Abstract:We propose an approach to analogical inference that marries the neuro-symbolic computational power of complex-sampled hyperdimensional computing (HDC) with Conceptual Spaces Theory (CST), a promising theory of semantic meaning. CST sketches, at an abstract level, approaches to analogical inference that go beyond the standard predicate-based structure mapping theories. But it does not describe how such an approach can be operationalized. We propose a concrete HDC-based architecture that computes several types of analogy classified by CST. We present preliminary proof-of-concept experimental results within a toy domain and describe how it can perform category-based and property-based analogical reasoning.
What problem does this paper attempt to address?
The core problem that this paper attempts to solve is **how to achieve analogical reasoning in the conceptual hyperspace**. Specifically, the author aims to combine Hyperdimensional Computing (HDC) and Conceptual Spaces Theory (CST) to propose a specific architecture for performing different types of analogical reasoning. The following are the main objectives and problems of the paper:
### 1. **Challenges of Analogical Reasoning**
Analogical reasoning is an important aspect of human cognition, involving multiple key processes: retrieval, mapping, and inference. In particular, the mapping task (i.e., identifying the relationship between A and B and applying this relationship to C to find X) is a combinatorial puzzle. Traditional connectionist and symbolic models have limitations in this regard.
### 2. **Limitations of Existing Methods**
- **Connectionist Models**: Such as ACME and DRAMA, although they can handle local concept representations, they are less efficient in handling complex analogical relationships.
- **Symbolic Models**: Such as the Structure Mapping Theory, although it can explicitly represent geometric graph structures, it encounters difficulties when computing the key conceptual properties of objects that need to be isolated.
- **Neural Networks**: Although they provide distributed representations, their concept embeddings do not have sufficient structural and combinatorial properties to perform analogical mapping.
### 3. **Proposed Solution**
The author proposes an HDC - based method, using high - dimensional vectors to represent data structures, thereby bridging the gap between symbolic and subsymbolic representations. Specifically:
- **Combining CST and HDC**: By combining the concept space of CST with the high - dimensional computing of HDC, a "conceptual hyperspace" is constructed, in which analogical reasoning can be carried out more effectively.
- **Specific Algorithms**: Specific algorithms are proposed to encode concepts, identify significant relationships, and perform mapping tasks. For example, the Parallelogram model is used to solve inter - category analogical problems.
### 4. **Experimental Verification**
The author conducted preliminary proof - of - concept experiments in the toy domain, demonstrating how to perform category - based and property - based analogical reasoning in the conceptual hyperspace.
### Summary
The core problem of the paper is **how to achieve efficient analogical reasoning in the conceptual hyperspace by combining HDC and CST**, and the effectiveness of this method has been verified through specific algorithms and experiments.