Self-similarity Analysis in Deep Neural Networks

Chengwen Qi; Hongfei Wang; Jian Gao; Jianshe Wu; Jingyi Ding; Licheng Jiao; Yuwei Guo

REVIEW 4 major objections 5 minor 37 references

Self-similarity Analysis in Deep Neural Networks

T0 review · 4 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read This paper claims that treating hidden-layer features as a complex network and regularizing their self-similarity improves classification accuracy by up to 6 percentage points, but only in MLP and attention architectures.

desk verdict Real accuracy gains from a self-similarity regularizer, but the metric is an unvalidated histogram statistic, so the paper's central claim is unsupported. read the letter →

arxiv 2507.17785 v1 pith:66UEWBF6 submitted 2025-07-23 cs.LG

Jingyi Ding , Chengwen Qi , Hongfei Wang , Jianshe Wu , Licheng Jiao , Yuwei Guo , Jian Gao This is my paper

classification cs.LG

keywords self-similaritycomplexnetworksfeaturehidden-layerfeaturesdifferentiableregularizationvisiontransformersMLP-MixerSSrate

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

This paper argues that the internal feature geometry of a deep network can be viewed as a complex network, and that its degree of self-similarity is a trainable quantity. The authors build a graph from hidden-layer activation vectors, define a differentiable self-similarity rate (SS rate), and use it as a regularizer during training. They report that this constraint improves accuracy on CIFAR-10, CIFAR-100, and Imagenette for MLP and attention architectures, by up to 6 percentage points, while leaving convolutional networks essentially unchanged. The reason, they argue, is that networks with global information aggregation genuinely possess self-similar feature structure, so enforcing that structure helps rather than hurts.

What carries the argument

The central object is $G_M$, a graph whose $D$ nodes are hidden-layer feature maps after dimension reduction, with edges defined by Euclidean distance below a threshold. Self-similarity is quantified by the SS rate, which measures deviation of the log of a box-count proxy from the linear law expected for a fractal network, with values near 0 indicating strong self-similarity and values near 1 indicating weak self-similarity. The key enabling device is the differentiable proxy $N_\theta = 1 + (D-1)\log_D(D + (1-D)p(C \le \theta))$, where $p(C \le \theta)$ is the fraction of normalized feature distances below threshold $\theta$; because it is differentiable through a sigmoid smoothing of the step count, it can enter the loss function directly and be optimized by backpropagation.

What would settle it

Take a small trained MLP or ViT on CIFAR-10 and compute the SS rate both with the paper's proxy and with an exact minimum-box-count algorithm, such as burning or cluster-growing, on the same thresholded feature graph. If the proxy disagrees with the exact box count in the ordering of layers or models, then the reported accuracy gains cannot be attributed to self-similarity.

Watch

Extended reading notes

Core claim

During training, the feature networks $G_M$ of most architectures move from high self-similarity to lower self-similarity. The paper's method penalizes deviation of the SS rate from a target value, namely the post-training SS rate of an unregularized model, through the loss $L_{\text{total}} = L_{\text{task}} + \alpha(SS_{\text{rate}}(G_M^{(k)}) - \gamma)^2$. This constraint consistently improves ResMLP, MLP-Mixer, and ViT small/base, with the largest gain being ViT-Base moving from 61.40 to 67.15 on CIFAR-100; it does not help convolutional networks or PoolFormer, and slightly hurts Vgg. The architecture-dependent effect is corroborated by measurements of statistical scale invariance and geometric invariance, which the paper presents as indicators of intrinsic self-similarity.

Load-bearing premise

The load-bearing premise is that the paper's closed-form formula for the box count is a faithful stand-in for the true box-covering count of the feature graph. If that formula is wrong, the regularizer is just penalizing a clever nonlinear function of pairwise feature distances, and the whole self-similarity story does not carry.

