Train local model with SVM instead of NN in federated learning

Issue

I have a dataset with numeric features and labels. I am building a federated learning model using TensorFlow (TFF).
Basically, the model that I have is the (neural network) which is always explained in the TFF tutorials.
I want to ask if there is a chance to build another model for the local clients, such as SVM? since it suits my dataset.

My neural network:

def create_keras_model():
  initializer = tf.keras.initializers.Zeros()
  return tf.keras.models.Sequential([
      tf.keras.layers.Input(shape=(18,)),
      tf.keras.layers.Dense(128),
      tf.keras.layers.Dense(4, kernel_initializer= initializer),
      tf.keras.layers.Softmax(),
  ])
def model_fn():
  keras_model = create_keras_model()
  return tff.learning.from_keras_model(
      keras_model,
      input_spec=train_data[0].element_spec,
      loss=tf.keras.losses.SparseCategoricalCrossentropy(),
      metrics=[tf.keras.metrics.SparseCategoricalAccuracy()]
      )

Solution

TFF supports a wide variety of models, including just about any model you can write in tf.keras.

You can also create a TFF model directly by subclassing https://www.tensorflow.org/federated/api_docs/python/tff/learning/Model with the code for your forward pass. If you are interested in a more functional approach, you could also define an SVM model via TFF’s FunctionalModel https://www.tensorflow.org/federated/api_docs/python/tff/learning/models/FunctionalModel.

Answered By – Zachary Charles

This Answer collected from stackoverflow, is licensed under cc by-sa 2.5 , cc by-sa 3.0 and cc by-sa 4.0