Pith. sign in

REVIEW 4 major objections 7 minor 31 references

MNIST-Gen: A Modular MNIST-Style Dataset Generation Using Hierarchical Semantics, Reinforcement Learning, and Category Theory

T0 review · 4 major / 7 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 MNIST-Gen builds custom MNIST-style image datasets on demand, auto-labeling 85% of images with modest human review.

desk verdict A useful dataset-generation pipeline whose headline accuracy claims are unvalidated and partially circular; deserves peer review but needs reworked evaluation. read the letter →

arxiv 2507.11821 v1 pith:JXL77V2G submitted 2025-07-16 cs.LG cs.AIcs.CVcs.HC

Pouya Shaeri , Arash Karimi , Ariane Middel This is my paper · ORCID
classification cs.LGcs.AIcs.CVcs.HC
keywords hierarchicalsemanticcategorizationCLIPMNIST-styledatasetgenerationreinforcementlearninghuman-in-the-loopcategorytheoryFood-MNISTTree-MNIST
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

The paper presents MNIST-Gen, an automated framework that generates MNIST-style grayscale image datasets for user-defined categories. It claims that combining CLIP-based semantic embeddings, hierarchical category definitions, and a reinforcement-learning agent with human feedback lets the system auto-categorize 85% of images in smart-batch mode while cutting annotation time by about 80%, from 25.0 to 5.2 hours on a 5,000-image food dataset. Two proof-of-concept datasets, Tree-MNIST and Food-MNIST, are produced and benchmarked, and the RL-filtered versions improve downstream classification accuracy by 5–8% over unfiltered or CLIP-only versions. The authors argue this lowers the barrier to task-specific benchmark creation in domains where standard datasets are missing or costly to publish.

What carries the argument

The argument is carried by the hierarchical semantic score, which combines CLIP text–image similarity, characteristic matching, and visual attributes; by the category-theoretic functor $F: \mathcal{I} \to \mathcal{M}$ that guarantees compositionality of the pipeline; and by a DQN agent whose reward balances semantic confidence, class-balance entropy, model accuracy, and redundancy. The functor framing licenses replacing or reordering any transformation step without changing the overall dataset semantics, and the RL reward is what drives the reported 5–8% downstream accuracy improvement.

What would settle it

Run the smart-batch pipeline on an image set with fixed ground-truth labels, such as a held-out subset of Food-101, and measure the auto-assigned categories against those labels; if the auto-categorization agreement is far below 85% while human review still corrects many errors, the headline accuracy and the claimed time savings would not survive a labeled benchmark.

Watch

Extended reading notes

Core claim

The central claim is that a dataset-generation pipeline can be made compositional and adaptive enough that researchers can construct small, publication-ready MNIST-style benchmarks for arbitrary hierarchical categories with minimal manual labeling. The authors establish this through a six-stage pipeline—keyword retrieval, hierarchical category definition, CLIP semantic analysis, three processing modes, RL-based sample filtering, and a transformation chain modeled as composable morphisms—and support it with two generated datasets. On Food-MNIST's smart-batch mode, the system auto-categorizes 85% of images with an overall accuracy of 89.7% against human review, and the RL-filtered version yields 73.84% CNN accuracy on the 10-class task; on Tree-MNIST the CNN reaches 86.51%.

Load-bearing premise

The evaluation assumes that CLIP-based semantic similarity, validated by human review, is a good enough proxy for true category membership, because the images come from unlabeled public repositories and the paper states there is no definitive ground truth.

Editorial extensions

If this is right

  • Researchers can create task-specific MNIST-style benchmarks without collecting and hand-labeling data from scratch.
  • In smart-batch mode, a 5,000-image dataset can be curated in about 5 hours rather than 25, with 15% of images needing human review.
  • Reinforcement-learning filtering, trained on the same pipeline, adds 5–8% downstream classification accuracy over CLIP-only selection.
  • Because the pipeline is modeled as composable morphisms, swapping in a new semantic model or preprocessing step should preserve consistency across stages.

