic lives.

The call() method may not create state (except in its first invocation, wrapping the creation of variables or other resources in tf.init_scope()). It is recommended to create state, including tf.Variable instances and nested Layer instances,

in __init__(), or in the build() method that is

called automatically before call() executes for the first time.

参数:

  • inputs –

    Input tensor, or dict/list/tuple of input tensors. The first positional inputs argument is subject to special rules: - inputs must be explicitly passed. A layer cannot have zero

    arguments, and inputs cannot be provided via the default value of a keyword argument.

    • NumPy array or Python scalar values in inputs get cast as tensors.

    • Keras mask metadata is only collected from inputs.

    • Layers are built (build(input_shape) method) using shape info from inputs only.

    • input_spec compatibility is only checked against inputs.

    • Mixed precision input casting is only applied to inputs. If a layer has tensor arguments in *args or **kwargs, their casting behavior in mixed precision should be handled manually.

    • The SavedModel input specification is generated using inputs only.

    • Integration with various ecosystem packages like TFMOT, TFLite, TF.js, etc is only supported for inputs and not for tensors in positional and keyword arguments.

  • *args – Additional positional arguments. May contain tensors, although this is not recommended, for the reasons above.

  • **kwargs –

    Additional keyword arguments. May contain tensors, although this is not recommended, for the reasons above. The following optional keyword arguments are reserved: - training: Boolean scalar tensor of Python boolean indicating

    whether the call is meant for training or inference.

    • mask: Boolean input mask. If the layer’s call() method takes a mask argument, its default value will be set to the mask generated for inputs by the previous layer (if input did come from a layer that generated a corresponding mask, i.e. if it came from a Keras layer with masking support).

返回:

A tensor or list/tuple of tensors.

class secretflow.security.privacy.mechanism.tensorflow.layers.GaussianEmbeddingDP(*args, **kwargs)[源代码]#

基类：EmbeddingDP

Embedding differential privacy perturbation using gaussian noise

Methods:

__init__(noise_multiplier, batch_size, ...)

param epnoise_multipliers:

Epsilon for pure DP.

call(inputs)

Add gaussion dp on embedding.

privacy_spent_rdp(step[, orders])

Get accountant using RDP.

privacy_spent_gdp(step, sampling_type)

Get accountant using GDP.
__init__(noise_multiplier: float, batch_size: int, num_samples: int, l2_norm_clip: float = 1.0, delta: Optional[float] = None, is_secure_generator: bool = False) → None[源代码]#
参数:

  • epnoise_multipliers – Epsilon for pure DP.

  • batch_size – Batch size.

  • num_samples – Number of all samples.

  • l2_norm_clip – The clipping norm to apply to the embedding.

  • is_secure_generator – whether use the secure generator to generate noise.

call(inputs)[源代码]#

Add gaussion dp on embedding.

参数:

inputs – Embedding.

privacy_spent_rdp(step: int, orders: Optional[List] = None)[源代码]#

Get accountant using RDP.

参数:

  • step – The current step of model training or prediction.

  • orders – An array (or a scalar) of RDP orders.

privacy_spent_gdp(step: int, sampling_type: str)[源代码]#

Get accountant using GDP.

参数:

  • step – The current step of model training or prediction.

  • sampling_type – Sampling type, which must be “poisson” or “uniform”.

class secretflow.security.privacy.mechanism.tensorflow.layers.LabelDP(eps: float)[源代码]#

基类：object

Label differential privacy perturbation

Methods:

__init__(eps)

param eps:

epsilon for pure DP.

privacy_spent()

__init__(eps: float) → None[源代码]#
参数:

eps – epsilon for pure DP.

privacy_spent()[源代码]#

secretflow.security.privacy.mechanism.tensorflow.mechanism_fl#

Classes:

GaussianModelDP(noise_multiplier, num_clients)

global model differential privacy perturbation using gaussian noise
class secretflow.security.privacy.mechanism.tensorflow.mechanism_fl.GaussianModelDP(noise_multiplier: float, num_clients: int, num_updates: Optional[int] = None, l2_norm_clip: float = 1.0, delta: Optional[float] = None, is_secure_generator: bool = False, is_clip_each_layer: bool = False)[源代码]#

基类：object

global model differential privacy perturbation using gaussian noise

Methods:

__init__(noise_multiplier, num_clients[, ...])

param epnoise_multipliers:

Epsilon for pure DP.

privacy_spent_rdp(step[, orders])

Get accountant using RDP.
__init__(noise_multiplier: float, num_clients: int, num_updates: Optional[int] = None, l2_norm_clip: float = 1.0, delta: Optional[float] = None, is_secure_generator: bool = False, is_clip_each_layer: bool = False) → None[源代码]#
参数:

  • epnoise_multipliers – Epsilon for pure DP.

  • num_clients – Number of all clients.

  • num_updates – Number of Clients that participate in the update.

  • l2_norm_clip – The clipping norm to apply to the parameters or gradients.

  • is_secure_generator – whether use the secure generator to generate noise.

  • is_clip_prelayer – The 2norm of each layer is dynamically assigned.

privacy_spent_rdp(step: int, orders: Optional[List] = None)[源代码]#

Get accountant using RDP.

参数:

  • step – The dp current step of model training or prediction.

  • orders – An array (or a scalar) of RDP orders.