Pith. sign in

REVIEW 4 major objections 4 minor 54 references

Tensor-to-Tensor Models with Fast Iterated Sum Features

T0 review · 4 major / 4 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read Corner trees make a large class of image iterated sums computable in linear time.

desk verdict A genuinely new linear-time building block for iterated-sum features, with an unproven complexity claim and empirical results that need more rigor. read the letter →

arxiv 2506.06041 v1 pith:Q5Z27DUG submitted 2025-06-06 cs.CV cs.LG

Joscha Diehl , Rasheed Ibraheem , Leonard Schmitz , Yue Wu This is my paper
classification cs.CVcs.LG MSC 05A0568Q2568T07
keywords iteratedsumstwo-parametersignaturecornertreesstate-spacemodelsimageclassificationanomalydetectionlinear-timelayertensor-to-tensor
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

Images are two-parameter data, and iterated-sum features that are cheap for sequences become expensive for images. This paper claims that a large class of such two-parameter sums—a subset of the two-parameter sums signature—can actually be computed in linear time and space by reinterpreting each sum as an occurrence count of a "corner tree." The authors package that algorithm as the Fast Iterated Sums (FIS) layer, a tensor-to-tensor layer usable in image networks. If correct, the FIS layer gives a subquadratic building block whose expressiveness comes from the same mathematical object that characterizes image data up to natural equivalence; in experiments, replacing ResNet blocks with FIS matches a larger ResNet's accuracy while cutting parameters and multiply-adds. That matters because it offers a linear-cost alternative to quadratic attention for high-dimensional image and video tensors.

What carries the argument

Corner trees, adapted from permutation-pattern counting, are the governing object: a rooted tree whose edges carry labels from the eight cardinal directions $\{N, NE, E, SE, S, SW, W, NW\}$, each interpreted as a two-point predicate (e.g., $NE(r,s)$ means $s$ is northeast of $r$). The corner tree sum over all assignments of vertices to pixel positions satisfying every edge predicate is re-expressed through the corner-tree pre-sum recurrence, which computes, at each pixel, the node function times the cumulative sums over each child's subtree along the relevant direction. Because cumulative sums along rows and columns make each level cost $O(T_1T_2)$, the whole tree costs $O(nT_1T_2)$; the same recurrence works over any commutative semiring, and the max-plus version is what the experiments use.

What would settle it

Run the FIS layer on tensors of growing size and check that runtime stays proportional to $T_1T_2$: a ratio that grows with input size would falsify the linear-time claim. To test expressiveness, compare FIS features with the full two-parameter sums signature on small images; if the full signature separates image pairs that every random few-node corner tree fails to separate, the subset is too narrow.

Watch

Extended reading notes

Core claim

The paper's central claim is that a significant subset of the two-parameter sums signature of [24] can be computed in linear time and space using corner trees. A corner tree sum assigns each vertex a learned function of the pixel value and each edge one of eight cardinal-direction predicates; the sum over all point configurations consistent with the tree is shown by Theorem 5 to equal the sum over all points of a recursively defined pre-sum built from cumulative sums. That recursion is the algorithm, and it runs in $O(n T_1 T_2)$ time and space for a tree with $n$ vertices and an order-two tensor of size $T_1 \times T_2$, with the same pattern extending to order-$p$ tensors. Packaged as the FIS layer using random trees and the max-plus semiring, it replaces convolutional blocks in a ResNet: one configuration reaches 94.47% top-1 accuracy on CIFAR-10, matching ResNet56's 94.37% with about 20% fewer parameters and 23% fewer multiply-adds, and an FIS encoder in an autoencoder scores an average AUROC of 97.3% on the MVTec texture set.

Load-bearing premise

The FIS layer's practical value rests on the unproven premise that the subset of the two-parameter sums signature computable by a few random corner trees is large and expressive enough to capture the pixel relationships that image tasks need.

Editorial extensions

If this is right

  • A strictly larger subset of the two-parameter sums signature is computable in linear time and space than was previously known, removing the polynomial bottleneck on those terms.
  • The FIS layer can serve as a drop-in replacement for convolutional or attention-like blocks: on CIFAR-10 the ResNet44-downsample configuration reaches 94.47% accuracy versus 94.37% for ResNet56, with about 20% fewer parameters and 23% fewer multiply-adds.
  • After discretization, the same algorithm gives the first practical route to non-diagonal terms of the id-signature of images, making more of that signature accessible for applications.
  • Because the method generalizes to order-$p$ tensors, video and multimodal tensors inherit the same linear-cost feature computation.
  • An autoencoder whose encoder is made of FIS layers reaches an average image-level AUROC of 97.3% on the MVTec AD texture categories, 1.2 points above the same autoencoder with a convolutional encoder.

