Pith. sign in

REVIEW 5 major objections 5 minor 103 references

AMCR: A Framework for Assessing and Mitigating Copyright Risks in Generative Models

T0 review · 5 major / 5 minor · reviewed 2026-08-05 · deepseek-v4-flash

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

Pith's one-line read AMCR: a diffusion-pipeline framework that sanitizes prompts, detects partial copyright infringement via attention-weighted trajectory comparison, and mitigates it during generation.

desk verdict A well-intended pipeline for copyright-safe text-to-image generation, but the central evidence for its detection and mitigation claims doesn't hold up as written. read the letter →

arxiv 2509.00641 v1 pith:OQAYGFLW submitted 2025-08-31 cs.LG cs.CRcs.CV

Zhipeng Yin , Zichong Wang , Avash Palikhe , Zhen Liu , Jun Liu , Wenbin Zhang This is my paper
classification cs.LGcs.CRcs.CV
keywords copyrightinfringementdiffusionmodelstext-to-imagegenerationpromptsanitizationattentionmapslatentimagesimilaritydetectionAIsafety
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 tries to establish that hidden copyright infringement in text-to-image models can be caught and removed during generation, not just filtered at the prompt. It proposes AMCR, a three-stage wrapper on latent diffusion models: rewriting risky prompts into semantically similar safe ones, detecting partial infringement by comparing attention-weighted local features across diffusion trajectories, and fine-tuning during generation with a risk loss that balances infringement reduction against image quality. The authors show their detector outperforms global similarity metrics such as L2, LPIPS, and SSCD on the D-Rep and LAION-5B datasets, and visually show outputs that no longer reproduce Mario, Apple's logo, or Lakers jerseys. If correct, AMCR gives deployers a practical end-to-end pipeline for safer generation without changing the base model.

What carries the argument

The load-bearing mechanism is two-trajectory alignment inside the latent diffusion process. At every diffusion step, the generated latent and the reference latent are denoised and compared through cross-attention-derived soft masks; patch-level CLIP embeddings are weighted by these masks and scored with a log-sum-exp cosine similarity. The same similarity score doubles as the mitigation objective, making detection and avoidance share one differentiable signal.

What would settle it

Take a set of images with localized copyrighted overlays (a logo in a small patch), run AMCR's latent-space CLIP detector, and separately run CLIP on the decoded RGB versions of the same latents; if the latent-space scores fail to rank cases the RGB CLIP scores and human annotators agree on, the partial-similarity signal is an artifact of feeding latents out of distribution.

Watch

Extended reading notes

Core claim

The paper claims that latent copyright infringement can be assessed and mitigated in three integrated stages. First, user prompts are parsed into semantic slots by an LLM, scored against a curated risk corpus using CLIP text embeddings, and risky phrases are replaced with alternatives that reduce risk while preserving meaning. Second, partial infringement is detected by aligning two diffusion trajectories: one from the sanitized prompt and one from the reference copyrighted image, then comparing attention-mask-weighted local CLIP embeddings of the denoised latents at matched timesteps, aggregated with a log-sum-exp similarity. Third, that same partial-similarity score is used as a differenti

Load-bearing premise

The whole approach assumes that CLIP embeddings computed on noisy latent-space images measure the same visual similarity as CLIP on ordinary natural images, even though CLIP was trained on RGB images.

Editorial extensions

If this is right

  • Deployers could wrap existing latent diffusion models with AMCR as an end-to-end copyright-safety layer, without retraining or swapping the base generator.
  • Partial infringements such as logos, character outfits, or small iconic objects would be detected even when the overall generated image looks unlike the reference.
  • Prompt sanitization preserves user intent by replacing only high-risk phrases with semantically aligned alternatives, rather than blocking prompts outright.
  • The mitigation loss balances infringement risk with generation quality, so safer outputs do not require sacrificing visual fidelity.
  • If the detection results hold, global similarity metrics like L2, LPIPS, and SSCD are insufficient benchmarks for copyright screening in practice.

Reading between the lines

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

  • The same trajectory-alignment trick could be reused for other localized similarity tasks, such as style-transfer attribution or detecting whether a specific object appears in a generated scene, not just copyright infringement.
  • The prompt sanitizer could double as an automated risky-prompt generator: starting from a small seed set of protected entities, it could produce diverse benign-sounding prompts that still trigger infringement, giving deployers a stress-test suite.
  • Because mitigation is done through LoRA fine-tuning, the risk-avoidance behavior could in principle be customized per copyright holder's reference image set, turning AMCR into an updatable wrapper rather than a one-time retraining.
  • The legal framing assumes the set of protected references is known in advance, which fits litigation workflows but leaves open the harder proactive-filtering scenario where the copyrighted work is not specified ahead of time.
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

5 major / 5 minor

Summary. The paper introduces AMCR, a three-stage framework for copyright risk in text-to-image latent diffusion models. Stage 1 sanitizes user prompts by parsing them into semantic slots, scoring slot text against a curated risk corpus with CLIP-text embeddings, and replacing high-risk elements with LLM-generated candidates. Stage 2 detects partial infringement by aligning a generated-image diffusion trajectory with a reference-image trajectory, extracting cross-attention masks, and computing a log-sum-exp cosine similarity between CLIP patch embeddings weighted by the mask. Stage 3 fine-tunes the diffusion model with a combined loss consisting of the standard v-prediction loss, the stage-2 similarity as a risk loss, and a CLIP-based semantic consistency loss. Experiments report detection accuracy/F1 on D-Rep and LAION-5B relative to L2/LPIPS/SSCD and qualitative mitigation examples against SDXL, DALL-E, and Midjourney.