Reading between the lines

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

  • The no-ground-truth evaluation means the headline 85% accuracy measures agreement with CLIP-guided human review, not agreement with a fixed external label set; a fair reader should treat it as a process claim rather than a label-quality guarantee.
  • A direct testable extension would be to run the smart-batch pipeline on a labeled benchmark such as a Food-101 subset and report auto-categorization agreement with the known labels, converting the framework's claims into falsifiable numbers.
  • The RL reward's model-accuracy term creates a feedback loop that may inflate the measured 5–8% gain; an ablation with that term zeroed would isolate the semantic-filtering contribution.
  • The functoriality guarantee holds only for deterministic transformations, so the modularity claim would need re-testing if stochastic or learnable stages are inserted into the pipeline.
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 / 7 minor

Summary. The paper presents MNIST-Gen, a modular pipeline for generating MNIST-style grayscale image datasets from user-specified hierarchical category definitions. The system combines keyword-driven image retrieval, CLIP-based semantic analysis, three processing modes (individual review, smart batch, fast batch), a Deep Q-Learning agent for adaptive sample filtering, and a category-theoretic formalization of the transformation pipeline. As proof of concept, the authors construct Tree-MNIST and Food-MNIST and evaluate them by training classical and deep classifiers. The headline claims are 85% automatic categorization accuracy, an 89.7% label accuracy in smart batch mode, a 5–8% classification accuracy improvement from RL+semantic filtering over random selection, and an 80% reduction in manual annotation time compared with manual curation.

Significance. If the empirical claims were properly supported, MNIST-Gen would be a practically useful contribution for researchers who need small, task-specific benchmark datasets without publishing full datasets. The modular architecture, the hierarchical semantic representation with explicit semantic characteristics, and the integration of CLIP with a DQN-based selection policy are interesting and clearly described. The paper also includes honest limitations (Section V) and provides two new datasets, though the datasets themselves are not yet released. However, the evaluation as it stands does not justify the quantitative claims: the RL reward includes the same downstream accuracy metric used for evaluation, the categorization accuracy is not benchmarked against an independent ground truth even though one exists (Food-11), ambiguous images are removed before evaluation without accounting for the resulting selection bias, and no error bars or significance tests are reported. These issues are load-bearing because they concern exactly the numbers announced in the abstract and conclusion.

