SmoothGrad
Algorithms AIMech:SmoothGrad DRAFT
A sensitivity-map smoothing method that averages maps computed on Gaussian-noise perturbations of an input.
Identity Notes & Concept Scope
SmoothGrad is an averaging wrapper around a sensitivity-map method, not a new trained network, a training-noise procedure or an exact equivalent of its underlying gradient map.
Claims & Grounded Evidence
1 Claim
Mechanism
#attribution-method
Sample multiple noisy copies of the input, compute the selected sensitivity map on each, and average the maps to estimate a local smoothing of that map function.
⚠️ Curated Limitations & Negative Results
- The result depends on the base map, noise scale and sample count.
- The paper's visual-coherence comparisons and proposed explanation for noise do not establish universal attribution faithfulness.
Artificial Intelligence Ontology (AIO) Grounding
Mapped AIO Classes
| AIO Target Class | Mapping Relation | Mapping Rationale |
|---|---|---|
AIO:ArtificialNeuralNetwork
Artificial Neural Network
|
related_match | The defining method smooths sensitivity maps of artificial-neural-network outputs. This direct analysis relationship is not an assertion that SmoothGrad is itself a network. |
Evidence Sources & Primary Literature
1 Source| Source ID | Publication / Document | Type / Access | Reference & Checksum |
|---|---|---|---|
original-paper
|
Daniel Smilkov, Nikhil Thorat, Been Kim, Fernanda Viégas, Martin Wattenberg
|
primary paper
full_text
|
ID: arXiv:1706.03825v1
sha256:5a018a58df31…
|
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/SmoothGrad.yaml