Significance. If the claims were fully supported, AMCR would be a practically useful end-to-end wrapper for reducing copyright infringement in text-to-image generation. The paper's strengths are the choice of a human-labeled infringement benchmark (D-Rep), the emphasis on partial/localized similarity rather than global image distance, and the explicit combination of prompt-, detection-, and generation-time mitigation in one pipeline. The experimental setup is reproducible in principle (optimizer, batch size, LoRA, etc.). However, the current evidence is not sufficient: the detector is evaluated with an undefined threshold, the core CLIP embedding step is applied to latent-space representations without justification, and the mitigation results are qualitative and partially circular. These issues are load-bearing for all three claimed contributions.

major comments (5)
  1. [Section 4.2, Eq. (12)] The detector passes denoised latent images z_hat_0(t) and z_hat_0,r(t) into CLIP-ViT. CLIP-ViT is trained on normalized RGB patches, whereas LDM latents are 4-channel VAE representations with different statistics. No decoding or channel adaptation is described, so Eq. (12) is undefined as written. Since Eq. (13) and Eq. (15) are built from these embeddings, both the detection and mitigation claims rest on this step. The authors must either specify the decode-to-pixel step or provide evidence that CLIP embeddings of raw latents are valid similarity measures. Without this, Table 1 and Figure 4 do not support the central claims.
  2. [Section 5.3, Eq. (1)] The classification experiments require a threshold tau, but the paper never states how tau is selected for AMCR or for baselines. Accuracy/F1 in Table 1 are threshold-dependent. The reported improvements could be an artifact of threshold tuning. Report the threshold-selection protocol (e.g., validation set, operating point) and include confidence intervals or significance tests over multiple runs/random seeds.
  3. [Section 4.3 and Section 5.4] The mitigation loss L_r in Eq. (15) is exactly the detection similarity S_img from Eq. (13), up to weighting and expectation. Thus showing that AMCR reduces S_img after fine-tuning is true by construction. To demonstrate reduced actual infringement, the authors need an independent evaluation: e.g., human-annotated infringement rates on a held-out set or one of the baseline copy-detection metrics as an external check. Moreover, the claim that image quality is preserved is supported only by three qualitative examples (Figure 4); no quantitative quality metric (FID, LPIPS, CLIP-score) or user study is provided. Contribution (iii) is therefore not established.
  4. [Section 4.2, Eq. (11)] The reference trajectory relies on a 'minimally conditioned' base model v_base(cond_star). The paper does not define cond_star operationally (which tokens/embeddings, what conditioning) nor justify that this yields a neutral baseline. The choice may substantially affect patch alignments and therefore S_img. Provide a precise specification and a sensitivity analysis (e.g., varying cond_star or ablating the two-trajectory alignment).
  5. [Section 5.3] The comparison to L2/LPIPS/SSCD is asymmetric: AMCR uses internal attention maps, reference latents, and the diffusion model's weights, while baselines receive only the two images. This is a legitimate system-level comparison only if stated clearly. As reported, it conflates the value of the similarity measure with the value of extra access. Add an image-only AMCR variant or an explicit discussion of the comparison's scope. Also, the dataset is called 'D-Rep' in Section 5.1 and 'L-Rep' in Table 1; please correct the inconsistency.
minor comments (5)
  1. [Section 5.2] There is a duplicated phrase: 'which subtly which subtly references a well-known character (Mario)'.
  2. [Section 5.4] 'DALL·E 4o' appears to be a typo for a specific DALL-E version; please verify the model name.
  3. [Eq. (8)] The trade-off parameter lambda is not constrained; state its range and how it is chosen in practice.
  4. [Section 4.2, Figure 3] Figure 3 is not explicitly referenced in the text, and the caption 'How to train your dragon' is unclear without surrounding context.
  5. [Section 5.1] The dataset description says 'D-Rep dataset [47]' and '4,000 test images'; confirm the exact dataset name, version, and license, since reference [47] is about image copy detection for diffusion models.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: detection is externally benchmarked, mitigation uses the detector score as an explicit control objective rather than as evidence, and self-citations are not load-bearing.