major comments (4)
  1. [Section IV.E and Eq. (16)] The RL reward in Eq. (16) includes λ3·ModelAcc, the accuracy of a lightweight model trained on the current batch. The paper then reports in Section IV.E that the RL+Semantic filtering variant achieves a 5–8% improvement in classification accuracy over random and CLIP-only filtering. This is circular: the filter is optimized against the same type of accuracy signal that is later used as the outcome measure. To support the claim, the authors should train the final downstream classifiers on a fixed, pre-registered test split that is never exposed to the RL loop, and report accuracy on that independent test set for all three filtering variants (random, CLIP-only, RL+CLIP). Without this separation, the reported improvement is partly a selection artifact.
  2. [Section IV.A and Section V] The paper states in Section V that “there is no definitive ground truth against which to benchmark performance,” yet Section IV.A says the 5,000 Food-MNIST images were “retrieved from the Kaggle Food-11 dataset provided by [30] to compare and validate subcategory labeling.” Food-11 is a labeled dataset; if the original labels were available, categorization accuracy should be measured against them rather than against human-review labels that were themselves generated with CLIP suggestions (Section III.G). At minimum, the authors should explain why Food-11 labels cannot serve as ground truth, and they should define exactly what the “Accuracy %” column in Table I measures (agreement with a human reviewer who saw CLIP suggestions, agreement with Food-11 labels, or something else).
  3. [Section IV.A] The evaluation excludes “visually ambiguous images that contained multiple food types from different categories (e.g., eggs and fries on the same plate)” before computing classification accuracy. This removes exactly the hard cases that a real-world dataset-generation pipeline would encounter, and the degree of filtering may affect the three processing modes differently. The authors should report the number and proportion of removed images, show the classification accuracy on the removed set, and run the processing-mode comparison on the full retrieved set (or at least demonstrate that the exclusion rule does not change the ranking of modes). Without this, the reported accuracies in Tables I–III are conditional on an unquantified favorable filter.
  4. [Tables II–III and Section IV.E] All accuracy numbers are reported as point estimates without error bars, confidence intervals, or significance tests, and it is unclear how many independent runs were performed. With dataset sizes of 1,500 and 5,000 images and an 80/20 split, a single run is not sufficient to establish a 5–8% difference between filtering variants; the observed gap could be within run-to-run noise. Please provide results over multiple seeds with standard deviations and, ideally, a paired statistical test for the RL+Semantic versus Semantic-only comparison.
minor comments (7)
  1. [Footnote 1] The code and data links are placeholders ("(mnist-gen-code)" and "(data-batch1)"); actual URLs or a DOI should be provided before publication so the claimed reproducibility can be checked.
  2. [Table I] The "Accuracy %" column needs a precise definition: accuracy against which reference (human review, original Food-11 labels, or something else), how the human review was conducted, and whether the human reviewers saw CLIP's category suggestions.
  3. [Eq. (15) and Appendix A.2] Please clarify which binarization method was actually used in the experiments: Eq. (15) defines a fixed threshold θ, while Appendix A.2 mentions Otsu's method and adaptive RL-learned thresholds.
  4. [Section III.E] The natural transformation diagram is not rendered correctly in the text; please typeset the commutative diagram properly and define the objects and arrows in the diagram explicitly.
  5. [Table V] Table V shows only three of the ten Food-MNIST main categories; label it as a representative excerpt and include the full table in a supplement if space is limited.
  6. [Section IV.E] The phrase "near-optimal class distribution entropy" should be quantified, for example by reporting the actual entropy value relative to the uniform distribution.
  7. [Section II.C and Section III.C–E] The category-theoretic formalism is not connected to any measurable prediction or experimental outcome in the paper; consider stating explicitly what explanatory work the formalism does, or presenting it more briefly as a conceptual framing.

Circularity Check

2 steps flagged · score 6.0 of 10

The reported RL-driven 5-8% accuracy gain is an optimized term of the paper's own reward function, and the 89.7% categorization accuracy is validated only against human-review labels made with the same CLIP-based suggestions; both are partial self-validation.

  1. fitted input called prediction [Section III.I, Eq. (16); Section IV.E]
    "R(s, a) =λ1 · SemanticConf(s) + λ2 · Entropy(ClassDist) + λ3 · ModelAcc − λ4 · Redundancy Where SemanticConf is CLIP-based semantic confidence score, ClassDist is current class distribution entropy, ModelAcc is accuracy from a lightweight model trained on current batch"

    The DQN agent is trained to maximize this reward, so the λ3·ModelAcc term directly rewards sample selections that raise the accuracy of a classifier trained on the current batch. Section IV.E then reports "The RL + Semantic version showed a 5–8% improvement in classification accuracy" as evidence of dataset quality. The evaluation metric is therefore the same objective the agent was optimized against, so the reported gain is a designed optimization outcome, not an independent validation of filtering. The final CNN is a different model from the lightweight reward classifier, but the target quantity—classification accuracy on the filtered batch—is built into Eq. (16) by construction.

  2. self definitional [Section III.G(a); Section V; Table I]
    "For maximum control, each image is reviewed with AI predictions, confidence scores, and comprehensive category suggestions."

