Global Workspace Competition + Broadcast + Working Memory with Ignition Dynamics

What problem does it solve?

Orchestrates multi-modal encoder outputs into a fixed-capacity global workspace using ignition dynamics to gate broadcasts and memory updates, enabling stable, cross-modal integration across timesteps.

Core Features & Use Cases

• Deterministic, fixed-modality token competition with explicit 4-term scoring (content, salience, novelty, task) and top-K gating.
• Iterative rounds with ignition dynamics that decide when to commit broadcasts and persist workspace content in memory.
• Slot-based workspace outputs of shape (B, K, D) with broadcast adapters to temporal, symbolic, and decision modules.
• Extensible architecture supporting modular encoders, working memory backends, and telemetry-friendly logging.
• Use cases include robust multi-modal reasoning, conscious-like broadcasting across cognitive modules, and persistent context across time.

Quick Start

Instantiate the workspace with sample modalities and run a single forward pass to observe ignition-gated broadcast.