full rationale

The paper is not circular in its central claims. The partial-infringement detector is defined independently in Section 4.2 (Eqs. 12-13) as a CLIP-based, attention-mask-weighted similarity score, and its detection performance is evaluated against external, human-annotated D-Rep and LAION-5B benchmarks in Table 1; this provides independent grounding for the detection claim. The mitigation objective in Eq. 15 is indeed exactly the detector similarity Simg from Eq. 13, up to the timestep weighting wr(t). This is a deliberate control-loop design, not a hidden equivalence between input and output. The paper does not claim to validate mitigation by showing that Simg fell; Section 5.4 instead provides qualitative visual comparisons against SDXL, DALL-E, and Midjourney, showing removal or modification of identifiable copyrighted elements. That evidence is independent of the detector metric. The prompt-sanitization stage (Eqs. 3-8) is also self-contained, using CLIP text embeddings, an externally maintained risk corpus, and LLM candidates. The many self-citations in the reference list are contextual or related-work citations and are not used as a load-bearing uniqueness theorem, ansatz, or definitional premise. The noted issue that Eq. 12 feeds denoised latent representations into CLIP-ViT, a model trained on RGB images, is a substantive validity/correctness concern about whether the similarity scores are meaningful in latent space, but it is not a circularity: the paper's equations do not reduce to each other by construction in a way that makes its claims true by definition. Thus no circular steps are identified.

Assumptions & free parameters 6 free parameters · 5 assumptions · 0 invented entities

The framework rests on several unstated assumptions: CLIP similarity as a legal proxy, attention maps as infringement localizers, the neutrality of the reference trajectory, the validity of the benchmarks, and quality preservation. Many hyperparameters (lambda, beta, tau, lambda_r, lambda_a, weighting functions, stopping criteria) are left unspecified, making the method difficult to evaluate or reproduce.

free parameters (6)
  • lambda (candidate replacement trade-off) = unknown
    Eq 8 balances risk reduction vs semantic alignment in replacement selection; no value given.
  • beta (LSE sharpness) = unknown
    Eq 13 defines Simg with beta controlling sensitivity to the most similar patch; used in detection and mitigation; value not reported.
  • tau (infringement threshold) = unknown
    Eq 1 defines infringement as similarity above tau; Table 1 accuracy/F1 depend on tau but its selection is not described.
  • lambda_r and lambda_a = unknown
    Eq 17 balances generative loss, risk loss, and alignment loss; values not reported.
  • attention aggregation weights = unknown
    Section 4.2 says 'using learned or uniform weights' to aggregate attention maps; not specified.
  • replacement budget, marginal improvement gamma, risk quantile = unknown
    Stopping criteria for prompt sanitization (Section 4.1); no values given.
assumptions (5)
  • domain assumption CLIP embeddings capture legal 'substantial similarity' for copyright infringement.
    The whole detection and mitigation depend on cosine similarity in CLIP space being a valid proxy for infringement (Eqs 4 and 13).
  • domain assumption Cross-attention maps localize infringing visual regions.
    Section 4.2 derives soft masks from attention maps and uses them to weight the generated image before comparison.
  • ad hoc to paper The two-trajectory alignment with a minimally conditioned base model is a meaningful neutral reference.
    Eq 11 defines the reference trajectory with minimal conditioning; no evidence that this baseline is neutral or comparable.
  • domain assumption D-Rep and LAION-5B provide valid ground-truth infringement labels.
    Section 5.1 uses D-Rep human similarity scores >= 4 as infringement and LAION-5B as a test set, but the LAION protocol is not described.
  • domain assumption Fine-tuning with the composite loss preserves generative quality.
    Eq 17 claims image quality is preserved, but no quality metric is reported.

how reviews work

0 comments
Cite this review

Pith. "Pith review of AMCR: A Framework for Assessing and Mitigating Copyright Risks in Generative Models." pith.science (2026). https://pith.science/paper/OQAYGFLW

@misc{pith2026250900641,
  author       = {Pith},
  title        = {Pith review of: AMCR: A Framework for Assessing and Mitigating Copyright Risks in Generative Models},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/OQAYGFLW}},
  note         = {Machine review of arXiv:2509.00641}
}
read the original abstract

Generative models have achieved impressive results in text to image tasks, significantly advancing visual content creation. However, this progress comes at a cost, as such models rely heavily on large-scale training data and may unintentionally replicate copyrighted elements, creating serious legal and ethical challenges for real-world deployment. To address these concerns, researchers have proposed various strategies to mitigate copyright risks, most of which are prompt based methods that filter or rewrite user inputs to prevent explicit infringement. While effective in handling obvious cases, these approaches often fall short in more subtle situations, where seemingly benign prompts can still lead to infringing outputs. To address these limitations, this paper introduces Assessing and Mitigating Copyright Risks (AMCR), a comprehensive framework which i) builds upon prompt-based strategies by systematically restructuring risky prompts into safe and non-sensitive forms, ii) detects partial infringements through attention-based similarity analysis, and iii) adaptively mitigates risks during generation to reduce copyright violations without compromising image quality. Extensive experiments validate the effectiveness of AMCR in revealing and mitigating latent copyright risks, offering practical insights and benchmarks for the safer deployment of generative models.