    Table I reports Smart Batch "Accuracy %" of 89.7 as the headline categorization accuracy. Section V concedes "there is no definitive ground truth against which to benchmark performance," so this figure is an agreement rate with human-review labels. Those labels are produced in the same review interface that presents CLIP-based "AI predictions, confidence scores, and comprehensive category suggestions" from the categorizer being evaluated. The reference is therefore not independent: reviewers are primed by the model's own suggestions, and agreement with a reviewer shown the model's answer is not evidence of true label correctness. Food-11 (Section IV.A) is a curated labeled source that could provide an external check, but no comparison against Food-11 labels is reported.

full rationale

The strongest load-bearing step is in Eq. (16): the RL reward includes ModelAcc, the accuracy of a lightweight model trained on the current batch, and Section IV.E uses model-accuracy improvement as evidence that RL filtering helps; this reduces that reported 5-8% gain to the optimization objective itself (fitted_input_called_prediction). The second step is the evaluation of categorization accuracy: Section V explicitly disclaims any definitive ground truth, yet Table I reports 89.7% accuracy, and the human-review labels used as the reference are generated in an interface showing the same AI predictions and confidence scores (Section III.G); this is a self-referential validation, not an external benchmark. Section IV.A states that Food-11 images were retrieved "to compare and validate subcategory labeling," which would be an independent ground truth, but no comparison against Food-11 labels appears in the paper, so the one external check is acknowledged and then unused. The 80% time-saving comparison (25.0 vs 5.2 hours, Table I) and the downstream classifier scores on the generated datasets are not circular. Self-citations appear in the references but none is load-bearing for the core framework; the category-theoretic formalization is descriptive rather than derivational. Overall: partial circularity, score 6.

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

The central empirical claims rest on three unverified legs: CLIP scores stand in for ground truth, ambiguous images are excluded, and the RL reward uses the same model accuracy later cited as evidence. No new physical or mathematical entities are introduced.

free parameters (5)
  • Score weights alpha, beta, gamma = not reported
    Equation (4) computes semantic similarity as a weighted sum; the weights are not specified or fitted to any external benchmark, so the categorization scores depend on unstated choices.
  • Reward weights lambda1..lambda4 = 0.4, 0.3, 0.2, 0.1
    Appendix A.2 specifies these values by hand; they directly determine which samples the RL agent keeps.
  • CLIP filtering similarity threshold = 0.3
    Appendix A.2; threshold for accepting retrieved images, chosen by hand.
  • Smart batch confidence thresholds = 0.85 and 0.4
    Section III.G(b); these thresholds determine the 85% auto-categorization rate claim.
  • Binarization threshold theta = manually set or RL-learned, value not reported
    Equation (15); choice affects final image content and downstream classification.
assumptions (4)
  • domain assumption CLIP embedding similarity is a valid proxy for fine-grained visual category membership
    Equation (4) and Section III.A.2 use TextSim and CharSim to score images against subcategories; the paper states in Section V that no ground truth exists, so this proxy is load-bearing.
  • ad hoc to paper Removing visually ambiguous multi-object images does not bias the reported accuracy
    Section IV.A excludes images with multiple food types to avoid labeling conflicts; this exclusion can inflate both categorization and downstream accuracy.
  • ad hoc to paper RL reward including ModelAcc on the current training batch is a valid signal for dataset quality
    Equation (16) includes lambda3 * ModelAcc, where ModelAcc comes from a lightweight model trained on the current batch; this creates a feedback loop with the later evaluation.
  • domain assumption Preprocessing functions are pure and deterministic, so functoriality holds
    Appendix C.1 'proof' assumes pure deterministic transformations to conclude F(g∘f)=F(g)∘F(f); this is an implementation assumption, not a mathematical theorem.

how reviews work

0 comments
Cite this review

Pith. "Pith review of MNIST-Gen: A Modular MNIST-Style Dataset Generation Using Hierarchical Semantics, Reinforcement Learning, and Category Theory." pith.science (2026). https://pith.science/paper/JXL77V2G