Editorial extensions

If this is right

The S2-Con regularizer can yield accuracy gains near the 6-point ceiling, for instance ViT-Base on CIFAR-100 from 61.40 to 67.15 and MLP-Mixer on CIFAR-10 from 86.81 to 88.27.
The same constraint leaves Vgg, ResNet, and PvT essentially flat or slightly lower, so the benefit is not generic regularization but depends on the architecture's intrinsic self-similarity.
Self-similarity of feature networks is not static: it degrades during training, which gives a new training-dynamics observable for how hidden representations specialize.
Sampling one hidden layer's $G_M$ per iteration keeps the SS-rate regularization computationally tractable for deep models.
The paper notes that the fixed target $\gamma$ may constrain a model's autonomous optimization pathway, pointing toward time-variant self-similarity targets as future work.

Reading between the lines

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

As an editorial extension, the fixed target $\gamma$ is read from the very unregularized training run the method is meant to improve, so the paper's own caveat about fixed targets points toward a curriculum or self-tuning version that has not been tested.
As an editorial extension, the paper's global-aggregation explanation yields a testable prediction the authors did not run: a convolutional network with dilated global receptive fields should start showing S2-Con gains.
As an editorial extension, one could replace the SS-rate regularizer with another differentiable function of the same pairwise-distance matrix; if gains persist, the effect is a generic feature-distance regularizer rather than self-similarity per se.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Desk Editor's Note

read the letter

Short version: the paper gets some real accuracy gains from a new regularizer, but it rests on a metric that does not actually measure what it claims. The stress-test note is correct. Equation 9 defines Nθ as a function of p(C≤θ), the fraction of pairwise distances below θ. For each θ, that is a single scalar from the distance histogram, so the SS rate cannot distinguish between two feature configurations with the same pairwise distance distribution but different network geometry. Real box-covering counts depend on which nodes are close to which. The authors call this an 'approximate simulation' (Appendix A) but never compare it to any actual box-covering algorithm. Without that, the self-similarity story is unsupported.

What is genuinely new here: the GM construction, the differentiable SS rate, and its use as a regularizer. The authors are also honest about their experimental setup—10 runs, mean/std, public code, multiple architectures and datasets. The main empirical observation, Table 3, shows consistent gains on ResMLP, MLP-Mixer, and ViT, and no gains on CNNs. Those numbers are worth checking.

The soft spots are load-bearing. The same metric is used as the training objective (Eq. 6) and as the evidence of self-similarity change (Table 2), so the claim 'constraining self-similarity improves performance' is partly circular. What is not circular is the accuracy improvement, but it is confounded by lack of comparison to simpler regularizers (e.g., penalizing distance histogram skewness, feature decorrelation). The PoolFormer results contradict the narrative—CIFAR-10/100 drop under the constraint—and the paper's explanation (local vs global) is plausible but post-hoc. The auxiliary invariance metrics in Tables 4 and 5 (power-law exponent std, correlation dimension fluctuation) are interesting but are different from SS rate and are also not validated.

If the authors add a direct box-covering comparison on small graphs and show Nθ tracks the true box count, the paper would be substantially stronger. If they also compare against a generic distance-distribution regularizer, they could separate the self-similarity effect from a generic smoothing effect. As is, I would not accept the central claim, but the paper deserves a serious referee because the empirical result is concrete and fixable.

For you: cite it only if you need a reference for 'differentiable self-similarity regularizer' with all caveats. I'd send it to review if I were an editor, but with a request for major revision and the box-covering validation.

Referee Report

4 major / 5 minor

Summary. The paper proposes a complex-network representation GM built from hidden-layer activations, defines a self-similarity rate (SS rate) through a differentiable proxy for box-covering, and adds an MSE regularizer that penalizes the squared difference between the SS rate and a target value gamma. Experiments on CIFAR-10, CIFAR-100, and Imagenette across MLP, CNN, and attention architectures report accuracy gains for ResMLP, MLP-Mixer, and ViT models, and the authors argue that models with intrinsic self-similarity benefit most from the constraint.

