Brute¶

Brute aggregation rule, most-clumped-subset baseline.

Reference:: El Mahdi El Mhamdi, Rachid Guerraoui, and Sébastien Rouault. “The Hidden Vulnerability of Distributed Learning in Byzantium.” In Proceedings of the 35th International Conference on Machine Learning (ICML 2018).

class aggregators.brute.Brute[source]¶

Bases: Aggregator

Brute aggregation rule, most-clumped subset selection.

For every subset \(R\) of size \(n - f\) from the submitted gradients, define its clumping score as \(\max_{i, j \in R} \|V_i - V_j\|^2\). The aggregator picks the subset with the smallest clumping score and returns its mean. This enumerates \(\binom{n}{n-f}\) subsets, so it is only feasible when that count is small (the paper uses \(6\) honest + \(5\) Byzantine workers, giving \(\binom{11}{6} = 462\) subsets).

classmethod aggregate(gradients: Sequence[Tensor] | Tensor, /, out: Tensor | None = None, *, n: int, f: int, **specialized: Any) → Tensor[source]¶

Aggregate gradients by selecting the most-clumped \(n - f\) subset.

Parameters:

gradients – Sequence of 1-D tensors containing gradients from workers.
out – Optional pre-allocated tensor to write the result into.
n – Total number of workers.
f – Number of Byzantine workers to tolerate.
**specialized – Additional keyword arguments.

Returns:

Mean of the selected ``n - f`` subset, of shape `` (d,)

Raises:

ValueError – If \(n\), \(f\), or the gradients count is invalid.