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

103 extracted references · 73 canonical work pages

  1. [1]

    S. K. Alhabeeb and A. A. Al-Shargabi. Text-to-image synthesis with generative models: Methods, datasets, performance metrics, chal- lenges, and future direction. IEEE Access, 12:24412–24427, 2024

    2024

  2. [2]

    A. Amon, Z. Yin, Z. Wang, A. Palikhe, and W. Zhang. Uncertain boundaries: Multidisciplinary approaches to copyright issues in gener- ative ai. ACM SIGKDD Explorations Newsletter, 2025

    2025

  3. [3]

    Anantrasirichai and D

    N. Anantrasirichai and D. Bull. Artificial intelligence in the creative industries: a review. Artificial intelligence review , 55(1):589–656, 2022

    2022

  4. [4]

    A. Borji. Generated faces in the wild: Quantitative compari- son of stable diffusion, midjourney and dall-e 2. arXiv preprint arXiv:2210.00586

    arXiv

  5. [5]

    Bourtoule, V

    L. Bourtoule, V . Chandrasekaran, C. A. Choquette-Choo, H. Jia, A. Travers, B. Zhang, D. Lie, and N. Papernot. Machine unlearning. In 2021 IEEE symposium on security and privacy (SP) , pages 141–159. IEEE, 2021

    2021

  6. [6]

    Y . Cai, D. Youngstrom, and W. Zhang. Exploring approaches for teaching cybersecurity and ai for k-12. In 2023 IEEE International Conference on Data Mining Workshops (ICDMW), pages 1559–1564. IEEE, 2023

    2023

  7. [7]

    Carlini, J

    N. Carlini, J. Hayes, M. Nasr, M. Jagielski, V . Sehwag, F. Tramer, B. Balle, D. Ippolito, and E. Wallace. Extracting training data from diffusion models. In 32nd USENIX Security Symposium (USENIX Se- curity 23), pages 5253–5270, 2023

    2023

  8. [8]

    Chiba-Okabe

    H. Chiba-Okabe. Probabilistic analysis of copyright disputes and gen- erative ai safety. arXiv preprint arXiv:2410.00475, 2024

    work page arXiv 2024

Show all 103 references
  1. [9]

    Chiba-Okabe and W

    H. Chiba-Okabe and W. J. Su. Tackling copyright issues in ai image generation through originality estimation and genericization.Scientific Reports, 15(1):10621, 2025

    2025

  2. [10]

    S. V . Chinta, K. Fernandes, N. Cheng, J. Fernandez, S. Yazdani, Z. Yin, Z. Wang, X. Wang, W. Xu, J. Liu, et al. Optimization and improve- ment of fake news detection using voting technique for societal benefit. In 2023 IEEE International Conference on Data Mining Workshops (I...

    2023

  3. [11]

    S. V . Chinta, Z. Wang, Z. Yin, N. Hoang, M. Gonzalez, T. L. Quy, and W. Zhang. Fairaied: Navigating fairness, bias, and ethics in educa- tional ai applications. arXiv preprint arXiv:2407.18745, 2024

    2024

  4. [12]

    S. V . Chinta, Z. Wang, A. Palikhe, X. Zhang, A. Kashif, M. A. Smith, J. Liu, and W. Zhang. Ai-driven healthcare: Fairness in ai healthcare: A survey. PLOS Digital Health, 4(5):e0000864, 2025

    2025

  5. [13]

    Z. Chu, S. Ni, Z. Wang, X. Feng, C. Li, X. Hu, R. Xu, M. Yang, and W. Zhang. History, development, and principles of large language models-an introductory survey. arXiv preprint arXiv:2402.06853 , 2024

    2024 arXiv

  6. [14]

    Z. Chu, Z. Wang, and W. Zhang. Fairness in large language models: A taxonomic survey. ACM SIGKDD Explorations Newsletter, 2024 , pages 34–48, 2024

    2024

  7. [15]

    T. V . Doan, Z. Wang, N. N. M. Hoang, and W. Zhang. Fairness in large language models in three hours. In Proceedings of the 33rd ACM In- ternational Conference on Information and Knowledge Management , pages 5514–5517, 2024

    2024

  8. [16]

    Dzuong, Z

    J. Dzuong, Z. Wang, and W. Zhang. Uncertain boundaries: Multidisci- plinary approaches to copyright issues in generative ai. arXiv preprint arXiv:2404.08221, 2024

    2024 arXiv

  9. [17]

    Guyet, W

    T. Guyet, W. Zhang, and A. Bifet. Incremental mining of fre- quent serial episodes considering multiple occurrences. In 22nd In- ternational Conference on Computational Science , pages 460–472. Springer, 2022

    2022

  10. [18]

    Hartwig, D

    S. Hartwig, D. Engel, L. Sick, H. Kniesel, T. Payer, P. Poonam, M. Glöckler, A. Bäuerle, and T. Ropinski. A survey on quality metrics for text-to-image generation. arXiv preprint arXiv:2403.11821, 2024

    2024 arXiv

  11. [19]

    C. M. Hayes. Generative artificial intelligence and copyright: Both sides of the black box. Available at SSRN 4517799, 2023

    2023

  12. [20]

    Joyce, T

    C. Joyce, T. T. Ochoa, M. W. Carroll, M. A. Leaffer, and P. Jaszi. Copyright law, volume 85. Carolina Academic Press Durham, NC, 2016

    2016

  13. [21]

    C. Kim, K. Min, M. Patel, S. Cheng, and Y . Yang. Wouaf: Weight modulation for user attribution and fingerprinting in text-to-image dif- fusion models. In Proceedings of the IEEE/CVF Conference on Com- puter Vision and Pattern Recognition, pages 8974–8983, 2024

