Pith. sign in

REVIEW 4 major objections 5 minor 38 references

Johnny: Structuring Representation Space to Enhance Machine Abstract Reasoning Ability

T0 review · 4 major / 5 minor · reviewed 2026-08-15 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read Johnny's claim: a learned representation space of wrong-answer stand-ins lifts RPM accuracy to 99.4% on RAVEN, 99.6% on I-RAVEN, and 99.0% on PGM.

desk verdict A well-specified architecture with plausible small benchmark gains, but the central sub-enumeration mechanism is self-referential and the PGM metadata experiments look contaminated. read the letter →

arxiv 2506.01970 v1 pith:UIAJPLMQ submitted 2025-05-13 cs.LG cs.CV

Ruizhuo Song , Beiming Yuan This is my paper · ORCID
classification cs.LGcs.CV
keywords abstractreasoningRaven'sProgressiveMatricesrepresentationspacesub-enumerationtokenizationTransformerinter-headcommunicationvisual
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

Johnny is a solver for Raven's Progressive Matrices (RPM), a family of visual multiple-choice puzzles. The paper argues that end-to-end models that pick from an option pool are limited by the wrong answers they are shown: training fits a decision boundary from the observed incorrect options, so an incomplete pool leaves that boundary loose and caps accuracy. To remove the cap, Johnny tokenizes image representations and aligns them to a learned, discrete 'representation space'; an extra loss scores every vector in that space and treats the vector closest to the correct answer's token as a stand-in for missing wrong answers. This 'sub-enumeration' of negative configurations is reported to push accuracy to 99.4% on RAVEN, 99.6% on I-RAVEN, and 99.0% on PGM, above the strongest published baselines, while a second contribution—the Spin-Transformer—adds explicit inter-head communication to capture positional relations, with a lightweight variant that keeps most of the gain. A reader should care because the claim is that abstract reasoning improves by structuring the latent space rather than by scaling the network.

What carries the argument

The load-bearing object is the learned representation space, a finite set of $K=256$ optimizable vectors in the same dimension as one image token. Johnny's extraction module maps each of the 16 images of an RPM instance into token vectors $\{z_{ij}\}$; losses $\ell_1$, $\ell_2$, and $\ell_3$ align each token to its nearest codebook vector, update the codebook, and reconstruct the original image from the selected vectors through a decoder $D$, making the space a complete coding of observable images. The sub-enumeration term $\ell_4$ then computes a reasoning score for every codebook vector and applies a softmax cross-entropy whose positive target is the vector $\tilde{k}=\arg\min_k \|z_{\alpha j}-T_k\|_2^2$ nearest the correct option's token; this is the mechanism that manufactures synthetic negative options. The Spin-Transformer is secondary machinery: its Spin Block computes pose vectors $P_{jkl}=H_{jk}W_{jkl}$, sums them with a squash nonlinearity, and adds them to output tokens, giving attention heads a route to communicate local position information; the Straw variant shares the mapping matrices across token positions and folds the weighted sum into a masked cross-attention over a learned anchor vector.

What would settle it

Run Johnny ACT3 identically but replace the target of $\ell_4$ with a randomly chosen codebook vector, or with the nearest vector from a frozen, randomly initialized codebook, and compare per-subset accuracy on RAVEN D-9/OIG and PGM Neutral against ACT2: if the gain persists, $\ell_4$ is a regularizer rather than sub-enumeration; if it vanishes, check whether the learned codebook vectors individually correspond to coherent human-annotated attributes, because without such alignment the codebook cannot be supplying the missing negative configurations.