@misc{pith2026250711821,
  author       = {Pith},
  title        = {Pith review of: MNIST-Gen: A Modular MNIST-Style Dataset Generation Using Hierarchical Semantics, Reinforcement Learning, and Category Theory},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/JXL77V2G}},
  note         = {Machine review of arXiv:2507.11821}
}
read the original abstract

Neural networks are often benchmarked using standard datasets such as MNIST, FashionMNIST, or other variants of MNIST, which, while accessible, are limited to generic classes such as digits or clothing items. For researchers working on domain-specific tasks, such as classifying trees, food items, or other real-world objects, these data sets are insufficient and irrelevant. Additionally, creating and publishing a custom dataset can be time consuming, legally constrained, or beyond the scope of individual projects. We present MNIST-Gen, an automated, modular, and adaptive framework for generating MNIST-style image datasets tailored to user-specified categories using hierarchical semantic categorization. The system combines CLIP-based semantic understanding with reinforcement learning and human feedback to achieve intelligent categorization with minimal manual intervention. Our hierarchical approach supports complex category structures with semantic characteristics, enabling fine-grained subcategorization and multiple processing modes: individual review for maximum control, smart batch processing for large datasets, and fast batch processing for rapid creation. Inspired by category theory, MNIST-Gen models each data transformation stage as a composable morphism, enhancing clarity, modularity, and extensibility. As proof of concept, we generate and benchmark two novel datasets-\textit{Tree-MNIST} and \textit{Food-MNIST}-demonstrating MNIST-Gen's utility for producing task-specific evaluation data while achieving 85\% automatic categorization accuracy and 80\% time savings compared to manual approaches.

Figures

Figures reproduced from arXiv: 2507.11821 by the authors.

Figure 1
Figure 1. Overview of the MNIST-Gen workflow. The pipeline includes keyword-based image retrieval, hierarchical semantic clustering via CLIP embeddings, [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Sample images from the Tree-MNIST dataset, illustrating morpho [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗
Figure 3
Figure 3. Sample images from the Food-MNIST dataset, illustrating visual [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

31 extracted references · 27 canonical work pages

  1. [30]

    Food/non-food image classifi- cation and food categorization using pre-trained googlenet model,

    A. Singla, L. Yuan, and T. Ebrahimi, “Food/non-food image classifi- cation and food categorization using pre-trained googlenet model,” in Proceedings of the 2nd international workshop on multimedia assisted dietary management, 2016, pp. 3–11. APPENDIX A. Implementation Details A.1 System Architecture and Dependencies: MNIST-Gen is implemented in Python 3....

    work page 2016

  2. [1]

    Gradient-based learning applied to document recognition,

    Y . LeCun, L. Bottou, Y . Bengio, and P. Haffner, “Gradient-based learning applied to document recognition,” Proceedings of the IEEE , vol. 86, no. 11, pp. 2278–2324, 1998

    1998

  3. [2]

    Fashion-mnist: a novel image dataset for benchmarking machine learning algorithms,

    H. Xiao, K. Rasul, and R. V ollgraf, “Fashion-mnist: a novel image dataset for benchmarking machine learning algorithms,” arXiv preprint arXiv:1708.07747, 2017

    arXiv 2017

  4. [3]

    Revisiting unreasonable effectiveness of data in deep learning era,

    C. Sun, A. Shrivastava, S. Singh, and A. Gupta, “Revisiting unreasonable effectiveness of data in deep learning era,” in Proceedings of the IEEE international conference on computer vision , 2017, pp. 843–852

    work page 2017

  5. [4]

    Food-101–mining dis- criminative components with random forests,

    L. Bossard, M. Guillaumin, and L. Van Gool, “Food-101–mining dis- criminative components with random forests,” in European conference on computer vision . Springer, 2014, pp. 446–461

    work page 2014

  6. [5]

    Fine-grained plant classification using convolutional neural networks for feature extraction

    N. S ¨underhauf, C. McCool, B. Upcroft, and T. Perez, “Fine-grained plant classification using convolutional neural networks for feature extraction.” in CLEF (Working notes), 2014, pp. 756–762

    work page 2014

  7. [6]

