Full Gradient Negation¶

Full-gradient negation attack.

class attacks.full_gradient_negation.FullGradientNegationAttack[source]¶

Bases: Attack

Full-gradient negation attack.

Each Byzantine worker sends the negation of the full-dataset gradient scaled by a factor \(\kappa\). The attack assumes knowledge of the true gradient over the entire dataset.

classmethod generate(honest_gradients: Sequence[Tensor] | Tensor, /, out: Tensor | None = None, *, f: int, full_gradient: Tensor, kappa: float = 100.0, **specialized: Any) → Tensor[source]¶

Generate Byzantine gradients.

Parameters:

honest_gradients – Sequence of \(h\) gradient vectors, one per honest worker, each of shape \((d,)\).
out – Optional pre-allocated tensor of shape \((f, d)\) to write the result into and return.
f – Number of Byzantine gradients to generate.
full_gradient – Tensor of shape \((d,)\) containing the gradient computed over the entire dataset.
kappa – Scale factor applied to the negated full gradient.
**specialized – Additional keyword arguments.

Returns:

Byzantine gradients of shape `` (f, d)
empty tensor of shape `` (0, d)

Raises:

ValueError – If \(\kappa\) or \(f\) is negative, there are no honest gradients, or full_gradient is not 1-D.
TypeError – If the honest gradients do not use a floating-point dtype.