Mamba sequence architecture
Architectures AIMech:Mamba DRAFT
Aliases: Mamba
A neural sequence architecture built from repeated blocks containing input-dependent selective state-space transformations.
Identity Notes & Concept Scope
The original Mamba architecture, not its trained checkpoints, all state-space models, or Mamba-2. Existing AIO recurrent and deep-network categories do not specify selective state-space blocks.
Claims & Grounded Evidence
1 Claim
Mechanism
#mechanism
Input-dependent B, C and time-step parameters control selective state updates; a scan computes the recurrence inside repeated gated blocks.
⚠️ Curated Limitations & Negative Results
- This record covers the original architecture, not every architecture called a state-space model.
- Reported throughput and benchmark results were not independently reproduced or generalized to arbitrary hardware or tasks.
Artificial Intelligence Ontology (AIO) Grounding
Mapped AIO Classes
| AIO Target Class | Mapping Relation | Mapping Rationale |
|---|---|---|
AIO:DeepNeuralNetwork
Deep Neural Network
|
broad_match | Repeated learned blocks form a multilayer neural network; the target is broader than this specific architecture. |
AIO:RecurrentNeuralNetwork
Recurrent Neural Network
|
related_match | Selective state updates use a temporal recurrence, but this mapping does not equate all recurrent networks with Mamba. |
Evidence Sources & Primary Literature
1 Source| Source ID | Publication / Document | Type / Access | Reference & Checksum |
|---|---|---|---|
original-paper
|
Albert Gu, Tri Dao
|
primary paper
full_text
|
ID: arXiv:2312.00752v2
sha256:adf70ed1803c…
|
Curation Provenance
Curator: Codex
Method: ai_assisted
Model: GPT-6
Created: 2026-10-11T03:02:19Z
Updated: 2026-10-11T03:02:19Z
Source File:
data/curated/Mamba.yaml