Reading between the lines

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

  • The paper leaves the coverage of the corner-tree subset unquantified; a natural follow-up is to measure, on small images, how much of the full sums signature's separating power is retained by random few-node trees, and whether larger or learned trees narrow the gap.
  • Viewed as a higher-order generalization of state-space models, the FIS block suggests a family of image architectures where long-range pixel interactions are accumulated by directional cumulative sums; combining such blocks with a selection mechanism, as in sequential state-space models, is a testable extension the paper does not explore.
  • The same corner-tree recurrence could in principle be run backward to provide per-pixel attribution or saliency maps, since each vertex's contribution to the final sum is already a tensor at the layer's native resolution.
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.

Referee Report

4 major / 4 minor

Summary. The paper introduces corner-tree sums, a class of two-parameter iterated sums over image grids, and proposes a recursive dynamic program (Eq. (4)) to compute them. It defines the Fast Iterated Sums (FIS) layer, a differentiable tensor-to-tensor layer based on these sums, and evaluates it on CIFAR-10/100 image classification and MVTec AD texture anomaly detection. The central claims are that the corner-tree construction yields a substantially larger linear-time-computable subset of the two-parameter sums signature of [24] than previously known, that the algorithm runs in O(n T1 T2) time and space, and that replacing ResNet blocks with FIS blocks gives competitive accuracy with fewer parameters and multiply-adds.

Significance. If the complexity and expressiveness claims hold, this is a useful contribution: it connects permutation-pattern counting techniques to tensor feature extraction and provides a novel linear-cost layer with a clean mathematical interpretation. The empirical results, especially the CA-FIS ablation and the anomaly detection results, give initial evidence that the layer is more than a theoretical curiosity. The paper ships code and the recursive formula is machine-checkable, which is a strength. However, the significance is tempered by the lack of a formal complexity proof in the text, unquantified expressiveness claims, the absence of uncertainty estimates for the headline experimental numbers, and the omission of key hyperparameter values.

major comments (4)
  1. [Section 3.2, Eqs. (4)-(5)] The complexity claim O(n T1 T2) is asserted without a cost model for the cumsum operation. Eq. (5) defines cumsum as a sum over a quadrant, and a naive per-pixel evaluation of this sum would cost O(T1 T2) per pixel, i.e., O(T1^2 T2^2) per edge. The text does not state that each directional cumsum is computed by a constant number of prefix/infix scans over the grid, nor does it provide pseudocode to substantiate the bound. Since the experiments use linear/NE trees of depth proportional to n (Section 5.1), the recursive computation along a chain performs one full-grid CTPS per level; this is still O(n T1 T2) only if each level's cumsum is linear in the grid size. Please provide a formal proof or annotated pseudocode with an explicit cost model.
  2. [Section 1 and Section 5.1] The paper claims that a 'substantially larger subset' of the two-parameter sums signature is computable in linear time, but it never quantifies the size of this subset relative to the linear-time subset of [24] or to the full signature. The practical success of the FIS layer is therefore supported only by the specific random-tree configurations used in the experiments; no evidence is given that the chosen corner trees are representative or expressive enough for general image features. This is not a criticism of the algorithm's correctness, but it is a load-bearing assumption for the claim that the FIS layer is a generally useful primitive.
  3. [Abstract and Tables 1, 3, 5] The headline claim in the abstract that a smaller network with FIS blocks achieves 'similar accuracy (with a difference of only 0.1%)' rests on a single comparison: ResNet44 Downsample (94.47%) versus ResNet56 Base (94.37%) in Table 1. Other configurations show larger gaps (e.g., ResNet32 L23 is 2.90% below base), and Table 1 reports a single run per architecture with no standard deviations or seeds. Table 5 similarly reports single AUROC values per category. Given that Table 3 shows run-to-run variation of up to 0.85% standard deviation in a related setting, the 0.1% difference is within plausible training noise and should not be presented without repeated-seed results.
  4. [Section 5.1, Section 6.1.3, Appendix A] The hyperparameters of the FIS layer, most importantly the number of corner trees N_T and the number of nodes per tree, are never reported for the experiments in Tables 1-6. Section 5.1 lists these as hyperparameters, but Section 6.1.3 only states that the max-plus semiring with random tree structure was adopted, and the appendix tables cover semiring, tree type, top-n threshold, latent dimension, and backbone, but not N_T or node counts. The code is a substitute, but the paper should state the values used so that the results are reproducible from the text alone.
minor comments (4)
  1. [Section 3, contribution bullet] The contribution bullet claims the algorithm 'yields the first practical method for computing non-diagonal terms of the integrals-signature,' but Section 3 only gives a single example (the second-derivative approximation in Example 2) and does not explain how arbitrary non-diagonal terms are obtained; either expand this argument or soften the claim.
  2. [Table 5 caption] The caption states that the PatchCore comparison uses the model 'with the best average AUROC' chosen from [51], while FIS-AE is reported with a single default configuration; this asymmetry should be stated in the main text so that the comparison is not read as a like-for-like best-versus-default evaluation.
  3. [Section 6.2.2] There is a typo in 'ImgaNet-pretrained ResNet' (should be 'ImageNet'); similar typos include 'conestellations' in Example 2, 'exaple' in the caption of Figure 10, and 'Secation' in Section 6.2.6.
  4. [Section 4] The generalization to order-p tensors is described as 'analogously' linear-time without details; if this is to remain a stated contribution, a brief explanation of how the directional scans generalize to p dimensions would be helpful.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the corner-tree sums are independently defined and their computation is justified by a self-contained recursion theorem, with self-citations serving as background only.

