incerto.active.BALDAcquisition

class incerto.active.BALDAcquisition(num_samples=20)

Bases: BaseAcquisition

Bayesian Active Learning by Disagreement (BALD).

Selects samples that maximize the mutual information between predictions and model parameters.

I[y;θ|x] = H[y|x] - E_θ[H[y|x,θ]]

Reference:

Houlsby et al., “Bayesian Active Learning for Classification” (ICML 2011)

Parameters:

num_samples (int)

__init__(num_samples=20)

Initialize BALD acquisition.

Parameters:

num_samples (int) – Number of MC samples for Bayesian inference

Methods

__init__([num_samples])	Initialize BALD acquisition.
score(model, x, **kwargs)	Compute BALD scores.

__init__(num_samples=20)

Initialize BALD acquisition.

Parameters:

num_samples (int) – Number of MC samples for Bayesian inference

score(model, x, **kwargs)

Compute BALD scores.

Parameters:

• model (Module)

• x (Tensor)

Return type:

Tensor