Skip to content

relax

muppet.explainers.relax

RELAX explainer for embedding-based model explanations.

This module implements the RELAX explainer, a perturbation-based XAI method that generates explanations by analyzing how perturbations affect the model's embeddings. It can be seen as a extention of RISE, from classification to representation learning task.

MUPPET Component Integration
  • Explorer: RandomMasksExplorer - generates random binary masks for input perturbation
  • Perturbator: SetToZeroPerturbator - applies zero-masking to perturb input regions
  • Attributor: EmbeddingDistanceAttributor - measures changes in model embeddings
  • Aggregator: WeightedSumAggregator - combines perturbations weighted by embedding changes

Classes:

  • RELAXExplainer –

    Implementation of the RELAX method for embedding-based explanations.

Refenrecs

Wickstrøm, Kristoffer K., et al. "RELAX: Representation learning explainability." International Journal of Computer Vision 131.6 (2023): 1584-1610. https://arxiv.org/pdf/2112.10161

Classes

RELAXExplainer 
RELAXExplainer(
    model, nmasks, mask_dim, mask_proba, seed=None
)

Bases: MuppetExplainer

RELAX explainer implementation for embedding-based explanations.

Implements the RELAX method that explains representation learning models based on the RISE method principles.

Key characteristics: - Uses random masking perturbations similar to RISE - Analyzes embedding output changes

By measuring how perturbations affect these embeddings, RELAX can identify which parts of the input are most critical for the embedding understanding.

Initialize the RELAX explainer for representation learning explanations.

Parameters:

  • nmasks (int) –

    Number of random masks to generate.

  • mask_dim (int) –

    The size of the squared grade (down-scaled mask).

  • mask_proba (float) –

    The probability of setting, independently, each pixel of the mask to 1.

  • seed (int, default: None ) –

    Random seed for reproducibility.

Source code in muppet/explainers/relax.py 
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
def __init__(
    self,
    model: torch.nn.Module,
    nmasks: int,
    mask_dim: int,
    mask_proba: float,
    seed: int | None = None,
):
    """Initialize the RELAX explainer for representation learning explanations.

    Args:
        nmasks (int): Number of random masks to generate.
        mask_dim (int): The size of the squared grade (down-scaled mask).
        mask_proba (float): The probability of setting, independently, each pixel of the mask to 1.
        seed (int, optional): Random seed for reproducibility.
    """
    self.nmasks = nmasks
    self.mask_dim = mask_dim
    self.mask_proba = mask_proba

    # existing components
    explorer = RandomMasksExplorer(
        nmasks=self.nmasks,
        mask_dim=self.mask_dim,
        mask_proba=self.mask_proba,
        seed=seed,
    )
    perturbator = SetToZeroPerturbator()

    attributor = EmbeddingDistanceAttributor()

    aggregator = WeightedSumAggregator()

    super().__init__(
        model=model,
        explorer=explorer,
        perturbator=perturbator,
        attributor=attributor,
        aggregator=aggregator,
    )