full rationale

The paper's central derivation is self-contained and does not reduce to its own inputs. The corner tree sum CTS is defined independently in Definition 1 via the Allowed predicate, and the recursive formula CTPS in Eq. (4) together with Theorem 5 gives a genuine proof that CTS(T,z) equals the sum over CTPS(T,z). The algorithm's linear-time claim is asserted with a pointer to the implementation rather than fully proved in the text, but that is a rigor or verification gap, not a circularity: no equation is shown to be equivalent to another by construction, and no fitted parameter is renamed as a prediction. The experimental pipeline is standard: hyperparameters such as the max-plus semiring and random tree structures are chosen via validation and then evaluated on held-out test sets, with ablations comparing FIS against convolutional baselines. Self-citations to [24] (two-parameter sums signature) and [41] (edge-labeled corner trees) supply background and reformulations, but the paper does not rely on an unverified self-citation to force its central conclusion; the relevant definitions and the recursion proof are stated fully in the manuscript. The expressiveness of the computed subset is an assumption supported only by experiments, and the complexity proof is incomplete in the writeup, but neither constitutes circularity under the stated criteria. The honest finding is therefore no significant circularity, with score 0.

Assumptions & free parameters 6 free parameters · 6 assumptions · 0 invented entities

The central algorithm is self-contained and parameter-free; all free parameters listed are empirical hyperparameters of the FIS layer and the anomaly detector. The axioms are standard algorithmic ingredients or domain assumptions inherited from prior work; the expressiveness of the corner-tree subset is the main unproven premise.

free parameters (6)
  • Tree size (nodes per corner tree)
    Hyperparameter chosen by the user; controls the expressiveness and cost of the FIS layer. No value is given in the paper text.
  • Number of corner trees NT
    Hyperparameter that sets the output channel count of the FIS layer.
  • Semiring type = max-plus
    Tuned on validation (Section 6.1.3); max-plus with random trees gave the best classification accuracy.
  • Tree structure type = random
    Tuned on validation (Section 6.1.3); random trees outperformed linear and linear-NE in the CA-FIS ablation (Table 8).
  • Latent dimension l (FIS-AE) = 32
    Default for the anomaly detection autoencoder; ablation in Table 10 shows 32 balances AUROC and variance.
  • Top-n anomaly score threshold = 10
    Mean of the top 10 reconstruction errors is used as the anomaly score; Table 9 shows other values give similar results.
assumptions (6)
  • standard math The recursive factorization of corner tree sums (Theorem 5) is valid for the extended 8-direction edge label set C.
    Proven in the paper via the decomposition of Allowed(T, r) into root-edge predicates and child subtree predicates.
  • standard math Two-dimensional dominance sums (cumsum) for each cardinal direction can be computed in O(T1 T2) time.
    Invoked in Eq. (5) and Section 3.2; standard prefix/suffix sum algorithms give linear time.
  • standard math The max-plus semiring is a commutative semiring, so the recursion holds with max and plus operations.
    Used in Section 3.1 for the max-plus variant of corner tree sums.
  • domain assumption A discrete second derivative z approximates the continuous second derivative and yields the id-signature on the diagonal word ii.
    Used in Example 2 to connect CTS to the id-signature of [25]; not derived here.
  • domain assumption Arbitrary state-space model transition matrices can be approximated by polynomials in the input and time, so iterated sums can approximate any SSM.
    This heuristic in Section 2.2 motivates the interpretation of FIS as a higher-order SSM generalization, but is not formalized.
  • ad hoc to paper The corner-tree computable subset of the two-parameter sums signature is expressive enough for image feature extraction.
    The paper asserts a 'substantially larger subset' but does not quantify coverage; empirical performance is the only evidence.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Tensor-to-Tensor Models with Fast Iterated Sum Features." pith.science (2026). https://pith.science/paper/Q5Z27DUG