Watch

Extended reading notes

Core claim

The paper's central claim has two stages. First, training an end-to-end RPM solver is equivalent to fitting a probability distribution: the mean is set by correct options and the variance by incorrect options, so the model converges its decision boundary only from the negatives it has seen. Second, Johnny replaces exhaustive enumeration of all possible wrong answers with a bounded discrete codebook of $K=256$ learnable vectors $\{T_k\}_{k=1}^{K}$, aligned with every image token through three losses (alignment, codebook update, and reconstruction via a decoder). The sub-enumeration loss $\ell_4$ makes the reasoning module score every codebook vector and drives the probability mass, at temperature $\tau=0.01$, onto the vector closest to the correct option's token. In the reported experiments, the CE-only baseline achieves 98.6/99.0 on RAVEN/I-RAVEN, adding the representation space reaches 98.8/99.2, and adding $\ell_4$ reaches 99.4/99.6; on PGM the comparable sequence is 97.9, 98.2, 99.0.

Load-bearing premise

The method depends on the learned set of 256 representative vectors genuinely separating the concepts in the puzzle images, so that the vector closest to the correct answer is a meaningful stand-in for a missing wrong answer; if the vectors have not captured the concepts, the extra training term only reinforces the model's own habits and the reported gains are regularization, not reasoning.

Editorial extensions

If this is right

  • With the representation space and sub-enumeration loss, RAVEN accuracy rises from 98.6% to 99.4% and I-RAVEN from 99.0% to 99.6%, with the largest subset gains on D-9 and OIG, which require positional rules.
  • On PGM, the same additions lift overall accuracy from 97.9% to 99.0% and interpolation generalization from 81.0% to 87.3%, while extrapolation remains near 18-19%.
  • Swapping the standard Transformer encoder for the Spin-Transformer in two strong published solvers, under compute-matched settings, improves their accuracy on 3×3 Grid and OIG subsets; the lightweight Straw variant retains most of the gain with a reduction in trainable mapping matrices from $O(N)$ to $O(1)$.
  • Because the decoder can reconstruct images from selected codebook components, the trained Johnny can in principle generate a candidate answer image rather than choose one, which the paper proposes for future generative RPM solving.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • If the mechanism is what carries the gain, the nearest-codebook-target loss should transfer to other multiple-choice reasoning formats with sparse negatives, such as diagrammatic reasoning or attribute-based visual question answering; the paper only tests RPM-style benchmarks.
  • The paper fixes $K=256$; the sub-enumeration story predicts a non-monotonic dependence on $K$, since too small a codebook cannot cover the concepts and too large a one makes the nearest-neighbor target nearly arbitrary, so sweeping $K$ would separate the mechanism from generic regularization.
  • The Gaussian-boundary argument implies that end-to-end solvers' accuracy should respond systematically to the information content of the option pool; varying the number or similarity of wrong answers across existing RPM datasets would test this prediction, which the paper does not run.
  • The future-work section's generative extension depends on an external judge for the synthesized answers, and the paper itself notes that current symbolic solvers are not accurate enough to serve as that judge.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 5 minor

Summary. The paper argues that end-to-end RPM-solving models are limited by their reliance on the specific incorrect options present in the training option pool, and proposes a learned representation space to 'sub-enumerate' possible incorrect configurations. It introduces the Johnny architecture, which separates a tokenized representation extractor (ViT backbone) from a reasoning module that scores individual tokens and, in the ACT3 configuration, also scores components of a learned codebook under a new loss l4 (Eq. 8). The paper further proposes Spin-Transformer and a lightweight Straw Spin-Transformer variant that modify inter-head communication in the transformer encoder. Experiments are reported on RAVEN, I-RAVEN, and PGM, with claims of state-of-the-art accuracy (e.g., 99.4% on RAVEN, 99.6% on I-RAVEN, 99.0% on PGM), and an additional metadata-conditional experiment on PGM generalization sub-problems is presented in Table V.

