Emergent-ARC v2.0

Emergent-ARC is a neurosymbolic architecture for solving abstract reasoning tasks (ARC-AGI). It evolves compact Spiking Neural Networks (SNNs) that compose programs from a physics-inspired Domain Specific Language (DSL).

Features

• Object-Centric Perception: Reduces grid to object slots.
• Hierarchical DSL: Physics, relational, and generative primitives.
• Compact SNN Policy: ~12k parameters, efficient evolution.
• Online Program Induction: Test-time hypothesis validation.
• Veteran Population: Continuous knowledge transfer.

Installation

pip install -r requirements.txt

Usage

python main.py

Structure

• emergent_arc/: Core package
  • detection/: Object detection and feature extraction
  • dsl/: Domain Specific Language primitives and executor
  • network/: Spiking Neural Network architecture
  • evolution/: Evolutionary algorithms (PGPE)
  • memory/: Subroutine library and veteran pool
  • inference/: Online program induction
• tests/: Unit tests
• notebooks/: Analysis and exploration