    MID-L: Matrix-Interpolated Dropout Layer with Layer-wise Neuron Selection

    P. Shaeri and A. Middel, “Mid-l: Matrix-interpolated dropout layer with layer-wise neuron selection,” 2025. [Online]. Available: https://arxiv.org/abs/2505.11416

    work page Pith review arXiv 2025

  8. [7]

    Learning visual features from large weakly supervised data,

    A. Joulin, L. Van Der Maaten, A. Jabri, and N. Vasilache, “Learning visual features from large weakly supervised data,” in European confer- ence on computer vision . Springer, 2016, pp. 67–84

    work page 2016

Show all 31 references
  1. [8]

    A semi-supervised fake news detection using sentiment encoding and lstm with self-attention,

    P. Shaeri and A. Katanforoush, “A semi-supervised fake news detection using sentiment encoding and lstm with self-attention,” in 2023 13th International Conference on Computer and Knowledge Engineering (ICCKE). IEEE, 2023, pp. 590–595

    2023

  2. [9]

    A new chebyshev operational matrix formulation of least-squares support vector regression for solving fractional integro-differential equations,

    A. A. Aghaei, A. G. Nejad, S. Yousefi, and K. Parand, “A new chebyshev operational matrix formulation of least-squares support vector regression for solving fractional integro-differential equations,” Journal of Computational and Applied Mathematics , p. 116816, 2025

    2025

  3. [10]

    A machine learning framework for efficiently solving fokker–planck equations,

    A. N. Firoozsalari, A. A. Aghaei, and K. Parand, “A machine learning framework for efficiently solving fokker–planck equations,” Computa- tional and Applied Mathematics , vol. 43, no. 6, p. 389, 2024

    2024

  4. [11]

    Emnist: Extending mnist to handwritten letters,

    G. Cohen, S. Afshar, J. Tapson, and A. van Schaik, “Emnist: Extending mnist to handwritten letters,” in 2017 International Joint Conference on Neural Networks (IJCNN) . IEEE, 2017, pp. 2921–2926

    2017

  5. [12]

    Sentiment and social signals in the climate crisis: A survey on analyzing social media responses to extreme weather events,

    P. Shaeri, Y . Mohammadpour, A. Beigi, A. Middel, and H. Liu, “Sentiment and social signals in the climate crisis: A survey on analyzing social media responses to extreme weather events,” 2025. [Online]. Available: https://arxiv.org/abs/2504.18837

    2025

  6. [13]

    Can llms improve multimodal fact-checking by asking relevant questions?

    A. Beigi, B. Jiang, D. Li, Z. Tan, P. Shaeri, T. Kumarage, A. Bhattachar- jee, and H. Liu, “Can llms improve multimodal fact-checking by asking relevant questions?” arXiv preprint arXiv:2410.04616 , 2024

    2024 arXiv

  7. [14]

    A multimodal physics-informed neural network approach for mean radiant temperature modeling,

    P. Shaeri, S. AlKhaled, and A. Middel, “A multimodal physics-informed neural network approach for mean radiant temperature modeling,”

  8. [15]

    Webmrt: An online tool to predict summertime mean radiant tempera- ture using machine learning,

    S. R. AlKhaled, A. Middel, P. Shaeri, I. Buo, and F. A. Schneider, “Webmrt: An online tool to predict summertime mean radiant tempera- ture using machine learning,” Sustainable Cities and Society , vol. 115, p. 105861, 2024

    2024

  9. [16]

    Urban form and composition of street canyons: A human-centric big data and deep learning approach,

    A. Middel, J. Lukasczyk, S. Zakrzewski, M. Arnold, and R. Maciejew- ski, “Urban form and composition of street canyons: A human-centric big data and deep learning approach,” Landscape and Urban Planning , vol. 183, pp. 122–132, 2019

    2019

  10. [17]

