By Siyu Wu
Review Details
Reviewer has chosen not to be Anonymous
Overall Impression: Average
Content:
Technical Quality of the paper: Average
Originality of the paper: Yes
Adequacy of the bibliography: Yes
Presentation:
Adequacy of the abstract: Yes
Introduction: background and motivation: Good
Organization of the paper: Needs improvement
Level of English: Satisfactory
Overall presentation: Average
Detailed Comments:
This paper discusses an intriguing application of Vector Symbolic Architectures (VSA), also referred to as Hyperdimensional Computing (HDC) in related literature. VSA is a family of models for representing and manipulating data in a high-dimensional space, originally proposed in cognitive psychology and neuroscience as a connectionist model for symbolic reasoning.
The topic is notably promising, as it aims to advance VSA towards computational universality. However, the paper seems to overstate its contribution to the proposed implementation of VSA at a macroscopic level, due to weak empirical results and a lack of technical accuracy. Nonetheless, it provides contributions such as proposing an improved notion of belief for modal encoding within semantic vectors to implement partial knowledge, outlining basic steps of knowledge encoding such as binding and bundling, and introducing a novel data structure called associative map with the proposed macroscopic level implementation. The paper still faces significant challenges, particularly in the experimental testing phase and in some theoretical aspects, with most of its content focusing on the macroscopic implementation of VSA.
The empirical results presented are underwhelming. Although the authors argue that the main contribution of the paper is the implementation of VSA at a macroscopic level, the results primarily discuss the computation costs associated with VSA rather than practical implementation details. The preliminary experimental section is notably weak; the authors themselves admit it is far from complete. The only substantive detail is the representation of semantics in the example 'Luigi 0.5 eats thisPizza,' where 0.5 represents an introduced modality—a partial knowledge modal encoding innovation highlighted in my previous comments on paper contributions. In the 'Binding Magnitude Verification' section, there is a reference to a non-existent Table 4, which undermines the credibility of the results discussed.
In the "Knowledge Structure Encoding" section, Table 1 describes dictionaries/maps as lacking enumeration capabilities. This is confusing because, typically, maps or dictionaries do allow for the enumeration of keys, values, or key-value pairs, contrary to what the authors have described.
Figure 5, found in the "Implementation at the Macroscopic Scale" section, presents a mystery. The figure's basic declaration does not clearly convey the operational details or behaviors associated with adding, updating, or removing symbols as described in the caption. It merely shows the data structure's format without any operational logic or methods that manipulate the map's contents.
Furthermore, in the section on "Binding Canonical Representation," the author claims that applying the r1 and r2 operators recursively guarantees no residual bindings/unbindings. This statement lacks a formal proof or detailed example, which is necessary to verify its accuracy and applicability.
Lastly, I recommend that the authors check for typos in this paper, e.g., on page 9: 'guaranty' should be 'guarantee.' Although this shall not impact the final decision, it will need to be corrected.
In conclusion, while the paper sets forth a promising framework for advancing VSA in universal semantic computation, it falls short in providing empirical support and clarity in its theoretical assertions. Future work should focus on enhancing the experimental designs, clarifying theoretical explanations, and providing proofs for the claims made.