Significance. If the reported results and the proposed mechanism are substantiated, the paper would make three contributions: a diagnosis of end-to-end RPM solvers' dependence on option-pool configurations with a concrete remedy; an architecture that couples tokenized representation learning with a reasoning module that scores both options and representation-space components; and an attention architecture variant with a lightweight version that improves positional-relational reasoning. The ACT1-ACT4 decomposition is a useful experimental design because it isolates the representation-space losses from the backbone change, and the paper evaluates on three standard benchmarks against several strong baselines. The claims are falsifiable, and the loss equations are stated in sufficient detail to reimplement the method. However, the central ACT3 mechanism currently lacks direct evidence: the l4 target is self-referential, and the reported gains are small on saturated metrics without repeated-run statistics or code. The contribution is therefore conditional until the mechanism analysis and reproducibility are supplied.

major comments (4)
  1. [§VI-D, Eq. (8)] The positive target in the sub-enumeration loss is \tilde{k} = argmin_k ||z_{\alpha j} - T_k||^2_2, where z_{\alpha j} is produced by the same encoder being trained and T_k are the same codebook vectors optimized by l1-l3 (Eqs. (2)-(5)). Final accuracy is measured against external labels, so the overall result is not circular, but the attribution of ACT3 gains to 'sub-enumeration of incorrect options' is not supported as stated. The paper reports no codebook usage statistics, no active-component counts, no analysis of which components are selected by \tilde{k}, and no ablation replacing the l4 target with a non-informative one (e.g., random components or a fixed target). Because Eqs. (2)-(4) are VQ-style losses without a commitment or usage penalty, codebook collapse is a concrete risk; if only a few components are active, \tilde{k} is effectively arbitrary and l4 can act as a generic regularizer. The ACT2-to-ACT3 gains (Table I: RAVEN 98.8 to 99.4; Table III: PGM 98.2 to 99.0) are then equally compatible with a regularization effect. Please provide codebook-usage analyses and target ablations, or revise the mechanistic claim.
  2. [§VIII, Tables I-III] All experiments are reported as single accuracies without seeds, standard deviations, or confidence intervals, and the experimental setup is delegated to self-citations [29], [32] ("the same settings and equipment"). On RAVEN/I-RAVEN many entries are at 99-100%, so a single-run difference of 0.1-0.5 percentage points (e.g., Johnny ACT3 99.4/99.6 vs Triple-CFN 98.9/99.1 in Table I) is within plausible run-to-run variability. Without repeated runs, the state-of-the-art claims and the cross-configuration comparisons ACT1-4 cannot be verified. Please report mean and standard deviation over multiple seeds, and release code or detailed hyperparameters, including the 'sliding window' schedule mentioned in §VI-C and the exact training schedule for adding l4.
  3. [§VIII-C, Eq. (33), Table V] The metadata experiments require clarification of what "enumerate all possible manifestations of metadata" means. If the representation space {Y_\beta | \beta in [1,L]} is sized to include metadata manifestations that appear only in held-out generalization splits, then construction of the space uses knowledge of the test distribution, and the high accuracies in Table V (92.2% on Interpolation, 98.0% on Held-out Pairs of Triples and Attribute Pairs) may reflect this prior knowledge rather than the method's generalization. Please specify whether L is chosen from training metadata only, whether any test metadata or test instances are used in building or training the space, and report repeated-run statistics for Table V. The candid limitation paragraph at the end of §VIII.C is appreciated, but it does not resolve the test-distribution question.
  4. [§VII.B, Tables II] The lightweight claim for Straw Spin-Transformer is asserted but never quantified. Table II compares accuracy but gives no parameter counts, FLOPs, or training/inference time for Spin-Transformer versus Straw Spin-Transformer; the only justification is the architectural reduction in the number of mapping matrices (Eqs. (14)-(25)). Please add quantitative complexity measurements to support the 'lightweight' label and to justify the 'computational parity' argument used in the replacement experiments (half the number of layers).
minor comments (5)
  1. [Abstract and §II] PGM is introduced with citation [16], but the correct reference is [17] (Barrett et al.).
  2. [Figure 8 and §VI.B] 'Patttern Extractor' should be 'Pattern Extractor'; similarly, 'fallows' in §VII.A should be 'follows'.
  3. [Table IV] The table title says 'Valen' instead of 'PGM', and the header layout 'Model/Task' is confusing; please clarify which column corresponds to the task.
  4. [Eq. (14)] The placeholder '' with the caption 'on-demand broadcasting' is undefined; please replace it with explicit indexing notation.
  5. [§VII.A] The statement that 'when N=1, the Spin-Transformer collapses into a regular Transformer-Encoder' is not justified by Eqs. (10)-(13) unless the pose matrices are set to zero; please explain the collapse explicitly.

