Crate omega_attention 

Omega Attention - Brain-Like Selective Processing

Implements 39 attention mechanisms inspired by transformer architectures and neuroscience research, plus brain-like attention control systems.

Features

• 39 Attention Mechanisms: Flash, Linear, Sparse, Hyperbolic, Graph, Memory-augmented
• Top-Down Attention: Goal-driven, task-relevant selection
• Bottom-Up Attention: Stimulus-driven, salience-based capture
• Working Memory Gating: Input/output/forget gates for WM control
• Attention Spotlight: Winner-take-all competition

Architecture

┌─────────────────────────────────────────────────────────────┐
│                   ATTENTION SYSTEM                          │
├─────────────────────────────────────────────────────────────┤
│                                                              │
│  ┌────────────────────┐    ┌────────────────────┐           │
│  │   TOP-DOWN         │    │   BOTTOM-UP        │           │
│  │   (Goal-driven)    │    │   (Salience)       │           │
│  │                    │    │                    │           │
│  │  • Task relevance  │    │  • Novelty         │           │
│  │  • Expected value  │    │  • Contrast        │           │
│  │  • Memory match    │    │  • Motion          │           │
│  └────────┬───────────┘    └────────┬───────────┘           │
│           │                         │                        │
│           └───────────┬─────────────┘                        │
│                       ▼                                      │
│           ┌───────────────────────┐                         │
│           │   ATTENTION CONTROL   │                         │
│           │   (Priority Map)      │                         │
│           └───────────┬───────────┘                         │
│                       ▼                                      │
│           ┌───────────────────────┐                         │
│           │   ATTENTION MECHANISMS│                         │
│           │   (39 types)          │                         │
│           └───────────┬───────────┘                         │
│                       ▼                                      │
│           ┌───────────────────────┐                         │
│           │   WORKING MEMORY      │                         │
│           │   (Gated Access)      │                         │
│           └───────────────────────┘                         │
│                                                              │
└─────────────────────────────────────────────────────────────┘

Re-exports

pub use mechanisms::AttentionMechanism;

pub use mechanisms::AttentionType;

pub use mechanisms::AttentionOutput;

pub use mechanisms::ScaledDotProductAttention;

pub use mechanisms::FlashAttention;

pub use mechanisms::LinearAttention;

pub use mechanisms::SparseAttention;

pub use mechanisms::HyperbolicAttention;

pub use mechanisms::GraphAttention;

pub use mechanisms::MemoryAugmentedAttention;

pub use mechanisms::MultiHeadAttention;

pub use controller::AttentionController;

pub use controller::AttentionConfig;

pub use controller::PriorityMap;

pub use working_memory::WorkingMemory;

pub use working_memory::WMGate;

pub use working_memory::WorkingMemoryItem;

pub use salience::SalienceMap;

pub use salience::SalienceComputer;

pub use salience::SalienceFeature;

Modules

controller

Attention Controller

mechanisms

Attention Mechanisms

salience

Salience Computation

working_memory

Working Memory with Gating

Structs

AttentionState

Current state of attention system

AttentionSystem

Main attention system orchestrating all components

Enums

AttentionError

Errors that can occur in the attention module

Type Aliases

Result