    2024

  14. [22]

    T. P. V . Le. Copyright protection towards generative ai artworks: The “clash” between us v. china and the implications for the european union

  15. [23]

    H. Lee, P. Adamopoulos, V . G. Todri, and A. Ghose. The impact of generative ai on advertising effectiveness. 2024

    2024

  16. [24]

    H. Liao. Deepseek large-scale model: technical analysis and develop- ment prospect. Journal of Computer Science and Electrical Engineer- ing, 7(1):33–37, 2025

    2025

  17. [25]

    Z. Liu, R. Wang, N. Japkowicz, D. Tang, W. Zhang, and J. Zhao. Re- search on unsupervised feature learning for android malware detection based on restricted boltzmann machines.Future Generation Computer Systems, 120:91–108, 2021

    2021

  18. [26]

    Z. Liu, R. Wang, N. Japkowicz, H. M. Gomes, B. Peng, and W. Zhang. Segdroid: An android malware detection method based on sensitive function call graph learning. Expert Systems with Applications , page 121125, 2023

    2023

  19. [27]

    N. Lucchi. Chatgpt: a case study on copyright challenges for genera- tive artificial intelligence systems. European Journal of Risk Regula- tion, 15(3):602–624, 2024

    2024

  20. [28]

    J. Pallant. Creating Images Using AI: A Step-by-Step Guide to Mid- journey. CRC Press, 2024

    2024

  21. [29]

    Pizzi, S

    E. Pizzi, S. D. Roy, S. N. Ravindra, P. Goyal, and M. Douze. A self- supervised descriptor for image copy detection. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages 14532–14542, 2022

    2022

  22. [30]

    C. M. Poland. Generative ai and us intellectual property law. arXiv preprint arXiv:2311.16023, 2023

    2023 arXiv

  23. [31]

    Radford, J

    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. InInternational con- ference on machine learning, pages 8748–8763. PmLR, 2021

    2021

  24. [32]

    A. Rao, Y . Xiangli, Y . Guo, M. Tang, C. Meng, and M. Agrawala. Generative models for visual content editing and creation. In ACM SIGGRAPH 2024 Courses, pages 1–6. 2024

    2024

  25. [33]

    Saharia, W

    C. Saharia, W. Chan, H. Chang, C. Lee, J. Ho, T. Salimans, D. Fleet, and M. Norouzi. Palette: Image-to-image diffusion models. In ACM SIGGRAPH 2022 conference proceedings, pages 1–10, 2022

    2022

  26. [34]

    Samuelson

    P. Samuelson. Ongoing lawsuits could affect everyone who uses gen- erative ai. Science, 381:6654, 2023

    2023

  27. [35]

    Sandiumenge

    I. Sandiumenge. Copyright implications of the use of generative ai. Available at SSRN 4531912, 2023

    2023

  28. [36]

    N. A. Saxena, W. Zhang, and C. Shahabi. Missed opportunities in fair ai. In Proceedings of the 2023 SIAM International Conference on Data Mining (SDM), pages 961–964. SIAM, 2023

    2023

  29. [37]

    Schuhmann, R

    C. Schuhmann, R. Beaumont, R. Vencu, C. Gordon, R. Wightman, M. Cherti, T. Coombes, A. Katta, C. Mullis, M. Wortsman, et al. Laion-5b: An open large-scale dataset for training next generation image-text models. Advances in neural information processing sys- tems, 35, 2022

    2022

  30. [38]

    Somepalli, V

    G. Somepalli, V . Singla, M. Goldblum, J. Geiping, and T. Goldstein. Diffusion art or digital forgery? investigating data replication in diffu- sion models. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 6048–6058, 2023

    2023

  31. [39]

    Somepalli, V

    G. Somepalli, V . Singla, M. Goldblum, J. Geiping, and T. Goldstein. Understanding and mitigating copying in diffusion models. Advances in Neural Information Processing Systems, 36:47783–47803, 2023

    2023

  32. [40]

    X. Tang, W. Zhang, Y . Yu, K. Turner, T. Derr, M. Wang, and E. Ntoutsi. Interpretable visual understanding with cognitive attention network. In International Conference on Artificial Neural Networks , pages 555–

  33. [41]

    Thongmeensuk

    S. Thongmeensuk. Rethinking copyright exceptions in the era of gen- erative ai: Balancing innovation and intellectual property protection. The Journal of World Intellectual Property, 27(2):278–295, 2024

    2024

  34. [42]

    Tirumala, A

    K. Tirumala, A. Markosyan, L. Zettlemoyer, and A. Aghajanyan. Memorization without overfitting: Analyzing the training dynamics of large language models. volume 35

  35. [43]

    Vartiainen and M

    H. Vartiainen and M. Tedre. Using artificial intelligence in craft educa- tion: crafting with text-to-image generative models.Digital Creativity, 34(1):1–21, 2023

    2023

  36. [44]

    N. Vyas, S. M. Kakade, and B. Barak. On provable copyright protec- tion for generative models. In International conference on machine learning, pages 35277–35299. PMLR, 2023