Circularity Check

1 steps flagged · score 5.0 of 10

The sub-enumeration loss (Eq. 8) uses a self-referential nearest-codebook target; final accuracy is externally labeled, so the circularity is partial and mechanism-level.

  1. self definitional [Section VI-D, Eq. (8); compare Eqs. (2)-(3)]
    "ℓ4 = − Σ_{j=1}^N log ( e^{score_{k~j}/τ} / Σ_{k=1}^K e^{score_{kj}/τ} ), where k~ = argmin_{k∈[1,K]} ||z_{αj} − T_k||_2^2 ... with the optimization goal that the component T_{k~} most similar to the correct reference token z_{αj} (associated with the ground-truth option x_α) should receive the highest probability mass."

    By construction, the positive component k~ in Eq. (8) is selected from the same learnable codebook {T_k} and the same encoder outputs z_{αj} that Eqs. (2)-(3) optimize: l1/l2 pull T_k toward z_{ij}, and l4 then rewards the Reasoning Module for massing probability on whichever T_k the model has itself made nearest to the correct option token. The 'incorrect' components suppressed by the softmax are therefore not independently enumerated negative configurations; they are only the components that the model's own geometry did not select. The loss enforces self-consistency between the encoder, codebook, and reasoner, and can be minimized by any regularizing alignment between these modules.

full rationale

The paper does not derive its benchmark numbers from its own model: Johnny ACT1-ACT4 are compared on RAVEN, I-RAVEN, and PGM against external baselines and against its own ablated configurations, so the reported accuracies are not circularly forced. The Spin-Transformer contribution is also tested by swapping it into independently published RS-TRAN and Triple-CFN models, which is external evidence. The central concern is confined to the sub-enumeration mechanism: Eq. (8)'s target is the model's own nearest codebook vector, so the 'supplementing negative option configurations' explanation is self-referential and could be a generic regularization effect. The paper itself concedes related enumerability limitations for the metadata-based PGM generalization loss, which is an explicit limitation rather than hidden circularity. No load-bearing self-citation chain or renamed known result was found.

Assumptions & free parameters 4 free parameters · 5 assumptions · 0 invented entities

The central method depends on several hand-picked hyperparameters (K, lambda, tau) and on assumptions about Gaussian decision boundaries, finite tokenizable representations, independent token evaluation, and metadata completeness. These are not derived from first principles or validated by ablations, and the metadata assumption is a potential source of label leakage.