@misc{pith2026250606041,
  author       = {Pith},
  title        = {Pith review of: Tensor-to-Tensor Models with Fast Iterated Sum Features},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/Q5Z27DUG}},
  note         = {Machine review of arXiv:2506.06041}
}
read the original abstract

Data in the form of images or higher-order tensors is ubiquitous in modern deep learning applications. Owing to their inherent high dimensionality, the need for subquadratic layers processing such data is even more pressing than for sequence data. We propose a novel tensor-to-tensor layer with linear cost in the input size, utilizing the mathematical gadget of ``corner trees'' from the field of permutation counting. In particular, for order-two tensors, we provide an image-to-image layer that can be plugged into image processing pipelines. On the one hand, our method can be seen as a higher-order generalization of state-space models. On the other hand, it is based on a multiparameter generalization of the signature of iterated integrals (or sums). The proposed tensor-to-tensor concept is used to build a neural network layer called the Fast Iterated Sums (FIS) layer which integrates seamlessly with other layer types. We demonstrate the usability of the FIS layer with both classification and anomaly detection tasks. By replacing some layers of a smaller ResNet architecture with FIS, a similar accuracy (with a difference of only 0.1\%) was achieved in comparison to a larger ResNet while reducing the number of trainable parameters and multi-add operations. The FIS layer was also used to build an anomaly detection model that achieved an average AUROC of 97.3\% on the texture images of the popular MVTec AD dataset. The processing and modelling codes are publicly available at https://github.com/diehlj/fast-iterated-sums.

Figures

Figures reproduced from arXiv: 2506.06041 by the authors.