Significance. If the proposed metric were validated, the paper would offer a useful, architecture-agnostic tool for monitoring and steering the geometry of hidden representations during training. The experimental protocol is commendable in reporting means and standard deviations over ten runs and in providing public code, and the reported gains (e.g., ViT-Base CIFAR-100 from 61.40 to 67.15) are potentially interesting. However, the significance currently hinges on an unvalidated proxy for box-covering, and the claimed mechanism connecting self-similarity to accuracy is not established by the evidence presented.

major comments (4)

[Appendix A, Eq. (9); Algorithm 2] The proposed proxy Nθ = 1 + (D−1) log_D(D + (1−D)p(C≤θ)) is a fixed monotone function of the single scalar p(C≤θ), so it contains no information about which nodes are close to which; two feature configurations with identical pairwise-distance histograms yield identical SS rates even when their true box-covering counts differ. The paper calls this an approximate simulation of the box-cover method but reports no comparison to any actual box-covering algorithm (such as the burning or greedy algorithms described in Appendix A). Because the same quantity is used both as the training penalty in Eq. (6) and as the evidence of self-similarity in Table 2, the central narrative that constraining self-similarity improves MLP and attention models is unsupported unless Eq. (9) is shown to track true box-covering. A direct comparison on small feature networks, or a counterexample with matched distance histograms, would settle this.
[Eq. (6), Table 3] The loss defines gamma as the inherent self-similarity level of feature networks after standard training without regularization constraints. If the regularizer works, the S2-Con SS rates should move toward the baseline values; instead, Table 3 reports moves away from baseline in many cases (e.g., ResMLP CIFAR-10 0.29 to 0.36, MLP-Mixer CIFAR-10 0.15 to 0.25, ViT-small CIFAR-10 0.49 to 0.51). The paper never reports the gamma values used, nor does it verify that the constraint actually reduces |SS_rate − gamma|. The accuracy gains in Table 3 are therefore not shown to be attributable to the stated self-similarity constraint; they could arise from any smooth auxiliary penalty on pairwise feature distances.
[Section 'Architectural Self-Similarity: The Decisive Factor', Tables 4–5] The claim that performance improvements occur precisely for models with better statistical and geometric invariance is not supported by the tables as presented. ResMLP has the third-worst statistical scale invariance (0.3420) yet shows consistent gains; PoolFormer has a better geometric-invariance score (0.2735) than ViT (0.2772) but degrades on CIFAR-10 and CIFAR-100; and ResNet34/50 show small gains despite poor invariance rankings. No correlation coefficient, significance test, or per-layer analysis is provided, so the proposed mechanism that intrinsic architectural self-similarity governs constraint efficacy is not established.
[Experiments, Eq. (6)] The experimental protocol omits several quantities needed for reproducibility and for interpreting the effect size: the values of gamma per model and dataset, the regularization coefficient alpha (only 'order of magnitude 10−4' is given), the smoothing factor k and gradient scaling factor fac in Algorithm 2, the threshold epsilon in Eq. (1), the number of layers sampled in the random layer selection, and the batch size B used for GM construction. Without these details, the reported improvements cannot be reproduced, and it is unclear whether the results are sensitive to the auxiliary-loss hyperparameters.

minor comments (5)

[Title page] The affiliation contains a typo: 'Inteligence' should be 'Intelligence'.
[Conclusion] The phrase 'an self-similarity constraint' should be 'a self-similarity constraint'.
[After Table 2] The text refers to 'the universal SS rate elevation trend observed in Table 1', but the relevant table is Table 2.
[Figure 2] Figure 2 lacks axis labels and a clear legend, which makes the claimed clustering-versus-diffusion differences among architectures difficult to verify.
[Table 2] The meaning of 'Pre-train' and 'Post-train' in Table 2 should be stated explicitly, since it is unclear whether these denote random-initialization versus standard-trained models or pre-trained versus fine-tuned checkpoints.

