Object-Oriented Dynamic Networks

• Object-Oriented Dynamic Networks are a formal framework that unifies object-oriented paradigms with dynamic, graph-based generative mechanisms to encode, modify, and infer complex knowledge.
• They employ exploiters and modifiers to generate inhomogeneous classes and transform objects, establishing an upper semilattice structure that supports systematic knowledge closure.
• OODNs feature enriched inheritance, dynamic graph traversal, and fuzzy extensions, enabling precise modeling of evolving concepts and human-like reasoning.

Object-Oriented Dynamic Networks (OODNs) are a formal knowledge representation framework that unifies object-oriented paradigms with dynamic, graph-based generative mechanisms for both objects and classes. OODNs provide a basis for encoding, manipulating, and evolving knowledge about entities, their types, and their inter-relations in a manner that supports explicit modification, set-theoretic composition, and network traversal as logical inference. The OODN model generalizes and extends the traditional OOP/class-frame view, introducing new concepts such as inhomogeneous classes, universal exploiters, closure construction, and a rich inheritance lattice, offering a rigorous base for modeling human-like reasoning and concept evolution (Terletskyi et al., 2015).

1. Formal Structure of OODNs

An OODN is defined as a 5-tuple:

$(O, C, R, E, M)$

where:

• $O$ is the set of objects, each represented by a specification $P(o)$ (properties) and signature $F(o)$ (methods).
• $C$ is the set of classes, each $c \in C$ defined by $c = (P(c), F(c))$.
• $R$ is the set of relations on $O \cup C$, including inheritance (subclassing), classification (instance-of), and aggregation.
• $E$ is the set of exploiters—external methods that generate new knowledge from existing objects/classes, without modifying their arguments.
• $O$0 is the set of modifiers—external methods that transform objects or classes, producing new entities (Terletskyi et al., 2015, Terletskyi, 2015, Terletskyi et al., 2016).

Types of Classes

• Homogeneous classes: All instances share identical $O$1 and $O$2.
• Inhomogeneous (heterogeneous) classes: Admit multiple “projections,” where the core covers common features and each projection $O$3 extends these for subgroups or types (Terletskyi et al., 2015, Terletskyi, 2015, Terletskyi, 2015).

Similarity

Two objects $O$4 are similar iff $O$5 and $O$6.

2. Graphical and Algebraic Representation

An OODN is naturally represented as a connected, directed graph $O$7 with nodes $O$8 and edges $O$9 encoding the different relation types, partitioned into:

• $P(o)$0: static (inheritance/classification) edges,
• $P(o)$1: modification edges,
• $P(o)$2: exploiter-derived (knowledge-generating) edges (Terletskyi et al., 2015).

The adjacency structure can be expressed by an adjacency matrix $P(o)$3 and subtype indicator matrices for each edge class.

3. Dynamic Knowledge Acquisition: Exploiters and Modifiers

Exploiters

Exploiters are side-effect-free operations with the capability to generate new classes or objects based on existing ones. The most universal exploiter is the union $P(o)$4, which produces inhomogeneous classes from arbitrary class subsets:

$P(o)$5

Systematic application of exploiters to a finite basic set $P(o)$6 generates a closure $P(o)$7 of all new inhomogeneous classes, the number and structure of which obey closed-form combinatorial laws:

• Number of new classes from $P(o)$8 basics: $P(o)$9,
• Each union of $F(o)$0 basics yields an inhomogeneous class with $F(o)$1 projections/types (Terletskyi, 2015).

The closure under $F(o)$2 forms an upper semilattice, with the join operation $F(o)$3 and the largest (top) element the union of all basics.

Modifiers

Modifiers effect structural change:

• Full: Alter every property and method;
• Partial: Alter a selected subset of properties/methods;
• Generating: Add new features;
• Destroying: Remove features;
• Commutable: Swap features (Terletskyi et al., 2015, Terletskyi et al., 2016).