Figure 1
Figure 1. Example point constellations for the iterated sum P 1≤t1<t2≤T −1 xt1 x 2 t2 . t1 symbolized by a blue circle, t2 by a red diamon. 2 In particular, the familiar maxplus semiring will be seen to be beneficial in the experiments. 3 [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Example point constellations for the sum (1). r 1 symbolized by a red circle, r 2 by a blue cross, r 3 by a green diamond, r 4 by a purple x. Such expressions are sufficiently expressive (they characterize the input, up to a natural equivalence relation), but are, in general, not computable in linear time [24]. In the current paper, only a subset of these sums is considered; namely those whose point constellation ca… view at source ↗
Figure 3
Figure 3. A corner tree and its occurrence in a permutation. An “occurence” of such a corner tree in a permutation is a map from the vertices of the corner tree to the indices of the permutation such that the labels of the edges “are respected”. More precisely, vertices of the tree are mapped to indices in the permutation and the east-west direction corresponds to the ordering of the indices whereas the north-south direction … view at source ↗
Figures from the paper (8 more)
Figure 4
Figure 4. Figure 4: Three examples of corner trees. Definition 1. A corner tree is a rooted, edge- and vertex-labeled tree T = (V (T ), E(T ), v : V (T ) → Fd, e : E(T ) → C). Here • V (T ) is a finite set of vertices, • E(T ) ⊂ V (T ) × V (T ) is a set of edges, such that (V (T ), E(T ))…
Figure 5
Figure 5. Figure 5: Visualization of point constellations for different corner trees. Example 3. With the corner tree T ′′ in [PITH_FULL_IMAGE:figures/full_fig_p007_5.png]
Figure 6
Figure 6. Figure 6: Point constellation for the corner tree in (6). The algorithm described in Section 3 works analogously for this case, with complexity again being linear in the input size. Example 6. The edge-labels can be thought of as declaring a “zone of influence” of the informatio…
Figure 7
Figure 7. Figure 7: Corner tree and its a three dimensional visualization. 5 FIS layers and FIS blocks This section focuses on the discussion of integrating Fast Iterated Sums (FIS), as described in Section 3, into existing neural network architectures to facilitate efficient image data p…
Figure 8
Figure 8. Figure 8: Illustraction of (a) FIS Layer and (b) FIS Block architecture. An FIS Block comprises two FIS Layers (in green) and an Adaptive Pooling Layer (in red) of output size (H′ , W′ ). A Batch Normalization and RELU layers (both in yellow) follow each FIS Layer. 6 Application…
Figure 9
Figure 9. Figure 9: ResNet20 architecture designed for CIFAR-10 dataset. This is the architecture used in [46]. A batch normalization and RELU layers follow each convolution layer. 12 [PITH_FULL_IMAGE:figures/full_fig_p012_9.png]
Figure 10
Figure 10. Figure 10: An exaple of modified ResNet20 architecture (L23) designed for CIFAR-10 dataset. Here, the last 2 convolution Basic Blocks are replaced by the FIS Block. A single FIS Block comprises 2 Fast Iterated Sums Layer and a Pooling Layer (which can be Max Pooling or Average P…
Figure 11
Figure 11. Figure 11: FIS-AE architecture. Features are extracted from the normal images using the ImageNet pre-trained ResNet, which are then fed to the autoencoder. These features are encoded by the FIS Layers using l channels. 6.2.3 Autoencoder as anomaly detector The way FIS-AE describ…

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

54 extracted references · 40 canonical work pages

  1. [24]

    Two-parameter sums signatures and corresponding quasisymmetric functions

    Joscha Diehl and Leonard Schmitz. Two-parameter sums signatures and corresponding quasisym- metric functions, 2022. arXiv:2210.14247

    work page Pith review arXiv 2022

  2. [1]

    Sequence to sequence learning with neural networks

    Ilya Sutskever, Oriol Vinyals, and Quoc V Le. Sequence to sequence learning with neural networks. Advances in neural information processing systems , 27, 2014

    2014

  3. [2]

    Improving language under- standing by generative pre-training

    Alec Radford, Karthik Narasimhan, Tim Salimans, Ilya Sutskever, et al. Improving language under- standing by generative pre-training. 2018

    2018

  4. [3]

    Bert: Pre-training of deep bidirectional transformers for language understanding

    Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. Bert: Pre-training of deep bidirectional transformers for language understanding. In Proceedings of the 2019 conference of the North American chapter of the association for computational linguistics: human language technolo- gies, volume 1 (long and short papers) , pages 4171–4186, 2019

    work page 2019

  5. [4]

    Attention is all you need

    Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N Gomez, Lukasz Kaiser, and Illia Polosukhin. Attention is all you need. Advances in neural information processing systems, 30, 2017. 20

    work page 2017

  6. [5]

    Llama: Open and efficient foundation language models

    Hugo Touvron, Thibaut Lavril, Gautier Izacard, Xavier Martinet, Marie-Anne Lachaux, Timoth´ ee Lacroix, Baptiste Rozi` ere, Naman Goyal, Eric Hambro, Faisal Azhar, et al. Llama: Open and efficient foundation language models. arXiv preprint arXiv:2302.13971 , 2023

    arXiv 2023

  7. [6]

    U-net: Convolutional networks for biomedical image segmentation

    Olaf Ronneberger, Philipp Fischer, and Thomas Brox. U-net: Convolutional networks for biomedical image segmentation. In Medical image computing and computer-assisted intervention–MICCAI 2015: 18th international conference, Munich, Germany, October 5-9, 2015, proceedings, part III 18 , pages 234–241. Springer, 2015

    2015

  8. [7]

    3d u-net: learning dense volumetric segmentation from sparse annotation

    ¨Ozg¨ un C ¸ i¸ cek, Ahmed Abdulkadir, Soeren S Lienkamp, Thomas Brox, and Olaf Ronneberger. 3d u-net: learning dense volumetric segmentation from sparse annotation. In Medical Image Computing and Computer-Assisted Intervention–MICCAI 2016: 19th International Conference, Athens, Greece, October 17-21, 2016, Proceedings, Part II 19 , pages 424–432. Springer, 2016

    work page 2016

Show all 54 references
  1. [8]

