Neural-Symbolic Units (NSU)

Overview

A Neural-Symbolic Unit (NSU) is the atomic entity within the Aevum Encyclopedia knowledge graph. Unlike traditional knowledge bases that rely solely on structured triples or purely on vector embeddings, NSUs embody a hybrid paradigm. Each NSU encapsulates both a neural component for semantic similarity and fuzzy matching, and a symbolic component for precise logic, rules, and verifiable facts.

This dual representation enables Aevum to perform complex reasoning tasks that require both the flexibility of deep learning and the rigor of symbolic AI, such as cross-disciplinary inference, contradiction detection, and multi-hop question answering.

💡 Key Concept

Think of an NSU as a bridge: the neural side understands how concepts relate intuitively, while the symbolic side defines how concepts relate formally. Together, they provide a robust, interpretable, and scalable unit of knowledge.

Architecture

The architecture of an NSU consists of three primary layers:

Neural Layer: A high-dimensional embedding vector trained on Aevum's corpus, capturing semantic nuance, context, and topical proximity.
Symbolic Layer: A structured data object containing identifiers, ontological classifications, logical rules, and verified assertions.
Fusion Layer: A dynamic interface that aligns neural and symbolic representations, enabling joint inference and consistency checks.

NSU Structural Model

🧠 Neural Embedding

⟷

🔗 Fusion Layer

⟷

⚖️ Symbolic Logic

Data Structure

Below is a simplified JSON representation of an NSU object as exposed via the Aevum API. Each unit contains unique metadata, neural parameters, and symbolic definitions.

NSU Schema Example JSON

{
  "nsu_id": "AE-NSU-89201",
  "concept": "Quantum Entanglement",
  "version": 3.1,
  "created_at": "2023-04-12T08:30:00Z",
  "neural": {
    "embedding_dim": 768,
    "cluster": "physics_quantum_mechanics",
    "similarity_threshold": 0.85,
    "context_vectors": [
      "superposition", "non-locality", "einstein_podolsky_rossen"
    ]
  },
  "symbolic": {
    "ontology_class": "PhysicalPhenomenon",
    "relations": {
      "is_subclass_of": "QuantumEffect",
      "interacts_with": [ "QuantumComputing", "Cryptography" ],
      "requires": "CompositeQuantumSystem"
    },
    "rules": [
      "IF distance > 0 AND particles are entangled THEN correlation persists instantaneously"
    ],
    "verified_sources": [
      "PhysRevLett.97.140401", "Nature.534.370"
    ]
  }
}

Capabilities

NSUs empower Aevum with advanced capabilities that transcend traditional search and retrieval:

Capability	Description	Layer Used
Semantic Retrieval	Finds relevant content based on meaning, not just keywords.	Neural
Logical Reasoning	Applies rules and ontologies to derive new facts.	Symbolic
Contradiction Detection	Identifies inconsistencies between neural context and symbolic rules.	Fusion
Multi-Hop Inference	Connects disparate concepts through chains of relations.	Fusion
Explainability	Provides traceable, rule-based explanations for AI outputs.	Symbolic

Inference Engine

The Neural-Symbolic Inference Engine orchestrates NSUs to perform complex queries. When a user asks a question, the engine:

Retrieves: Uses the neural layer to fetch candidate NSUs based on semantic similarity.
Filters: Applies symbolic constraints and ontology rules to narrow results.
Ranks: Scores results using a weighted combination of vector proximity and logical relevance.
Synthesizes: Constructs a coherent answer by traversing the NSU graph and validating against symbolic rules.

This process ensures high accuracy while maintaining the flexibility to handle ambiguous or novel queries.

⚠️ Developer Note

When integrating NSUs via the API, always specify the inference_mode parameter. Use "hybrid" for optimal results, or "symbolic_only" when strict rule compliance is required.

Integration

Developers can query NSUs directly through the Aevum REST API or use the Python SDK for advanced graph traversal. NSUs are compatible with standard RDF serialization for interoperability with external knowledge graphs.

Python SDK Query Python

import aevum

# Initialize client
client = aevum.Client(api_key="YOUR_API_KEY")

# Query NSUs with hybrid inference
results = client.nsu.query(
    concept="Quantum Computing",
    inference_mode="hybrid",
    depth=3,
    include_rules=True
)

for unit in results:
    print(unit.concept, unit.symbolic.relations)