    2023

  37. [45]

    S.-Y . Wang, A. A. Efros, J.-Y . Zhu, and R. Zhang. Evaluating data attribution for text-to-image models. In Proceedings of the IEEE/CVF International Conference on Computer Vision , pages 7192–7203, 2023

    2023

  38. [46]

    W. Wang, W. Zhang, Y . Sun, and Y . Yang. Bag of tricks and a strong baseline for image copy detection. arXiv preprint arXiv:2111.08004

    arXiv

  39. [47]

    W. Wang, Y . Sun, Z. Tan, and Y . Yang. Image copy detection for dif- fusion models. Advances in Neural Information Processing Systems , 37:14417–14456, 2024

    2024

  40. [48]

    Wang and W

    Z. Wang and W. Zhang. Group fairness with individual and censorship constraints. In 27th European Conference on Artificial Intelligence , 2024

    2024

  41. [49]

    Wang and W

    Z. Wang and W. Zhang. Fdgen: A fairness-aware graph generation model. In Proceedings of the 42nd International Conference on Ma- chine Learning. PMLR, 2025

    2025

  42. [50]

    Z. Wang, G. Narasimhan, X. Yao, and W. Zhang. Mitigating mul- tisource biases in graph neural networks via real counterfactual sam- ples. In 2023 IEEE International Conference on Data Mining (ICDM), pages 638–647. IEEE, 2023

    2023

  43. [51]

    Z. Wang, N. Saxena, T. Yu, S. Karki, T. Zetty, I. Haque, S. Zhou, D. Kc, I. Stockwell, A. Bifet, et al. Preventing discriminatory decision- making in evolving data streams. In Proceedings of the 2023 ACM Conference on Fairness, Accountability, and Transparency (FAccT) , 2023

    2023

  44. [52]

    Z. Wang, C. Wallace, A. Bifet, X. Yao, and W. Zhang. Fg2an: Fairness- aware graph generative adversarial networks. In Joint European Con- ference on Machine Learning and Knowledge Discovery in Databases, pages 259–275. Springer Nature Switzerland, 2023

    2023

  45. [53]

    Z. Wang, Y . Zhou, I. Haque, D. Lo, and W. Zhang. Towards fair machine learning software: Understanding and addressing model bias through counterfactual thinking. arXiv preprint arXiv:2302.08018 , 2023

    2023

  46. [54]

    Z. Wang, C. Chen, V . Sehwag, M. Pan, and L. Lyu. Evaluating and mitigating ip infringement in visual generative ai. arXiv preprint arXiv:2406.04662, 2024

    2024 arXiv

  47. [55]

    Z. Wang, Z. Chu, R. Blanco, Z. Chen, S.-C. Chen, and W. Zhang. Advancing graph counterfactual fairness through fair representation learning. In Joint European Conference on Machine Learning and Knowledge Discovery in Databases , pages 40–58. Springer Nature Switzerland, 2024

    2024

  48. [56]

    Z. Wang, Z. Chu, T. V . Doan, S. Ni, M. Yang, and W. Zhang. History, development, and principles of large language models-an introductory survey. AI and Ethics, 2024, 2024

    2024

  49. [57]

    Z. Wang, J. Dzuong, X. Yuan, Z. Chen, Y . Wu, X. Yao, and W. Zhang. Individual fairness with group awareness under uncertainty. In Joint European Conference on Machine Learning and Knowledge Discov- ery in Databases, pages 89–106. Springer Nature Switzerland, 2024

    2024

  50. [58]

    Z. Wang, A. Palikhe, Z. Yin, and W. Zhang. Fairness definitions in language models explained. arXiv preprint arXiv:2407.18454, 2024

    2024

  51. [59]

    Z. Wang, M. Qiu, M. Chen, M. B. Salem, X. Yao, and W. Zhang. Toward fair graph neural networks via real counterfactual samples. Knowledge and Information Systems, pages 1–25, 2024

    2024

  52. [60]

    Z. Wang, D. Ulloa, T. Yu, R. Rangaswami, R. Yap, and W. Zhang. Individual fairness with group constraints in graph neural networks. In 27th European Conference on Artificial Intelligence, 2024

    2024

  53. [61]

    Z. Wang, Z. Chu, T. Viet Doan, S. Wang, Y . Wu, V . Palade, and W. Zhang. Fair graph u-net: A fair graph learning framework inte- grating group and individual awareness. In proceedings of the AAAI conference on artificial intelligence, volume 39, pages 28485–28493, 2025

    2025

  54. [62]

    Z. Wang, N. Hoang, X. Zhang, K. Bello, X. Zhang, S. S. Iyengar, and W. Zhang. Towards fair graph learning without demographic informa- tion. In The 28th International Conference on Artificial Intelligence and Statistics, volume 258, pages 2107–2115, 2025

    2025

  55. [63]

    Z. Wang, F. Liu, S. Pan, J. Liu, F. Saeed, M. Qiu, and W. Zhang. fairgnn-wod: Fair graph learning without complete demographics. In Proceedings of the 34th International Joint Conference on Artificial Intelligence, 2025

    2025

  56. [64]