    Image-to-image translation with conditional adversarial networks

    Phillip Isola, Jun-Yan Zhu, Tinghui Zhou, and Alexei A Efros. Image-to-image translation with conditional adversarial networks. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 1125–1134, 2017

    2017

  2. [9]

    Video-to-video synthesis

    Ting-Chun Wang, Ming-Yu Liu, Jun-Yan Zhu, Guilin Liu, Andrew Tao, Jan Kautz, and Bryan Catanzaro. Video-to-video synthesis. In Proceedings of the 32nd International Conference on Neural Information Processing Systems, pages 1152–1164, 2018

    2018

  3. [10]

    An image is worth 16x16 words: Transformers for image recognition at scale

    Alexey Dosovitskiy, Lucas Beyer, Alexander Kolesnikov, Dirk Weissenborn, Xiaohua Zhai, Thomas Unterthiner, Mostafa Dehghani, Matthias Minderer, G Heigold, S Gelly, et al. An image is worth 16x16 words: Transformers for image recognition at scale. In International Conference on...

    2020

  4. [11]

    Differential equations driven by rough signals (i): An extension of an inequality of lc young

    Terry Lyons. Differential equations driven by rough signals (i): An extension of an inequality of lc young. Mathematical Research Letters, 1(4), 1994

    1994

  5. [12]

    Efficiently modeling long sequences with structured state spaces

    Albert Gu, Karan Goel, and Christopher R´ e. Efficiently modeling long sequences with structured state spaces. arXiv preprint arXiv:2111.00396 , 2021

    2021 arXiv

  6. [13]

    Mamba: Linear-time sequence modeling with selective state spaces

    Albert Gu and Tri Dao. Mamba: Linear-time sequence modeling with selective state spaces. arXiv preprint arXiv:2312.00752, 2023

    2023 arXiv

  7. [14]

    A discrete state-space model for linear image processing

    Robert Roesser. A discrete state-space model for linear image processing. IEEE transactions on automatic control, 20(1):1–10, 1975

    1975

  8. [15]

    State space representations of the roesser type for convolutional layers

    Patricia Pauli, Dennis Gramlich, and Frank Allg¨ ower. State space representations of the roesser type for convolutional layers. IF AC-PapersOnLine, 58(17):344–349, 2024

    2024

  9. [16]

    Vmamba: Visual state space model

    Yue Liu, Yunjie Tian, Yuzhong Zhao, Hongtian Yu, Lingxi Xie, Yaowei Wang, Qixiang Ye, Jianbin Jiao, and Yunfan Liu. Vmamba: Visual state space model. Advances in neural information processing systems, 37:103031–103063, 2024

    2024

  10. [17]

    Theo- retical foundations of deep selective state-space models

    Nicola Muca Cirone, Antonio Orvieto, Benjamin Walker, Cristopher Salvi, and Terry Lyons. Theo- retical foundations of deep selective state-space models. Advances in Neural Information Processing Systems, 37:127226–127272, 2024

    2024

  11. [18]

    Integration of paths, geometric invariants and a generalized baker-hausdorff formula

    Kuo-Tsai Chen. Integration of paths, geometric invariants and a generalized baker-hausdorff formula. Annals of Mathematics , 65(1):163–178, 1957

    1957

  12. [19]

    Neural rough differential equa- tions for long time series

    James Morrill, Cristopher Salvi, Patrick Kidger, and James Foster. Neural rough differential equa- tions for long time series. In International Conference on Machine Learning , pages 7829–7838. PMLR, 2021

    2021

  13. [20]

    Early prediction of lithium-ion cell degradation trajectories using signatures of voltage curves up to 4-minute sub-sampling rates

    Rasheed Ibraheem, Yue Wu, Terry Lyons, and Goncalo Dos Reis. Early prediction of lithium-ion cell degradation trajectories using signatures of voltage curves up to 4-minute sub-sampling rates. Applied Energy, 352:121974, 2023

    2023

  14. [21]

    FRUITS: Feature extraction using iterated sums for time series classification

    Joscha Diehl and Richard Krieg. FRUITS: Feature extraction using iterated sums for time series classification. Data Mining and Knowledge Discovery , 38(6):4122–4156, November 2024. 21

