Home / Curated / Mathematical Methods / Layer-wise relevance propagation

Layer-wise relevance propagation

Mathematical Methods AIMech:LayerWiseRelevancePropagation DRAFT

Aliases: LRP

A prediction-decomposition framework that redistributes relevance backward through a classifier's computational layers toward its input features.

Identity Notes & Concept Scope LRP is a family of constrained relevance-propagation constructions, demonstrated for neural networks and kernel-based Bag-of-Words classifiers; it is not a single gradient formula or a classifier-training algorithm.

Claims & Grounded Evidence

1 Claim

Mechanism #attribution-method

Starting from an output score, layer-specific rules allocate relevance messages to preceding units and sum incoming messages until the input is reached. The neural-network procedure supplies stabilized and positive/negative contribution rules.

supports Source: On Pixel-Wise Explanations for Non-Linear Classifier Decisions by Layer-Wise Relevance Propagation · Locator: Layer-wise relevance propagation, equation 2, PDF/printed pp. 3–4; Layer-wise relevance backpropagation, equations 55–62 and Algorithm 2, pp. 20–21

⚠️ Curated Limitations & Negative Results

Conservation alone does not uniquely determine meaningful attributions; the original paper gives a counterexample.
Bias terms and epsilon stabilization can absorb relevance, so exact conservation is not asserted for every practical rule.

Artificial Intelligence Ontology (AIO) Grounding

Mapped AIO Classes

AIO Target Class	Mapping Relation	Mapping Rationale
`AIO:ArtificialNeuralNetwork` Artificial Neural Network	related_match	The paper explicitly constructs LRP for artificial neural networks. This related mapping records that supported analysis relationship while retaining LRP's additional kernel-classifier scope; it asserts neither equivalence nor exclusive applicability.

Evidence Sources & Primary Literature

1 Source

Source ID	Publication / Document	Type / Access	Reference & Checksum
`original-paper`	On Pixel-Wise Explanations for Non-Linear Classifier Decisions by Layer-Wise Relevance Propagation Sebastian Bach, Alexander Binder, Grégoire Montavon, Frederick Klauschen, Klaus-Robert Müller, Wojciech Samek	primary paper full_text	ID: doi:10.1371/journal.pone.0130140 sha256:212147782b3e…

Curation Provenance

Curator: Codex

Method: ai_assisted

Model: GPT-6

Created: 2026-10-11T03:54:14Z

Updated: 2026-10-11T03:54:14Z

Source File: data/curated/LayerWiseRelevancePropagation.yaml