    Z. Wang, A. Wu, N. Moniz, S. Hu, B. Knijnenburg, Q. Zhu, and W. Zhang. Towards fairness with limited demographics via disentan- gled learning. In Proceedings of the 34th International Joint Confer- ence on Artificial Intelligence, 2025

    2025

  57. [65]

    Z. Wang, Z. Yin, F. Liu, Z. Liu, C. Lisetti, R. Yu, S. Wang, J. Liu, S. Ganapati, S. Zhou, et al. Graph fairness via authentic counterfactu- als: Tackling structural and causal challenges. ACM SIGKDD Explo- rations Newsletter, 26(2):89–98, 2025

    2025

  58. [66]

    Z. Wang, Z. Yin, A. Palikhe, and W. Zhang. Fairness in language models: A tutorial. In Proceedings of the 34th ACM International Conference on Information and Knowledge Management, 2025

    2025

  59. [67]

    Z. Wang, Z. Yin, L. Yang, J. Zhuang, R. Yu, Q. Kong, and W. Zhang. Fairness-aware graph representation learning with limited demo- graphic information. In Joint European Conference on Machine Learning and Knowledge Discovery in Databases . Springer Nature Switzerland, 2025

    2025

  60. [68]

    Z. Wang, Z. Yin, R. Yap, and W. Zhang. Ai fairness beyond complete demographics: Current achievements and future directions. In 28th European Conference on Artificial Intelligence, 2025

    2025

  61. [69]

    Z. Wang, Z. Yin, R. Yap, X. Zhang, S. Hu, and W. Zhang. Redefin- ing fairness: A multi-dimensional perspective and integrated evalua- tion framework. In Joint European Conference on Machine Learning and Knowledge Discovery in Databases. Springer Nature Switzerland, 2025

    2025

  62. [70]

    Z. Wang, Z. Yin, X. Zhang, Y . Zhang, X. He, S. Wang, H. Song, L. Yang, and W. Zhang. Towards fair graph-based machine learning software: unveiling and mitigating graph model bias. AI and Ethics , pages 1–18, 2025

    2025

  63. [71]

    Z. Wang, Z. Yin, Y . Zhang, L. Yang, T. Zhang, N. Pissinou, Y . Cai, S. Hu, Y . Li, L. Zhao, et al. Fg-smote: Towards fair node classification with graph neural network. ACM SIGKDD Explorations Newsletter , 26(2):99–108, 2025

    2025

  64. [72]

    Q. Xu, Z. Wang, X. He, L. Han, and R. Tang. Can large vision- language models detect images copyright infringement from genai? arXiv preprint arXiv:2502.16618, 2025

    2025 arXiv

  65. [73]

    Yazdani, N

    S. Yazdani, N. Saxena, Z. Wang, Y . Wu, and W. Zhang. A comprehen- sive survey of image and video generative ai: recent advances, variants, and applications. 2024

    2024

  66. [74]

    Z. Yin, S. Agarwal, A. Kashif, M. Gonzalez, Z. Wang, S. Liu, Z. Liu, Y . Wu, I. Stockwell, W. Xu, et al. Accessible health screening using body fat estimation by image segmentation. In 2024 IEEE Interna- tional Conference on Data Mining Workshops (ICDMW), pages 405– 414, 2024

    2024

  67. [75]

    Z. Yin, Z. Wang, and W. Zhang. Improving fairness in machine learn- ing software via counterfactual fairness thinking. In Proceedings of the 2024 IEEE/ACM 46th International Conference on Software Engi- neering: Companion Proceedings, pages 420–421, 2024

    2024

  68. [76]

    Z. Yin, Z. Wang, A. Palikhe, and W. Zhang. Uncertain boundaries: A tutorial on copyright challenges and cross-disciplinary solutions for generative ai. In Proceedings of the 34th ACM International Confer- ence on Information and Knowledge Management, 2025

    2025

  69. [77]

    Z. Yin, Z. Wang, W. Xu, J. Zhuang, P. Mozumder, A. Smith, and W. Zhang. Digital forensics in the age of large language models.arXiv preprint arXiv:2504.02963, 2025

    2025 arXiv

  70. [78]

    Zhang, K

    G. Zhang, K. Wang, X. Xu, Z. Wang, and H. Shi. Forget-me-not: Learning to forget in text-to-image diffusion models. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 1755–1764, 2024

    2024

  71. [79]

    Zhang, T

    H. Zhang, T. Nakamura, T. Isohara, and K. Sakurai. A review on ma- chine unlearning. SN Computer Science, 4(4):337, 2023

    2023

  72. [81]

    Zhang, Z

    J. Zhang, Z. Wang, A. Palikhe, Z. Yin, and W. Zhang. Datasets for fairness in language models: An in-depth survey, 2025. URL https: //arxiv.org/abs/2506.23411

    2025

  73. [82]

    Zhang, F

    M. Zhang, F. Zhang, J. Zhang, A. Chaddad, F. Guo, W. Zhang, J. Zhang, and A. Evans. Autoencoder for neuroimage. In Interna- tional conference on database and expert systems applications, pages 84–90. Springer, 2021

    2021

  74. [83]

