# tff.learning.algorithms.build_unweighted_mime_lite

<br />

|---|
| [![](https://www.tensorflow.org/images/GitHub-Mark-32px.png) View source on GitHub](https://github.com/tensorflow/federated/blob/v0.87.0 Version 2.0, January 2004 Licensed under the Apache License, Version 2.0 (the) |

Builds a learning process that performs Mime Lite.

    tff.learning.algorithms.build_unweighted_mime_lite(
        model_fn: Callable[[], https://www.tensorflow.org/federated/api_docs/python/tff/learning/models/VariableModel],
        base_optimizer: https://www.tensorflow.org/federated/api_docs/python/tff/learning/optimizers/Optimizer,
        server_optimizer: https://www.tensorflow.org/federated/api_docs/python/tff/learning/optimizers/Optimizer = sgdm.build_sgdm(1.0),
        model_distributor: Optional[https://www.tensorflow.org/federated/api_docs/python/tff/learning/templates/DistributionProcess] = None,
        model_aggregator: Optional[https://www.tensorflow.org/federated/api_docs/python/tff/aggregators/UnweightedAggregationFactory] = None,
        full_gradient_aggregator: Optional[https://www.tensorflow.org/federated/api_docs/python/tff/aggregators/UnweightedAggregationFactory] = None,
        metrics_aggregator: https://www.tensorflow.org/federated/api_docs/python/tff/learning/metrics/MetricsAggregatorType = https://www.tensorflow.org/federated/api_docs/python/tff/learning/metrics/sum_then_finalize,
        loop_implementation: https://www.tensorflow.org/federated/api_docs/python/tff/learning/LoopImplementation = https://www.tensorflow.org/federated/api_docs/python/tff/learning/LoopImplementation#DATASET_REDUCE
    ) -> https://www.tensorflow.org/federated/api_docs/python/tff/learning/templates/LearningProcess

This function creates a [`tff.learning.templates.LearningProcess`](https://www.tensorflow.org/federated/api_docs/python/tff/learning/templates/LearningProcess) that performs
Mime Lite algorithm on client models. The iterative process has the following
methods inherited from [`tff.learning.templates.LearningProcess`](https://www.tensorflow.org/federated/api_docs/python/tff/learning/templates/LearningProcess):

- `initialize`: A [`tff.Computation`](https://www.tensorflow.org/federated/api_docs/python/tff/Computation) with the functional type signature `( -> S@SERVER)`, where `S` is a [`tff.learning.templates.LearningAlgorithmState`](https://www.tensorflow.org/federated/api_docs/python/tff/learning/templates/LearningAlgorithmState) representing the initial state of the server.
- `next`: A [`tff.Computation`](https://www.tensorflow.org/federated/api_docs/python/tff/Computation) with the functional type signature `(<S@SERVER, {B*}@CLIENTS> -> <L@SERVER>)` where `S` is a [`tff.learning.templates.LearningAlgorithmState`](https://www.tensorflow.org/federated/api_docs/python/tff/learning/templates/LearningAlgorithmState) whose type matches the output of `initialize` and `{B*}@CLIENTS` represents the client datasets. The output `L` contains the updated server state, as well as aggregated metrics at the server, including client training metrics and any other metrics from distribution and aggregation processes.
- `get_model_weights`: A [`tff.Computation`](https://www.tensorflow.org/federated/api_docs/python/tff/Computation) with type signature `(S -> M)`, where `S` is a [`tff.learning.templates.LearningAlgorithmState`](https://www.tensorflow.org/federated/api_docs/python/tff/learning/templates/LearningAlgorithmState) whose type matches the output of `initialize` and `next`, and `M` represents the type of the model weights used during training.
- `set_model_weights`: A [`tff.Computation`](https://www.tensorflow.org/federated/api_docs/python/tff/Computation) with type signature `(<S, M> -> S)`, where `S` is a [`tff.learning.templates.LearningAlgorithmState`](https://www.tensorflow.org/federated/api_docs/python/tff/learning/templates/LearningAlgorithmState) whose type matches the output of `initialize` and `M` represents the type of the model weights used during training.

Each time the `next` method is called, the server model is communicated to
each client using the provided `model_distributor`. For each client, local
training is performed using `optimizer`, where its state is communicated by
the server, and kept intact during local training. The state is updated only
at the server based on the full gradient evaluated by the clients based on the
current server model state. The client full gradients are aggregated by
unweighted `full_gradient_aggregator`. Each client computes the difference
between the client model after training and its initial model. These model
deltas are then aggregated by unweighted `model_aggregator`. The aggregate
model delta is added to the existing server model state.

