Examples of implementing HEU in the framework of federal learning

[AustinApple]

Issue Type

Others

Source

binary

Secretflow Version

SecretFlow 1.2.0.dev20231009

OS Platform and Distribution

macOS 13.2.1 (22D68)

Python version

Python 3.8.18

Bazel version

No response

GCC/Compiler version

No response

What happend and What you expected to happen.

Are there any examples of implementing HEU in the framework of federal learning?

[ian-huu]

You can take a look at the implementation of FlLogisticRegressionMix.

https://github.com/secretflow/secretflow/blob/main/secretflow/ml/linear/fl_lr_mix.py

[AustinApple]

Thank you for the response. I am wondering how to implement HEU in the FLModel() class. For example, the following is my original code. I wanna use encrypted input to train the model, but I find that fit() function does not support heus argument.

heu = sf.HEU(heu_config('charlie', ['alice', 'bob']), spu.spu_pb2.FM128)
aggregator = SecureAggregator(server, [alice, bob])

fl_model = FLModel(
    server=server,
    device_list=device_list,
    model=model_def,
    aggregator=aggregator,
    strategy='fed_avg_w',
    backend="torch"
)
history = fl_model.fit(
    train_x,
    train_y,
    validation_data=(val_x, val_y),
    epochs=1,
    batch_size=128,
    aggregate_freq=1,
)

[ian-huu]

If you want to use HEU for aggregation, you can write a custom aggregator (you can refer to https://github.com/secretflow/secretflow/blob/main/secretflow/security/aggregation/spu_aggregator.py ).
If you want to use HEU for other purposes, you may need to inherit or even rewrite FLModel to achieve it.