    2024

  15. [22]

    Signature moments to characterize laws of stochastic pro- cesses

    Ilya Chevyrev and Harald Oberhauser. Signature moments to characterize laws of stochastic pro- cesses. Journal of Machine Learning Research , 23(176):1–42, 2022

    2022

  16. [23]

    Rough transformers: Lightweight and continuous time series modelling through signature patching

    Fernando Moreno-Pino, ´Alvaro Arroyo, Harrison Waldon, Xiaowen Dong, and ´Alvaro Cartea. Rough transformers: Lightweight and continuous time series modelling through signature patching. Ad- vances in Neural Information Processing Systems , 37:106264–106294, 2024

    2024

  17. [25]

    Harang, and Samy Tindel

    Joscha Diehl, Kurusch Ebrahimi-Fard, Fabian N. Harang, and Samy Tindel. On the signature of an image. Stochastic Processes and their Applications , 187:104661, 2025

    2025

  18. [26]

    Two-dimensional signature of images and texture classification

    Sheng Zhang, Guang Lin, and Samy Tindel. Two-dimensional signature of images and texture classification. Proceedings of the Royal Society A: Mathematical, Physical and Engineering Sciences , 478(2266), oct 2022

    2022

  19. [27]

    A topological approach to mapping space signatures

    Chad Giusti, Darrick Lee, Vidit Nanda, and Harald Oberhauser. A topological approach to mapping space signatures. Advances in Applied Mathematics , 163:102787, 2025

    2025

  20. [28]

    Signature matrices of membranes, 2024

    Felix Lotter and Leonard Schmitz. Signature matrices of membranes, 2024. arXiv:arxiv.org/abs/2409.11996

    2024

  21. [29]

    Two-dimensional signature of images and texture classification

    Sheng Zhang, Guang Lin, and Samy Tindel. Two-dimensional signature of images and texture classification. Proceedings of the Royal Society A , 478(2266):20220346, 2022

    2022

  22. [30]

    2dsig-detect: a semi-supervised framework for anomaly detection on image data using 2d-signatures, 2025

    Xinheng Xie, Kureha Yamaguchi, Margaux Leblanc, Simon Malzard, Varun Chhabra, Victoria Nock- les, and Yue Wu. 2dsig-detect: a semi-supervised framework for anomaly detection on image data using 2d-signatures, 2025

    2025

  23. [31]

    A multiplicative surface signature through its magnus expansion

    Ilya Chevyrev, Joscha Diehl, Kurusch Ebrahimi-Fard, and Nikolas Tapia. A multiplicative surface signature through its magnus expansion. arXiv preprint arXiv:2406.16856 , 2024

    2024 arXiv

  24. [32]

    The surface signature and rough surfaces

    Darrick Lee. The surface signature and rough surfaces. arXiv preprint arXiv:2406.16857 , 2024

    2024 arXiv

  25. [33]

    Random surfaces and higher algebra

    Darrick Lee and Harald Oberhauser. Random surfaces and higher algebra. arXiv preprint arXiv:2311.08366, 2023

    2023

  26. [34]

    Counting small permutation patterns

    Chaim Even-Zohar and Calvin Leng. Counting small permutation patterns. In Proceedings of the 2021 ACM-SIAM Symposium on Discrete Algorithms (SODA) , pages 2288–2302. SIAM, 2021

    2021

  27. [35]

    Mvtec ad–a comprehensive real-world dataset for unsupervised anomaly detection

    Paul Bergmann, Michael Fauser, David Sattlegger, and Carsten Steger. Mvtec ad–a comprehensive real-world dataset for unsupervised anomaly detection. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages 9592–9600, 2019

    2019

  28. [36]

    Time-warping invariants of multidimen- sional time series

    Joscha Diehl, Kurusch Ebrahimi-Fard, and Nikolas Tapia. Time-warping invariants of multidimen- sional time series. Acta Applicandae Mathematicae, 170(1):265–290, 2020

    2020

  29. [37]

    Tropical time series, iterated-sums signa- tures, and quasisymmetric functions

    Joscha Diehl, Kurusch Ebrahimi-Fard, and Nikolas Tapia. Tropical time series, iterated-sums signa- tures, and quasisymmetric functions. SIAM Journal on Applied Algebra and Geometry, 6(4):563–599, 2022