    Zhang, P

    R. Zhang, P. Isola, A. A. Efros, E. Shechtman, and O. Wang. The unreasonable effectiveness of deep features as a perceptual metric. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 586–595, 2018

    2018

  75. [84]

    W. Zhang. Learning fairness and graph deep generation in dynamic environments. 2020

    2020

  76. [85]

    W. Zhang. Ai fairness in practice: Paradigm, challenges, and prospects. Ai Magazine, 45(3):386–395, 2024

    2024

  77. [86]

    W. Zhang. Fairness with censorship: Bridging the gap between fair- ness research and real-world deployment. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 38, pages 22685–22685, 2024

    2024

  78. [87]

    W. Zhang. Online and customizable fairness-aware learning. Knowl- edge and Information Systems, pages 1–28, 2025

    2025

  79. [88]

    Zhang and E

    W. Zhang and E. Ntoutsi. Faht: an adaptive fairness-aware decision tree classifier. In Proceedings of the 28th International Joint Confer- ence on Artificial Intelligence, pages 1480–1486, 2019

    2019

  80. [89]

    Zhang and J

    W. Zhang and J. Wang. Content-bootstrapped collaborative filtering for medical article recommendations. In IEEE International Confer- ence on Bioinformatics and Biomedicine (BIBM), 2018

    2018

  81. [90]

    Zhang and J

    W. Zhang and J. Weiss. Fair decision-making under uncertainty. In 2021 IEEE International Conference on Data Mining (ICDM). IEEE, 2021

    2021

  82. [91]

    Zhang and J

    W. Zhang and J. C. Weiss. Longitudinal fairness with censorship. In proceedings of the AAAI conference on artificial intelligence , vol- ume 36, pages 12235–12243, 2022

    2022

  83. [92]

    Zhang and J

    W. Zhang and J. C. Weiss. Fairness with censorship and group con- straints. Knowledge and Information Systems, pages 1–24, 2023

    2023

  84. [93]

    Zhang and L

    W. Zhang and L. Zhao. Online decision trees with fairness. arXiv preprint arXiv:2010.08146, 2020

    2010 arXiv

  85. [94]

    Zhang, J

    W. Zhang, J. Tang, and N. Wang. Using the machine learning ap- proach to predict patient survival from high-dimensional survival data. In IEEE International Conference on Bioinformatics and Biomedicine (BIBM), 2016

    2016

  86. [95]

    Zhang, J

    W. Zhang, J. Wang, D. Jin, L. Oreopoulos, and Z. Zhang. A determin- istic self-organizing map approach and its application on satellite data based cloud type classification. In IEEE International Conference on Big Data (Big Data), 2018

    2018

  87. [96]

    Zhang, X

    W. Zhang, X. Tang, and J. Wang. On fairness-aware learning for non- discriminative decision-making. In International Conference on Data Mining Workshops (ICDMW), pages 1072–1079, 2019

    2019

  88. [97]

    Zhang, A

    W. Zhang, A. Bifet, X. Zhang, J. C. Weiss, and W. Nejdl. Farf: A fair and adaptive random forests classifier. In Pacific-Asia Conference on Knowledge Discovery and Data Mining, pages 245–256. Springer, 2021

    2021

  89. [98]

    Zhang, L

    W. Zhang, L. Zhang, D. Pfoser, and L. Zhao. Disentangled dynamic graph deep generation. In Proceedings of the SIAM International Con- ference on Data Mining (SDM), pages 738–746, 2021

    2021

  90. [99]

    Zhang, S

    W. Zhang, S. Pan, S. Zhou, T. Walsh, and J. C. Weiss. Fairness amidst non-iid graph data: Current achievements and future directions. arXiv preprint arXiv:2202.07170, 2022

    2022 arXiv

  91. [100]

    Zhang, T

    W. Zhang, T. Hernandez-Boussard, and J. Weiss. Censored fairness through awareness. InProceedings of the AAAI conference on artificial intelligence, volume 37, pages 14611–14619, 2023

    2023

  92. [101]

    Zhang, Z

    W. Zhang, Z. Wang, J. Kim, C. Cheng, T. Oommen, P. Ravikumar, and J. Weiss. Individual fairness under uncertainty. In 26th European Conference on Artificial Intelligence, pages 3042–3049, 2023

    2023

  93. [102]

    Zhang, S

    W. Zhang, S. Zhou, T. Walsh, and J. C. Weiss. Fairness amidst non-iid graph data: A literature review. AI Magazine, 46(1):e12212, 2025

    2025

  94. [103]

    Zhang et al

    W. Zhang et al. Flexible and adaptive fairness-aware learning in non- stationary data streams. In IEEE 32nd International Conference on Tools with Artificial Intelligence (ICTAI), pages 399–406, 2020

    2020

  95. [104]

    Zhong, J

    H. Zhong, J. Chang, Z. Yang, T. Wu, P. C. Mahawaga Arachchige, C. Pathmabandu, and M. Xue. Copyright protection and accountability of generative ai: Attack, watermarking and attribution. In Companion Proceedings of the ACM Web Conference 2023, pages 94–98, 2023

    2023

Pith tools

Reviewed August 5, 2026 · model on record in the stance chip above.