    Snorkel: Rapid training data creation with weak supervision,

    A. Ratner, S. Bach, H. Ehrenberg, J. Fries, S. Wu, and C. R ´e, “Snorkel: Rapid training data creation with weak supervision,” The VLDB Journal, vol. 29, no. 2, pp. 709–730, 2020

    2020

  11. [18]

    Label studio: Open-source data labeling tool,

    Heartex, “Label studio: Open-source data labeling tool,” https:// labelstud.io/, 2020

    2020

  12. [19]

    Google automl: cloud vision,

    E. Bisong, “Google automl: cloud vision,” in Building Machine Learning and Deep Learning Models on Google Cloud Platform: A Comprehen- sive Guide for Beginners . Springer, 2019, pp. 581–598

    2019

  13. [20]

    Nvidia tao toolkit,

    “Nvidia tao toolkit,” https://developer.nvidia.com/tao-toolkit, accessed July 2025

    2025

  14. [21]

    A survey of data collection methods for machine learning,

    H. Kaur, A. Malhi, and S. Kumar, “A survey of data collection methods for machine learning,” Journal of Big Data , vol. 8, no. 1, pp. 1–32, 2021

    2021

  15. [22]

    Docparsenet: Advanced semantic segmentation and ocr embeddings for efficient scanned document annotation,

    A. Mohammadshirazi, A. N. Firoozsalari, M. Zhou, D. Kulshrestha, and R. Ramnath, “Docparsenet: Advanced semantic segmentation and ocr embeddings for efficient scanned document annotation,” arXiv preprint arXiv:2406.17591, 2024

    2024 arXiv

  16. [23]

    Learning transferable visual models from natural language supervision,

    A. Radford, J. W. Kim, C. Hallacy, A. Ramesh, G. Goh, S. Agarwal, G. Sastry, A. Askell, P. Mishkin, J. Clark et al., “Learning transferable visual models from natural language supervision,” in International conference on machine learning . PmLR, 2021, pp. 8748–8763

    2021

  17. [24]

    Fong and D

    B. Fong and D. I. Spivak, An Invitation to Applied Category Theory: Seven Sketches in Compositionality. Cambridge University Press, 2019

    2019

  18. [25]

    Level- k reasoning, deep rein- forcement learning, and monte carlo decision process for fast and safe automated lane change and speed management,

    S. Karimi, A. Karimi, and A. Vahidi, “Level- k reasoning, deep rein- forcement learning, and monte carlo decision process for fast and safe automated lane change and speed management,” IEEE Transactions on Intelligent Vehicles, vol. 8, no. 6, pp. 3556–3571, 2023

    2023

  19. [26]

    Learning to reweight ex- amples for robust deep learning,

    M. Ren, W. Zeng, B. Yang, and R. Urtasun, “Learning to reweight ex- amples for robust deep learning,” International Conference on Machine Learning (ICML), 2018

    2018

  20. [27]

    Deliberate practice in data selection for efficient learning,

    L. Jiang, L. Mou, Z. Li, and G. Neubig, “Deliberate practice in data selection for efficient learning,” NeurIPS, 2021

    2021

  21. [28]

    Adaptive data selection for labeling via reinforcement learning,

    T. Wu, S. Ma, D. Dey, H. Zhang, T. Mitra, X. Geng, and Z. Yu, “Adaptive data selection for labeling via reinforcement learning,” ACL, 2023

    2023

  22. [29]

    U2-net: Going deeper with nested u-structure for salient object detection,

    X. Qin, Z. Zhang, C. Huang, M. Dehghan, O. R. Zaiane, and M. Jager- sand, “U2-net: Going deeper with nested u-structure for salient object detection,” Pattern recognition, vol. 106, p. 107404, 2020

    2020

  23. [2025]

    Available: https://arxiv.org/abs/2503.08482

    [Online]. Available: https://arxiv.org/abs/2503.08482

    arXiv

Pith tools

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