free parameters (4)
  • Representation space size K = 256
    Number of codebook components, chosen by hand; no sensitivity analysis reported.
  • Loss weight lambda for l2 = 0.25
    Weight for the codebook commitment loss l2, set by hand.
  • Temperature tau for l4 = 0.01
    Temperature used in the softmax of the sub-enumeration loss l4; no tuning study reported.
  • Initial temperature for l5 = 1e-6
    Optimizable temperature in the metadata alignment loss l5, initialized to 1e-6.
assumptions (5)
  • ad hoc to paper The end-to-end RPM solver's predictive distribution is Gaussian.
    Section IV.C, Eq. 1: the entire critique of end-to-end models is built on this assumed distribution; no empirical or theoretical justification is given.
  • domain assumption All observable RPM images can be represented by a finite discrete codebook of K components.
    Section VI.C: necessary for the representation space; continuous attributes such as precise positions may not be faithfully codified with 256 vectors.
  • domain assumption The reasoning module must evaluate each representation token independently.
    Section VI requirement 3: this restricts architecture and may discard useful inter-token dependencies that could aid reasoning.
  • ad hoc to paper Metadata enumeration can cover all possible manifestations, including those in held-out generalization tasks.
    Section VIII.C, Eq. 33: used to build the representation space for PGM generalization; if it includes held-out attribute values, test labels are accessed during training.
  • domain assumption Transformer self-attention limits inter-head communication, so explicit pose matrices are required.
    Section VII.A: stated as an observation without formal analysis or controlled ablations.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Johnny: Structuring Representation Space to Enhance Machine Abstract Reasoning Ability." pith.science (2026). https://pith.science/paper/UIAJPLMQ

@misc{pith2026250601970,
  author       = {Pith},
  title        = {Pith review of: Johnny: Structuring Representation Space to Enhance Machine Abstract Reasoning Ability},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/UIAJPLMQ}},
  note         = {Machine review of arXiv:2506.01970}
}
read the original abstract

This paper thoroughly investigates the challenges of enhancing AI's abstract reasoning capabilities, with a particular focus on Raven's Progressive Matrices (RPM) tasks involving complex human-like concepts. Firstly, it dissects the empirical reality that traditional end-to-end RPM-solving models heavily rely on option pool configurations, highlighting that this dependency constrains the model's reasoning capabilities. To address this limitation, the paper proposes the Johnny architecture - a novel representation space-based framework for RPM-solving. Through the synergistic operation of its Representation Extraction Module and Reasoning Module, Johnny significantly enhances reasoning performance by supplementing primitive negative option configurations with a learned representation space. Furthermore, to strengthen the model's capacity for capturing positional relationships among local features, the paper introduces the Spin-Transformer network architecture, accompanied by a lightweight Straw Spin-Transformer variant that reduces computational overhead through parameter sharing and attention mechanism optimization. Experimental evaluations demonstrate that both Johnny and Spin-Transformer achieve superior performance on RPM tasks, offering innovative methodologies for advancing AI's abstract reasoning capabilities.

Figures

Figures reproduced from arXiv: 2506.01970 by the authors.

Figure 2
Figure 2. Annotations of images within a RPM instance [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 1
Figure 1. RAVEN and PGM case B. Related work Image reasoning models such as CoPINet [20], LEN+teacher [19], and DCNet [21] focus on learning disparities and rules. Meanwhile, NCD [23], SCL [24], SAVIR-T [25], and neural symbolism systems (PrAE, NVSA, ALANS [26]–[28]) enhance both interpretability and accuracy. RS-CNN and RS-TRAN [29] excel in solving RPM problems, while Triple-CFN [32] stands out by implicitly extracting and … view at source ↗
Figure 3
Figure 3. The diagram of the structure of end-to-end RPM [PITH_FULL_IMAGE:figures/full_fig_p002_3.png] view at source ↗
Figures from the paper (10 more)
Figure 4
Figure 4. Figure 4: RPM-solving model training process augmented with [PITH_FULL_IMAGE:figures/full_fig_p004_4.png]
Figure 6
Figure 6. Figure 6: The process by which Johnny’s representation extractor [PITH_FULL_IMAGE:figures/full_fig_p004_6.png]
Figure 9
Figure 9. Figure 9: The architecture of the Pattern Extractor. [PITH_FULL_IMAGE:figures/full_fig_p005_9.png]
Figure 10
Figure 10. Figure 10: The architecture of the Representation Space. [PITH_FULL_IMAGE:figures/full_fig_p006_10.png]
Figure 11
Figure 11. Figure 11: The feedforward process of this block shown in [PITH_FULL_IMAGE:figures/full_fig_p007_11.png]
Figure 11
Figure 11. Figure 11: The feedforward process of a Spin block [1, N], k ∈ [1, M], l ∈ [1, N], Wjkl ∈ R(d/M)×d}, where the outer indices (j, k) specifically identify their corresponding local vector Hjk, while the inner index l ∈ [1, N] traverses all input token positions. 3) After summatio…
Figure 12
Figure 12. Figure 12: The architecture of the Spin-Transformer [PITH_FULL_IMAGE:figures/full_fig_p008_12.png]
Figure 13
Figure 13. Figure 13: The feedforward process of a Straw Spin Block [PITH_FULL_IMAGE:figures/full_fig_p009_13.png]
Figure 14
Figure 14. Figure 14: Masked attention 4) Due to the combined effects of the multi-head self￾attention and masking operations, the vector z˜0 ag￾gregates contextual information from the logical in￾puts {z˜j |j ∈ [1, N]}, and is ultimately transformed into the global vector z0. Within this …
Figure 15
Figure 15. Figure 15: The architecture of the Straw Spin-Transformer [PITH_FULL_IMAGE:figures/full_fig_p010_15.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

