Gradient-weighted class activation mapping
Algorithms AIMech:GradCAM DRAFT
Aliases: Grad-CAM
A convolutional-network localization method weighting activation maps by spatially averaged gradients of a selected target output.
Identity Notes & Concept Scope
Grad-CAM computes a coarse activation-space localization map. It differs from architecture-restricted CAM and from Guided Grad-CAM's additional input-space fusion.
Claims & Grounded Evidence
1 Claim
Mechanism
#explanation-method
Differentiate the target score with respect to convolutional feature maps, average gradients spatially to obtain channel weights, and apply ReLU to the weighted sum of activation maps.
⚠️ Curated Limitations & Negative Results
- Map resolution follows the selected convolutional layer, and ReLU retains positive weighted contributions.
- A localization map is not a proof of causal influence or universal faithfulness; evaluation claims are not curated here.
Artificial Intelligence Ontology (AIO) Grounding
Mapped AIO Classes
| AIO Target Class | Mapping Relation | Mapping Rationale |
|---|---|---|
AIO:ArtificialNeuralNetwork
Artificial Neural Network
|
related_match | The method analyzes convolutional artificial-neural-network activations and target gradients. This is a direct analysis relation, not architecture identity. |
Evidence Sources & Primary Literature
1 Source| Source ID | Publication / Document | Type / Access | Reference & Checksum |
|---|---|---|---|
original-paper
|
Ramprasaath R. Selvaraju, Michael Cogswell, Abhishek Das, Ramakrishna Vedantam, Devi Parikh, Dhruv Batra
|
primary paper
full_text
|
sha256:fa3192ab0f81…
|
Curation Provenance
Curator: Codex
Method: ai_assisted
Model: GPT-6
Created: 2026-10-11T04:02:04Z
Updated: 2026-10-11T04:02:04Z
Source File:
data/curated/GradCAM.yaml