Circularity Check

0 steps flagged · score 0.0 of 10

No circular derivation: the reported accuracy gains are measured on external benchmarks and are not forced by the SS-rate loss; the Eq. 9 proxy is a validity concern, not a circular step.

full rationale

The paper's derivation chain is: construct GM from hidden features (Algorithm 1); compute Nθ via the approximate formula in Eq. 9; define SS rate from Nθ (Eqs. 3-5); and add the MSE penalty (SS rate − γ)^2 to the task loss (Eq. 6). The only quantity forced by construction is the regularized SS rate itself, because the loss explicitly minimizes its squared deviation from γ. The paper does not rest its headline claim on that convergence; the headline claim is that S2-Con improves classification accuracy by up to about 6 percentage points (Table 3), and accuracy is an external benchmark that is not an input to the self-similarity loss. No load-bearing self-citation or imported uniqueness theorem appears in the references. The legitimate concern is construct validity, not circularity: Appendix A admits Eq. 9 is an "approximate simulation" of the box-cover method, yet no comparison against a true box-covering algorithm (burning, cluster-growing, or greedy) is reported, so SS rate may not track the box-cover count it is named after. That gap should be addressed by a validation experiment; it does not make the accuracy results circular.

Assumptions & free parameters 5 free parameters · 3 assumptions · 1 invented entities

The central claim depends on several free parameters (alpha, gamma, k, fac, epsilon) and on the unvalidated Nθ proxy. The most critical is the Nθ formula, which is the bridge between raw features and the 'self-similarity' measurement. The target gamma is fitted from the baseline model, making the regularizer's goal data-derived. GM as an invented entity further cements that the 'self-similarity' language is a construction rather than an externally verified quantity.

free parameters (5)

alpha (regularization strength) = 1e-4 (order of magnitude)
Set to 1e-4 without sensitivity analysis; only one order reported.
gamma (target SS rate) = post-training SS rate of baseline model
Computed from the unregularized baseline model's own feature networks, making the target a fitted quantity from data.
k (smoothing factor in sigmoid) = not specified
Controls the differentiability approximation of the indicator function in p(C<=θ); no value given.
fac (gradient scaling factor) = not specified
Scales the gradient flow; no value or tuning procedure given.
epsilon (adjacency threshold) = not specified
Used in Algorithm 1 to build the graph adjacency matrix, but apparently not used in the SS rate computation, which uses the distance matrix directly; its role is unclear.

assumptions (3)

ad hoc to paper The formula Nθ = 1 + (D−1) log_D(D + (1−D)p(C≤θ)) faithfully simulates the box-covering count of the feature network.
Stated without derivation or validation against standard box-covering algorithms or known fractal networks. Appears in Algorithm 2 and Eq. (9) of Appendix A. All self-similarity claims depend on this proxy.
domain assumption The SS rate, defined as a weighted integral of deviation from a line in log-log coordinates, ranges between 0 and 1 and monotonically reflects self-similarity strength.
Asserted in the Method section but never proven. The prefactor and integral form are not shown to produce a bounded, scale-invariant quantity.
domain assumption Hidden-layer feature vectors define a meaningful network where Euclidean distances between feature dimensions capture information relationships.
The paper assumes that treating feature dimensions as nodes and their Euclidean distances as graph distances reflects model topology, but no validation against alternative representations is given.

invented entities (1)

GM feature network
purpose: A graph abstraction of hidden-layer activations used to define self-similarity and the SS rate regularizer.
GM is a novel construct introduced by the authors. It is not validated against external benchmarks, and its connection to actual network topology is asserted rather than demonstrated.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Self-similarity Analysis in Deep Neural Networks." pith.science (2026). https://pith.science/paper/66UEWBF6