38 extracted references · 29 canonical work pages

  1. [29]

    Q. Wei, D. Chen, B. Yuan, Multi-viewpoint and multi-evaluation with felicitous inductive bias boost machine abstract reasoning ability, arXiv :2210 .14914, 2022

    work page 2022

  2. [32]

    Triple-CFN: Separating Concepts and Features Enhances Machine Abstract Reasoning Ability

    R.Song, B.Yuan. Triple-CFN: Restructuring Concept Spaces for Enhancing Abstract Reasoning process. Preprint at https://arxiv.org/abs/arXiv:2403.03190 (2024)

    work page Pith review arXiv 2024

  3. [1]

    Communications of the ACM, 63(11), 139-144 (2020)

    Goodfellow, I.et al.Generative adversarial networks. Communications of the ACM, 63(11), 139-144 (2020)

    work page 2020

  4. [2]

    P., & Welling, M

    Kingma, D. P., & Welling, M. Auto-encoding variational bayes. Preprint at https://arxiv.org/abs/1312.6114 (2014)

    arXiv 2014

  5. [3]

    Deep Residual Learning for Image Recognition

    He, K., Zhang, X., Ren, S., & Sun, J. Deep Residual Learning for Image Recognition. In IEEE Conference on Computer Vision and Pattern Recognition, 770-778 (2016)

    work page 2016

  6. [4]

    Li, Z., Liu, F., Yang, W., Peng, S., & Zhou, J. (2021). A survey of convolutional neural networks: analysis, applications, and prospects. IEEE transactions on neural networks and learning systems, 33(12), 6999-7019

    work page 2021

  7. [5]

    ”A survey of visual transformers.” IEEE Transactions on Neural Networks and Learning Systems (2023)

    Liu, Yang, et al. ”A survey of visual transformers.” IEEE Transactions on Neural Networks and Learning Systems (2023)

    work page 2023

  8. [6]

    In Advances in Neural Information Processing Systems, (2017)

    Vaswani, A.et al.Attention is All You Need. In Advances in Neural Information Processing Systems, (2017)

    work page 2017

Show all 38 references
  1. [7]

    In Advances in Neural Information Processing Systems, 1877-1901 (2020)

    Brown, T.et al.Language Models are Few-shot Learners. In Advances in Neural Information Processing Systems, 1877-1901 (2020)

    2020

  2. [8]

    W., Medina, J

    Otter, D. W., Medina, J. R., & Kalita, J. K. (2020). A survey of the usages of deep learning for natural language processing. IEEE transactions on neural networks and learning systems, 32(2), 604-624

    2020

  3. [9]

    Denoising diffusion probabilistic models

    Ho, J., Jain, A., & Abbeel, P. Denoising diffusion probabilistic models. In Advances in Neural Information Processing Systems, 33, 6840-6851 (2020)

