[−][src]Module xaynet_server::examples

A guide to getting started with the XayNet examples.

Examples

The XayNet examples code can be found under the rust/examples directory of the xaynet repository.

This Getting Started guide will cover only the general ideas around usage of the examples. Also see the source code of the individual examples themselves, which have plenty of comments.

Running an example typically requires having a coordinator already running, which is the core component of XayNet.

Federated Learning

A federated learning session over XayNet consists of two kinds of parties - a coordinator and (multiple) participants. The two parties engage in a protocol (called PET) over a series of rounds. The over-simplified idea is that in each round:

  1. The coordinator makes available a global model, from which selected participants will train model updates (or, local models) to be sent back to the coordinator.

  2. As a round progresses, the coordinator aggregates these updates into a new global model.

From this description, it might appear that individual local models are plainly visible to the coordinator. What if sensitive data could be extracted from them? Would this not be a violation of participants' data privacy?

In fact, a key point about this process is that the updates are not sent in the plain! Rather, they are sent encrypted (or masked) so that the coordinator (and by extension, XayNet) learns almost nothing about the individual updates. Yet, it is nevertheless able to aggregate them in such a way that the resulting global model is unmasked.

This is essentially what is meant by federated learning that is privacy-preserving, and is a key feature enabled by the PET protocol.

PET Protocol

It is worth describing the protocol very briefly here, if only to better understand some of the configuration settings we will meet later. It is helpful to think of each round being divided up into several contiguous phases:

Start. At the start of a round, the coordinator generates a collection of random round parameters for all participants. From these parameters, each participant is able to determine whether it is selected for the round and if so, which of the two roles it is:

  • update participants.

  • sum participants.

Sum. In the Sum phase, sum participants send sum messages to the coordinator (the details of which are not so important here, but vital for computing sum2 messages later).

Update. In the Update phase, each update participant obtains the global model from the coordinator, trains a local model from it, masks it, and sends it to the coordinator in the form of update messages. The coordinator will internally aggregate these (masked) local models.

Sum2. In the Sum2 phase, sum participants compute the sum of masks over all the local models, and sends it to the coordinator in the form of sum2 messages.

Equipped with the sum of masks, the coordinator is able to unmask the aggregated global model, for the next round.

This short description of the protocol skips over many details, but is sufficient for the purposes of this guide. For a much more complete specification, see the white paper.

Coordinator

The coordinator is configurable via various settings. The project contains various ready-made configuration files that can be used, found under the configs directory of the repository. Typically they look something like the following (in TOML format):

[api]
bind_address = "127.0.0.1:8081"

[pet]
min_sum_count = 1
min_update_count = 3
min_sum_time = 5
min_update_time = 10
sum = 0.4
update = 0.9

[mask]
group_type = "Prime"
data_type = "F32"
bound_type = "B0"
model_type = "M3"

[model]
size = 4

The actual files contain more settings than this, but we mention just the selection above because they will be the most relevant for this guide.

Settings

Going from the top, the ApiSettings include the address the coordinator should listen on for requests from participants. This address should be known to all participants.

The PetSettings specify various parameters of the PET protocol:

  • The most important are sum and update, which are the probabilities assigned to the selection of sum and update participants, respectively (note that if a participant is selected for both roles, the sum role takes precedence).

  • The settings min_sum_count and min_update_count specify, respectively, the minimum number of sum/sum2 and update messages the coordinator should accept. By default, they are set to the theoretical minimum in order for the protocol to function correctly.

  • To complement, the settings min_sum_time and min_update_time specify, respectively, the minimum amount of time (in seconds) the coordinator should wait for sum/sum2 and update messages. To allow for more messages to be processed, increase these times.

The MaskSettings determines the masking configuration, consisting of the group type, data type, bound type and model type. The ModelSettings specify the size of the model used. Both of these settings should be decided in advance with participants, and agreed upon by both.

Running

The coordinator can be run as follows:

$ git clone git://github.com/xaynetwork/xaynet
$ cd xaynet/rust
$ cargo run --bin coordinator -- -c ../configs/config.toml

We will assume configs/config.toml includes the values shown in the excerpt above.

Connecting with Participants

The below shows an example of how participants can be created to connect to the coordinator running as instructed above. It is a slightly adapted version of the test-drive-net example, for illustrative purposes.

use tokio::signal;
use xaynet_client::{
    api::{HttpApiClient, HttpApiClientError},
    Client,
    ClientError,
};
use xaynet_core::mask::{FromPrimitives, Model};

#[tokio::main]
async fn main() -> Result<(), ClientError<HttpApiClientError>> {
    let len = 4_usize;
    let model = Model::from_primitives(vec![0; len].into_iter()).unwrap();

    let mut clients = Vec::with_capacity(20_usize);
    for id in 0..20 {
        let api_client = HttpApiClient::new("http://127.0.0.1:8081");
        let mut client = Client::new(1, id, api_client)?;
        client.local_model = Some(model.clone());
        let join_hdl = tokio::spawn(async move {
            tokio::select! {
                _ = signal::ctrl_c() => {}
                result = client.start() => {
                    println!("{:?}", result);
                }
            }
        });
        clients.push(join_hdl);
    }

    for client in clients {
        let _ = client.await;
    }

    Ok(())
}

As this example is meant solely as a demonstration of the operation of the PET protocol, it does not perform any training as such (see other forthcoming examples for that). Instead, a "dummy" zero-Model is used throughout. Nevertheless, note that its length and contents, respectively, are required to match the size and mask configuration expected by the coordinator.

The example creates the expected number of participants (called Clients here), connecting to the address where the coordinator should be listening. The participants are then started, and run continuously until stopped (with CTRL + C).

Running

The actual test-drive-net example is a tidier version of the above, where the hard-coded numbers are made configurable. To run:

$ cargo run --example test-drive-net -- -l 4 -n 20 -u http://127.0.0.1:8081