@misc{pith2026250717785,
  author       = {Pith},
  title        = {Pith review of: Self-similarity Analysis in Deep Neural Networks},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/66UEWBF6}},
  note         = {Machine review of arXiv:2507.17785}
}

read the original abstract

Current research has found that some deep neural networks exhibit strong hierarchical self-similarity in feature representation or parameter distribution. However, aside from preliminary studies on how the power-law distribution of weights across different training stages affects model performance,there has been no quantitative analysis on how the self-similarity of hidden space geometry influences model weight optimization, nor is there a clear understanding of the dynamic behavior of internal neurons. Therefore, this paper proposes a complex network modeling method based on the output features of hidden-layer neurons to investigate the self-similarity of feature networks constructed at different hidden layers, and analyzes how adjusting the degree of self-similarity in feature networks can enhance the classification performance of deep neural networks. Validated on three types of networks MLP architectures, convolutional networks, and attention architectures this study reveals that the degree of self-similarity exhibited by feature networks varies across different model architectures. Furthermore, embedding constraints on the self-similarity of feature networks during the training process can improve the performance of self-similar deep neural networks (MLP architectures and attention architectures) by up to 6 percentage points.

Figures

Figures reproduced from arXiv: 2507.17785 by the authors.

**Figure 1.** Application workflow of the proposed method on a three-layer fully connected neural network. The diagram illustrates [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗

**Figure 2.** Comparative Visualization of GM Distributions Across Multiple Architectures Pre- and Post-Training low features close to the input are highly sensitive to pretraining, their differences are significant and follow clear patterns. In contrast, the deep features near the output undergo relatively smaller changes after fine-tuning on the target task using the pre-trained model, making them less sensitive to the pre-… view at source ↗

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

37 extracted references · 26 canonical work pages

[1]

, " * write output.state after.block = add.period write newline

ENTRY address archivePrefix author booktitle chapter edition editor eid eprint howpublished institution isbn journal key month note number organization pages publisher school series title type volume year label extra.label sort.label short.list INTEGERS output.state before.all mid.sentence after.sentence after.block FUNCTION init.state.consts #0 'before.a...
[2]

write newline

" write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 global.max substring 't := if while FUNCTION word.in bbl.in capitalize " " * FUNCT...
[3]

Barab \'a si, A.-L.; and Albert, R. 1998. Emergence of scaling in random networks, 1999. Science, 286(5439): 509

work page 1998
[4]

Dorogov, A. Y. 2021 a . Morphological model of self-similar multilayer neural networks. Procedia Computer Science, 186: 366--373

work page 2021
[5]

Dorogov, A. Y. 2021 b . Stratified model of self-similar multi-layer neural network. Procedia Computer Science, 186: 685--692

work page 2021
[6]

Dosovitskiy, A.; Beyer, L.; Kolesnikov, A.; Weissenborn, D.; Zhai, X.; Unterthiner, T.; Dehghani, M.; Minderer, M.; Heigold, G.; Gelly, S.; et al. 2020. An image is worth 16x16 words: Transformers for image recognition at scale. arXiv preprint arXiv:2010.11929

arXiv 2020
[7]

Erkaymaz, O.; Ozer, M.; and Perc, M. 2017. Performance of small-world feedforward neural networks for the diagnosis of diabetes. Applied Mathematics and Computation, 311: 22--28

work page 2017
[8]

Grassberger, P.; and Procaccia, I. 1983. Measuring the strangeness of strange attractors. Physica D: nonlinear phenomena, 9(1-2): 189--208

work page 1983

Show all 37 references

[9]

Han, K.; Xiao, A.; Wu, E.; Guo, J.; Xu, C.; and Wang, Y. 2021. Transformer in transformer. Advances in neural information processing systems, 34: 15908--15919

2021
[10]

He, K.; Zhang, X.; Ren, S.; and Sun, J. 2016. Deep residual learning for image recognition. In Proceedings of the IEEE conference on computer vision and pattern recognition, 770--778

2016
[11]

C.; Papesh, M

Hout, M. C.; Papesh, M. H.; and Goldinger, S. D. 2013. Multidimensional scaling. Wiley Interdisciplinary Reviews: Cognitive Science, 4(1): 93--103

