Deep Sets
Architectures AIMech:DeepSets DRAFT
A neural architecture family for unordered collections that combines shared element transformations with symmetric aggregation for invariant outputs, or shared element and pooled-context computations for permutation-equivariant outputs.
Identity Notes & Concept Scope
Deep Sets denotes the architecture family defined by Zaheer and colleagues, including invariant and equivariant constructions. It is not a synonym for all set functions, PointNet, Set Transformer or graph neural networks.
Claims & Grounded Evidence
1 Claim
Mechanism
#architecture-mechanism
The invariant construction applies a shared neural transformation to each element, sums the transformed representations, and processes the sum with an output network. The equivariant construction combines an element's features with a shared pooled summary using tied parameters, and such layers can be composed.
⚠️ Curated Limitations & Negative Results
- The source's exact set-function representation theorem treats a countable universe; its uncountable-domain discussion establishes a fixed-size case and states a broader conjecture. No unrestricted universality claim is curated.
- Permutation invariance and equivariance are different output properties. The selected architecture, aggregation and finite neural representation determine its actual scope.
Artificial Intelligence Ontology (AIO) Grounding
Mapped AIO Classes
| AIO Target Class | Mapping Relation | Mapping Rationale |
|---|---|---|
AIO:ArtificialNeuralNetwork
Artificial Neural Network
|
broad_match | The source constructs trainable neural transformations and shared neural layers for set-valued inputs; the artificial-neural-network target has broader scope. |
Evidence Sources & Primary Literature
1 Source| Source ID | Publication / Document | Type / Access | Reference & Checksum |
|---|---|---|---|
original-paper
|
Manzil Zaheer, Satwik Kottur, Siamak Ravanbakhsh, Barnabás Póczos, Ruslan Salakhutdinov, Alexander J. Smola
|
primary paper
full_text
|
sha256:ae1e9a4655b8…
|
Curation Provenance
Curator: Codex
Method: ai_assisted
Model: GPT-6
Created: 2026-10-11T04:11:44Z
Updated: 2026-10-11T04:11:44Z
Source File:
data/curated/DeepSets.yaml