    2020

  4. [10]

    L., & Parikh, D

    Antol, S., Agrawal, A., Lu, J., Mitchell, M., Batra, D., Zitnick, C. L., & Parikh, D. VQA: Visual question answering. In IEEE International Conference on Computer Vision, 2425-2433 (2015)

    2015

  5. [11]

    Girshick

    Johnson, J., Hariharan, B., Van Der Maaten, L., Fei-Fei, L., Lawrence Zitnick, C., & Girshick, R. Girshick. CLEVR: A Diagnostic Dataset for Compositional Language and Elementary Visual Reasoning. In IEEE Conference on Computer Vision and Pattern Recognition, 2901-2910 (2017)

    2017

  6. [12]

    H. Hu, S. Storks, R. L. Lewis, and J. Chai, ”In-context Analog- ical Reasoning with Pre-trained Language Models,” arXiv Preprint arXiv:2305.17626, 2023

    2023 arXiv

  7. [13]

    B. Li, Y . Zhang, D. Guo, R. Zhang, F. Li, H. Zhang, K. Zhang, P. Zhang, Y . Li, Z. Liu, and C. Li, ”LLaV A-OneVision: Easy Visual Task Transfer,” arXiv Preprint arXiv:2408.03326, 2024

    2024 arXiv

  8. [14]

    Zheng, J

    T. Zheng, J. Cheng, C. Li, H. Shi, Z. Wang, J. Bai, Y . Song, G. Y . Wong, and S. See, ”LogiDynamics: Unraveling the Dynamics of Logical Inference in Large Language Model Reasoning,” arXiv Preprint arXiv:2502.11176, 2025

    2025

  9. [15]

    Raven J. C. Raven’s Progressive Matrices. (Western Psychological Services, (1938)

    1938

  10. [16]

    Zhang, C., Gao, F., Jia, B., Zhu, Y ., & Zhu, S. C. Raven: A Dataset for Relational and Analogical Visual Reasoning. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 5317–5327 (2019)

    2019

  11. [17]

    Measuring Abstract Reasoning in Neural Networks

    Barrett, D., Hill, F., Santoro, A., Morcos, A., & Lillicrap, T. Measuring Abstract Reasoning in Neural Networks. In International Conference on Machine Learning, 511-520 (2018)

    2018

  12. [18]

    Preprint at https://arxiv.org/abs/2010.11929 (2020)

    Dosovitskiy, A.et al.An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale. Preprint at https://arxiv.org/abs/2010.11929 (2020)

    2020 arXiv

  13. [19]

    J., & Wei, W

    Zheng, K., Zha, Z. J., & Wei, W. Abstract Reasoning with Distracting Features. In Advances in Neural Information Processing Systems, (2019)

    2019

  14. [20]

    Zhang, C., Jia, B., Gao, F., Zhu, Y ., Lu, H., & Zhu, S. C. Learning Perceptual Inference by Contrasting. In Proceedings of Advances in Neural Information Processing Systems, (2019)

    2019

  15. [21]

    Effective Abstract Reasoning with Dual- Contrast Network

    Zhuo, T., & Kankanhalli, M. Effective Abstract Reasoning with Dual- Contrast Network. In Proceedings of International Conference on Learn- ing Representations, (2020)

    2020

  16. [22]

    Scale-Localized Abstract Reasoning

    Benny, Y ., Pekar, N., & Wolf, L. Scale-Localized Abstract Reasoning. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 12557-12565, (2021)

    2021

  17. [23]

    Unsupervised abstract reasoning for raven’s problem matrices

    Zhuo, Tao and Huang, Qiang & Kankanhalli, Mohan. Unsupervised abstract reasoning for raven’s problem matrices. IEEE Transactions on Image Processing, 8332–8341, (2021)

    2021

  18. [24]