2013
[12]

Howard, J.; et al. 2019. Imagenette: A smaller subset of 10 easily classified classes from imagenet

2019
[13]

T.; Nagy, M.; and Molontay, R

Kov \'a cs, P. T.; Nagy, M.; and Molontay, R. 2021. Comparative analysis of box-covering algorithms for fractal networks. Applied Network Science, 6(1): 73

2021
[14]

Krizhevsky, A.; Hinton, G.; et al. 2009. Learning multiple layers of features from tiny images

2009
[15]

Krizhevsky, A.; Sutskever, I.; and Hinton, G. E. 2012. Imagenet classification with deep convolutional neural networks. Advances in neural information processing systems, 25

2012
[16]

La Malfa, E.; La Malfa, G.; Nicosia, G.; and Latora, V. 2021. Characterizing learning dynamics of deep neural networks via complex networks. In 2021 IEEE 33rd International Conference on Tools with Artificial Intelligence (ICTAI), 344--351. IEEE

2021
[17]

La Malfa, E.; La Malfa, G.; Nicosia, G.; and Latora, V. 2024. Deep Neural Networks via Complex Network Theory: a Perspective. arXiv preprint arXiv:2404.11172

2024 arXiv
[18]

Larsson, G.; Maire, M.; and Shakhnarovich, G. 2016. Fractalnet: Ultra-deep neural networks without residuals. arXiv preprint arXiv:1605.07648

2016 arXiv
[19]

LeCun, Y.; Bengio, Y.; and Hinton, G. 2015. Deep learning. nature, 521(7553): 436--444

2015
[20]

Lee, S.; Lee, S.; Seong, H.; and Kim, E. 2023. Revisiting self-similarity: Structural embedding for image retrieval. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 23412--23421

2023
[21]

Liang, M.; Du, J.; Li, L.; Xue, Z.; Wang, X.; Kou, F.; and Wang, X. 2020. Video super-resolution reconstruction based on deep learning and spatio-temporal feature self-similarity. IEEE Transactions on Knowledge and Data Engineering, 34(9): 4538--4553

2020
[22]

Mahoney, M.; and Martin, C. 2019. Traditional and heavy tailed self regularization in neural network models. In International Conference on Machine Learning, 4284--4293. PMLR

2019
[23]

L.; Carneiro, T

Monteiro, R. L.; Carneiro, T. K. G.; Fontoura, J. R. A.; da Silva, V. L.; Moret, M. A.; and Pereira, H. B. d. B. 2016. A model for improving the learning curves of artificial neural networks. PloS one, 11(2): e0149874

2016
[24]

F.; and Bruno, O

Scabini, L. F.; and Bruno, O. M. 2023. Structure and performance of fully connected neural networks: Emerging complex network properties. Physica A: Statistical Mechanics and its Applications, 615: 128585

2023
[25]

Simonyan, K.; and Zisserman, A. 2014. Very deep convolutional networks for large-scale image recognition. arXiv preprint arXiv:1409.1556

2014 arXiv
[26]

Song, C.; Havlin, S.; and Makse, H. A. 2005. Self-similarity of complex networks. Nature, 433(7024): 392--395

2005
[27]

Testolin, A.; Piccolini, M.; and Suweis, S. 2020. Deep learning systems as complex networks. Journal of Complex Networks, 8(1): cnz018

2020
[28]

Theiler, J. 1990. Statistical precision of dimension estimators. Physical Review A, 41(6): 3038

1990
[29]

O.; Houlsby, N.; Kolesnikov, A.; Beyer, L.; Zhai, X.; Unterthiner, T.; Yung, J.; Steiner, A.; Keysers, D.; Uszkoreit, J.; et al

Tolstikhin, I. O.; Houlsby, N.; Kolesnikov, A.; Beyer, L.; Zhai, X.; Unterthiner, T.; Yung, J.; Steiner, A.; Keysers, D.; Uszkoreit, J.; et al. 2021. Mlp-mixer: An all-mlp architecture for vision. Advances in neural information processing systems, 34: 24261--24272

2021
[30]

Touvron, H.; Bojanowski, P.; Caron, M.; Cord, M.; El-Nouby, A.; Grave, E.; Izacard, G.; Joulin, A.; Synnaeve, G.; Verbeek, J.; et al. 2022. Resmlp: Feedforward networks for image classification with data-efficient training. IEEE transactions on pattern analysis and machine int...

