使用tensorflow进行联邦学习  

下载tensorflow_federated模块，此处使用较稳定的0.13.1版本，可根据实际情况选择版本  

pip3 install --upgrade tensorflow_federated==0.13.1 -i http://pypi.douban.com/simple --trusted-host pypi.douban.com

引入所需模块

import tensorflow as tf
import tensorflow_federated as tff

下载手写体数字识别数据集

source, _ = tff.simulation.datasets.emnist.load_data()
def client_data(n):
  return source.create_tf_dataset_for_client(source.client_ids[n]).map(
      lambda e: (tf.reshape(e['pixels'], [-1]), e['label'])
  ).repeat(10).batch(20)
  # Pick a subset of client devices to participate in training.
train_data = [client_data(n) for n in range(3)]
# Grab a single batch of data so that TFF knows what data looks like.
sample_batch = tf.nest.map_structure(
    lambda x: x.numpy(), iter(train_data[0]).next())

定义模型

# Wrap a Keras model for use with TFF.
def model_fn():
  model = tf.keras.models.Sequential([
      tf.keras.layers.Dense(10, tf.nn.softmax, input_shape=(784,),
                            kernel_initializer='zeros')
  ])
  return tff.learning.from_keras_model(
      model,
      dummy_batch=sample_batch,
      loss=tf.keras.losses.SparseCategoricalCrossentropy(),
      metrics=[tf.keras.metrics.SparseCategoricalAccuracy()])

训练模型

# Simulate a few rounds of training with the selected client devices.
trainer = tff.learning.build_federated_averaging_process(
  model_fn,
  client_optimizer_fn=lambda: tf.keras.optimizers.SGD(0.1))
state = trainer.initialize()
for _ in range(20):
  state, metrics = trainer.next(state, train_data)
  print (metrics.loss)

训练结果

Instructions for updating:
If using Keras pass *_constraint arguments to layers.
11.67063
11.65545
11.121943
10.477433
8.830841
7.7931805
7.192688
5.1021852
5.1880836
3.3151658
3.3807354
3.3242536
2.30087
1.7061236
1.5317304
1.1874161
0.7962298
0.94748044
0.7774023
0.6510306