    The Scattering Compositional Learner: Discovering Objects, Attributes, Relationships in Analogical Reasoning

    Wu, Y ., Dong, H., Grosse, R., & Ba, J. The Scattering Compositional Learner: Discovering Objects, Attributes, Relationships in Analogical Reasoning. Preprint at https://arxiv.org/abs/2007.04212 (2020)

    2020 arXiv

  19. [25]

    SA ViR-T: Spatially Attentive Visual Reasoning with Transformers

    Sahu, P., Basioti, K., & Pavlovic, V . SA ViR-T: Spatially Attentive Visual Reasoning with Transformers. Preprint at https://arxiv.org/abs/2206.09265 (2022)

    2022 arXiv

  20. [26]

    C., & Zhu, Y

    Zhang, C., Jia, B., Zhu, S. C., & Zhu, Y . Abstract Spatial-Temporal Reasoning via Probabilistic Abduction and Execution. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 9736-9746 (2021)

    2021

  21. [27]

    N., Zhu, S

    Zhang, C., Xie, S., Jia, B., Wu, Y . N., Zhu, S. C., & Zhu, Y . Learning Algebraic Representation for Systematic Generalization. In Proceedings of the European Conference on Computer Vision, (2022)

    2022

  22. [28]

    A Neuro-vector-symbolic Architecture for Solving Raven’s Progressive Matrices

    Hersche, M., Zeqiri, M., Benini, L., Sebastian, A., & Rahimi, A. A Neuro-vector-symbolic Architecture for Solving Raven’s Progressive Matrices. Preprint at https://arxiv.org/abs/2203.04571 (2022)

    2022 arXiv

  23. [30]

    ”Abstracting Concept-Changing Rules for Solving Raven’s Progressive Matrix Problems.” arxiv preprint arxiv:2307.07734 (2023)

    Shi, Fan, Bin Li, and Xangyang Xue. ”Abstracting Concept-Changing Rules for Solving Raven’s Progressive Matrix Problems.” arxiv preprint arxiv:2307.07734 (2023)

    2023 arXiv

  24. [31]

    Solving the bongard-logo problem by modeling a probabilistic model

    R.Song, B.Yuan. Solving the bongard-logo problem by modeling a probabilistic model. Preprint at https://arxiv.org/abs/ arXiv:2403.03173 (2024)

    2024 arXiv

  25. [33]

    Stratified Rule-Aware Network for Abstract Visual Reasoning

    Hu, S., Ma, Y ., Liu, X., Wei, Y ., & Bai, S. Stratified Rule-Aware Network for Abstract Visual Reasoning. In Proceedings of the AAAI Conference on Artificial Intelligence, 1567-1574 (2021). JOURNAL OF LATEX CLASS FILES, VOL. 14, NO. 8, AUGUST 2021 15

    2021

  26. [34]

    A., Just, M

    Carpenter, P. A., Just, M. A., & Shell, P. What One Intelligence Test Measures: a Theoretical Account of the Processing in the Raven Progressive Matrices Test. Psychological review, 97(3), 404, (1990)

    1990

  27. [35]

    Oord, A. V . D., Li, Y ., & Vinyals, O. Representation Learning with Con- trastive Predictive Coding. Preprint at https://arxiv.org/abs/1807.03748 (2019)

    2019 arXiv

  28. [36]

    Sabour, Sara, Nicholas Frosst, and Geoffrey E. Hinton. ”Dynamic routing between capsules.” Advances in neural information processing systems 30 (2017)

    2017

  29. [37]

    Galassi, A., Lippi, M., & Torroni, P. (2020). Attention in natural language processing. IEEE transactions on neural networks and learning systems, 32(10), 4291-4308

    2020

  30. [38]

    In NIPS Autodiff Workshop, (2017)

    Paszke, A.et al.Automatic Differentiation in Pytorch. In NIPS Autodiff Workshop, (2017)

    2017

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.