These mechanisms enable runtime class generation and dynamic network evolution.

4. Inheritance and Class Lattice

OODNs provide an enriched inheritance mechanism, classified along three axes:

• Single vs. Multiple: Number of parent classes.
• Full vs. Partial: All or subset of features inherited.
• Strong vs. Weak: Degree to which features are inherited (supports fuzzy/partial membership, $F(o)$4).

This yields $F(o)$5 inheritance types, generalizing classical OOP. Heterogeneous classes enable grouping variants without redundancy. Algorithmic construction ensures that exceptions, redundancy, and ambiguity are alleviated—features can be omitted or weakened, core/projection separation prevents conflict, and ambiguous slots are resolved by minimum aggregation or user rule (Terletskyi, 2015).

5. Reasoning, Traversal, and Network Algorithms

Reasoning in an OODN is formulated as graph traversal, typically using depth-first search augmented with modifier/application propagation. The central algorithm iterates through the network applying inheritance or exploiter edges, and for each modifier edge, applies state transformations, building a path through dynamically created and static classes/objects. Complexity is $F(o)$6 up to modifier cost, with termination enforced by visited-marking or depth-bounding (Terletskyi et al., 2015).

6. Extensions: Fuzzy Object-Oriented Dynamic Networks

OODNs generalize to Fuzzy Object-Oriented Dynamic Networks (FOODNs) by allowing fuzzy properties, classes, and relations:

• Fuzzy attributes: Quantitative or qualitative, expressed as fuzzy sets or fuzzy verification functions.
• Fuzzy relations: Instance-of, subclassing, etc., may carry membership weights $F(o)$7.
• Lifting exploiters/modifiers: Set-theoretic operations and structural changes apply via t-norms and t-conorms (Terletskyi et al., 2016, Terletskyi et al., 2016).

FOODNs thus provide fine control over imprecision, degrees of membership, and partial inheritance. They support dynamic, timestamped evolution of both conceptual structure and specific instances, and are applicable in domains requiring reasoning about evolving, fuzzy knowledge.

7. Examples, Applications, and Significance

Classical OODN construction is illustrated by the convex polygon example:

• Classes $F(o)$8 (polygons), $F(o)$9 (rhombuses), $C$0 (squares) defined by incremental property addition,
• Objects such as $C$1, $C$2,
• Modifiers effect deletion/addition of properties (e.g., $C$3 by deleting “all-angles-90°”),
• Exploiters (e.g., union) synthesize inhomogeneous classes.

OODNs enable experience acquisition, creative concept construction, and dynamic evolution, modeling key aspects of human knowledge refinement and generalization (Terletskyi et al., 2015). Beyond KR, analogous dynamic, object-oriented graph structures underlie systems such as highly connected dynamic artificial neural networks (Alten, 2023) and largenet2 for adaptive network simulation (Zschaler et al., 2012). Fuzzy and object-oriented dynamic predictors are fundamental in contemporary causal modeling and model-based RL (Zhu et al., 2018, Yu et al., 2024).

• "Object-Oriented Dynamic Networks" (Terletskyi et al., 2015)
• "Exploiters-Based Knowledge Extraction in Object-Oriented Knowledge Representation" (Terletskyi, 2015)
• "Inheritance in Object-Oriented Knowledge Representation" (Terletskyi, 2015)
• "Fuzzy Object-Oriented Dynamic Networks. I" (Terletskyi et al., 2016)
• "Fuzzy Object-Oriented Dynamic Networks. II" (Terletskyi et al., 2016)
• "Object-Oriented Dynamics Predictor" (Zhu et al., 2018)
• "Highly connected dynamic artificial neural networks" (Alten, 2023)
• "Learning Causal Dynamics Models in Object-Oriented Environments" (Yu et al., 2024)
• "Largenet2: an object-oriented programming library for simulating large adaptive networks" (Zschaler et al., 2012)