The Mime Lite algorithm is based on the paper
"Breaking the centralized barrier for cross-device federated learning."
Sai Praneeth Karimireddy, Martin Jaggi, Satyen Kale, Mehryar Mohri, Sashank
Reddi, Sebastian U. Stich, and Ananda Theertha Suresh.
Advances in Neural Information Processing Systems 34 (2021).
<https://proceedings.neurips.cc/paper/2021/file/f0e6be4ce76ccfa73c5a540d992d0756-Paper.pdf>

Note that Keras optimizers are not supported. This is due to the Mime Lite
algorithm applying the optimizer without changing it state at clients
(optimizer's [`tf.Variable`](https://www.tensorflow.org/api_docs/python/tf/Variable)s in the case of Keras), which is not possible with
Keras optimizers without reaching into private implementation details and
incurring additional computation and memory cost at clients.

<br />

<br />

<br />

| ## Args ||
|---|---|
| `model_fn` | A no-arg function that returns a [`tff.learning.models.VariableModel`](https://www.tensorflow.org/federated/api_docs/python/tff/learning/models/VariableModel). This method must *not* capture TensorFlow tensors or variables and use them. The model must be constructed entirely from scratch on each invocation, returning the same pre-constructed model each call will result in an error. |
| `base_optimizer` | A [`tff.learning.optimizers.Optimizer`](https://www.tensorflow.org/federated/api_docs/python/tff/learning/optimizers/Optimizer) which will be used for both creating and updating a global optimizer state, as well as optimization at clients given the global state, which is fixed during the optimization. |
| `server_optimizer` | A [`tff.learning.optimizers.Optimizer`](https://www.tensorflow.org/federated/api_docs/python/tff/learning/optimizers/Optimizer) which will be used for applying the aggregate model update to the global model weights. |
| `model_distributor` | An optional `DistributionProcess` that distributes the model weights on the server to the clients. If set to `None`, the distributor is constructed via `distributors.build_broadcast_process`. |
| `model_aggregator` | An optional [`tff.aggregators.UnweightedAggregationFactory`](https://www.tensorflow.org/federated/api_docs/python/tff/aggregators/UnweightedAggregationFactory) used to aggregate client updates on the server. If `None`, this is set to [`tff.aggregators.UnweightedMeanFactory`](https://www.tensorflow.org/federated/api_docs/python/tff/aggregators/UnweightedMeanFactory). |
| `full_gradient_aggregator` | An optional [`tff.aggregators.UnweightedAggregationFactory`](https://www.tensorflow.org/federated/api_docs/python/tff/aggregators/UnweightedAggregationFactory) used to aggregate the full gradients on client datasets. If `None`, this is set to [`tff.aggregators.UnweightedMeanFactory`](https://www.tensorflow.org/federated/api_docs/python/tff/aggregators/UnweightedMeanFactory). |
| `metrics_aggregator` | A function that takes in the metric finalizers (i.e., [`tff.learning.models.VariableModel.metric_finalizers()`](https://www.tensorflow.org/federated/api_docs/python/tff/learning/models/VariableModel#metric_finalizers)) and a [`tff.types.StructWithPythonType`](https://www.tensorflow.org/federated/api_docs/python/tff/types/StructWithPythonType) of the unfinalized metrics (i.e., the TFF type of [`tff.learning.models.VariableModel.report_local_unfinalized_metrics()`](https://www.tensorflow.org/federated/api_docs/python/tff/learning/models/VariableModel#report_local_unfinalized_metrics)), and returns a [`tff.Computation`](https://www.tensorflow.org/federated/api_docs/python/tff/Computation) for aggregating the unfinalized metrics. If `None`, this is set to [`tff.learning.metrics.sum_then_finalize`](https://www.tensorflow.org/federated/api_docs/python/tff/learning/metrics/sum_then_finalize). |
| `loop_implementation` | Changes the implementation of the training loop generated. See [`tff.learning.LoopImplementation`](https://www.tensorflow.org/federated/api_docs/python/tff/learning/LoopImplementation) for more details. |

<br />

<br />

<br />

<br />

| ## Returns ||
|---|---|
| A [`tff.learning.templates.LearningProcess`](https://www.tensorflow.org/federated/api_docs/python/tff/learning/templates/LearningProcess). ||

<br />