2022
[31]

Wang, W.; Xie, E.; Li, X.; Fan, D.-P.; Song, K.; Liang, D.; Lu, T.; Luo, P.; and Shao, L. 2021. Pyramid vision transformer: A versatile backbone for dense prediction without convolutions. In Proceedings of the IEEE/CVF international conference on computer vision, 568--578

2021
[32]

J.; and Strogatz, S

Watts, D. J.; and Strogatz, S. H. 1998. Collective dynamics of ‘small-world’networks. nature, 393(6684): 440--442

1998
[33]

Wewer, C.; Ilg, E.; Schiele, B.; and Lenssen, J. E. 2023. Simnp: Learning self-similarity priors between neural points. In Proceedings of the IEEE/CVF International Conference on Computer Vision, 8841--8852

2023
[34]

Xie, S.; Kirillov, A.; Girshick, R.; and He, K. 2019. Exploring randomly wired neural networks for image recognition. In Proceedings of the IEEE/CVF International Conference on Computer Vision, 1284--1293

2019
[35]

You, J.; Leskovec, J.; He, K.; and Xie, S. 2020. Graph structure of neural networks. In International Conference on Machine Learning, 10881--10891. PMLR

2020
[36]

Yu, W.; Luo, M.; Zhou, P.; Si, C.; Zhou, Y.; Wang, X.; Feng, J.; and Yan, S. 2022. Metaformer is actually what you need for vision. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, 10819--10829

2022
[37]

Zambra, M.; Maritan, A.; and Testolin, A. 2020. Emergence of network motifs in deep neural networks. Entropy, 22(2): 204

2020

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.

[1] [1]

, " * write output.state after.block = add.period write newline

ENTRY address archivePrefix author booktitle chapter edition editor eid eprint howpublished institution isbn journal key month note number organization pages publisher school series title type volume year label extra.label sort.label short.list INTEGERS output.state before.all mid.sentence after.sentence after.block FUNCTION init.state.consts #0 'before.a...

[2] [2]

write newline

" write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 global.max substring 't := if while FUNCTION word.in bbl.in capitalize " " * FUNCT...

[3] [3]

Barab \'a si, A.-L.; and Albert, R. 1998. Emergence of scaling in random networks, 1999. Science, 286(5439): 509

work page 1998

[4] [4]

Dorogov, A. Y. 2021 a . Morphological model of self-similar multilayer neural networks. Procedia Computer Science, 186: 366--373

work page 2021

[5] [5]

Dorogov, A. Y. 2021 b . Stratified model of self-similar multi-layer neural network. Procedia Computer Science, 186: 685--692

work page 2021

[6] [6]

Dosovitskiy, A.; Beyer, L.; Kolesnikov, A.; Weissenborn, D.; Zhai, X.; Unterthiner, T.; Dehghani, M.; Minderer, M.; Heigold, G.; Gelly, S.; et al. 2020. An image is worth 16x16 words: Transformers for image recognition at scale. arXiv preprint arXiv:2010.11929

arXiv 2020

[7] [7]

Erkaymaz, O.; Ozer, M.; and Perc, M. 2017. Performance of small-world feedforward neural networks for the diagnosis of diabetes. Applied Mathematics and Computation, 311: 22--28

work page 2017

[8] [8]

Grassberger, P.; and Procaccia, I. 1983. Measuring the strangeness of strange attractors. Physica D: nonlinear phenomena, 9(1-2): 189--208

work page 1983