    2022

  30. [38]

    Prefix sums and their applications

    Guy E Blelloch. Prefix sums and their applications. 1990

    1990

  31. [39]

    Seq2tens: An efficient representation of se- quences by low-rank tensor projections

    Csaba Toth, Patric Bonnier, and Harald Oberhauser. Seq2tens: An efficient representation of se- quences by low-rank tensor projections. In International Conference on Learning Representations , 2021

    2021

  32. [40]

    Towards a unifying sequence-to-sequence deep learning layer based on iterated sums

    Richard Krieg. Towards a unifying sequence-to-sequence deep learning layer based on iterated sums. Master’s thesis, University of Greifswald, 2024

    2024

  33. [41]

    On a generalization of corner trees

    Joscha Diehl and Emanuele Verri. On a generalization of corner trees. arXiv preprint arXiv:2408.08293, 2024

    2024 arXiv

  34. [42]

    Counting permutation patterns with multidimensional trees

    Gal Beniamini and Nir Lavee. Counting permutation patterns with multidimensional trees. arXiv preprint arXiv:2407.04971, 2024. 22

    2024 arXiv

  35. [43]

    Semirings and their Applications

    Jonathan S Golan. Semirings and their Applications . Springer Science & Business Media, 2013

    2013

  36. [44]

    Tropical geometry of deep neural networks

    Liwen Zhang, Gregory Naitzat, and Lek-Heng Lim. Tropical geometry of deep neural networks. In International Conference on Machine Learning , pages 5824–5832. PMLR, 2018

    2018

  37. [45]

    Deep residual learning for image recog- nition

    Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image recog- nition. In Proceedings of the IEEE conference on computer vision and pattern recognition , pages 770–778, 2016

    2016

  38. [46]

    Image classification codebase

    Yaofo Chen. Image classification codebase. https://github.com/chenyaofo/ image-classification-codebase, 2021

    2021

  39. [47]

    Learning multiple layers of features from tiny images

    Alex Krizhevsky, Geoffrey Hinton, et al. Learning multiple layers of features from tiny images. 2009

    2009

  40. [48]

    Texture image analysis and texture classification methods-a review

    Laleh Armi and Shervan Fekri-Ershad. Texture image analysis and texture classification methods-a review. arXiv preprint arXiv:1904.06554 , 2019

    1904 arXiv

  41. [49]

    Anomaly detection in medical imaging-a mini review

    Maximilian E Tschuchnig and Michael Gadermayr. Anomaly detection in medical imaging-a mini review. In International Data Science Conference , pages 33–38. Springer, 2021

    2021

  42. [50]

    Zero-shot versus many-shot: Un- supervised texture anomaly detection

    Toshimichi Aota, Lloyd Teh Tzer Tong, and Takayuki Okatani. Zero-shot versus many-shot: Un- supervised texture anomaly detection. In Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision , pages 5564–5572, 2023

    2023

  43. [51]

    Towards total recall in industrial anomaly detection

    Karsten Roth, Latha Pemula, Joaquin Zepeda, Bernhard Sch¨ olkopf, Thomas Brox, and Peter Gehler. Towards total recall in industrial anomaly detection. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages 14318–14328, 2022

    2022

  44. [52]

    Unsupervised anomaly detection for x-ray images

    Diana Davletshina, Valentyn Melnychuk, Viet Tran, Hitansh Singla, Max Berrendorf, Evgeniy Faer- man, Michael Fromm, and Matthias Schubert. Unsupervised anomaly detection for x-ray images. arXiv preprint arXiv:2001.10883 , 2020

    2001 arXiv

  45. [53]

    Anomaly detection using autoencoders with nonlinear dimen- sionality reduction

    Mayu Sakurada and Takehisa Yairi. Anomaly detection using autoencoders with nonlinear dimen- sionality reduction. In Proceedings of the MLSDA 2014 2nd workshop on machine learning for sensory data analysis , pages 4–11, 2014

    2014

  46. [54]

    Inverse-transform autoencoder for anomaly detection

    Chaoqing Huang, Jinkun Cao, Fei Ye, Maosen Li, Ya Zhang, and Cewu Lu. Inverse-transform autoencoder for anomaly detection. arXiv preprint arXiv:1911.10676 , 2(4), 2019. 23

    1911 arXiv

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.