Pith. sign in

REVIEW 4 major objections 5 minor 55 references

MATE: LLM-Powered Multi-Agent Translation Environment for Accessibility Applications

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

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

Pith's one-line read The paper claims that ModCon-Task-Identifier, a fine-tuned BERT model, recognizes modality conversion requests with 0.917 accuracy and 0.916 F1-score on the ModConTT dataset, outperforming GPT-3.5-Turbo and classical classifiers.

desk verdict A useful open-source accessibility system with a new but thinly documented synthetic dataset; the classifier's headline accuracy is real only on that distribution, so treat it as a proof-of-concept. read the letter →

arxiv 2506.19502 v2 pith:QBTP6NUQ submitted 2025-06-24 cs.MA cs.AIcs.LG

Aleksandr Algazinov , Matt Laing , Paul Laban This is my paper
classification cs.MAcs.AIcs.LG
keywords multi-agentsystemsaccessibilitymodalityconversiontasktyperecognitionfine-tunedBERTModConTTdatasetopen-sourceassistivetechnology
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 introduces MATE, an open-source multi-agent system that converts digital content (text, images, audio, video) into a format a user with disabilities can use. To route each request to the right conversion expert, the system must first recognize the type of modality conversion the user is asking for. The paper's central empirical claim is that a fine-tuned BERT model, called ModCon-Task-Identifier, performs this recognition with 0.917 accuracy and 0.916 F1-score on the authors' ModConTT dataset, substantially higher than GPT-3.5-Turbo (0.750 accuracy, 0.720 F1-score) and the best classical classifier (logistic regression on BERT embeddings, 0.783 accuracy, 0.779 F1-score). If the results transfer, a small model running locally could serve as the router in an accessibility assistant, keeping user data private. The paper also contributes the ModConTT benchmark of 600 AI-generated, human-verified prompts spanning ten task categories.

What carries the argument

The central object is the ModCon-Task-Identifier, a BERT model fine-tuned on the ModConTT dataset. It maps the user prompt $P_{user}$ to a task type $\hat{t} = \arg\max_{t \in T} P(t|P_{user}; \theta_M)$. It serves as the interpreter agent in the MATE architecture, deciding which of seven expert agents receives the input data. The expert agents then execute specific conversions, such as image-to-audio via image captioning plus text-to-speech, and audio-to-image via speech-to-text plus text-to-image.

What would settle it

Collect a set of natural language accessibility requests from real users (for instance, from accessibility forums or assistive technology support logs), label them by modality conversion type, and run ModCon-Task-Identifier on them; if accuracy drops substantially below the reported 0.917 on ModConTT, the synthetic dataset does not transfer.

Watch

Extended reading notes

Core claim

The paper claims that recognizing the intended modality conversion from a user's free-text request is best solved not by a large general-purpose LLM but by a task-specific fine-tuned BERT. On the ModConTT dataset, the fine-tuned model (ModCon-Task-Identifier) achieves accuracy 0.917 and F1-score 0.916, compared with 0.865/0.865 for GPT-3.5-Turbo, 0.835/0.833 for Llama-3.1-70B-Instruct, 0.774/0.774 for GLM-4-Flash, and 0.783/0.779 for the best classical baseline (logistic regression with BERT embeddings). The paper presents this as evidence that a specialized, locally runnable classifier can outperform much larger API-based models on this narrow routing task, and that the multi-agent architecture with an interpreter agent plus seven expert agents is justified.

Load-bearing premise

The load-bearing premise is that the ModConTT dataset, generated by LLMs and verified by humans, reflects the distribution of real accessibility requests well enough that the measured accuracy transfers to actual use.

Editorial extensions

If this is right

  • If ModCon-Task-Identifier generalizes, a user request like 'describe this image to me' can be routed automatically to the image-to-audio pipeline without asking the user to specify an output format.
  • The interpreter can be initialized with either a local fine-tuned model or an API LLM, so deployments can trade cost and privacy depending on hardware and sensitivity of the data.
  • The ModConTT dataset provides a 600-prompt benchmark for future accessibility-routing research, including task types the current system cannot yet execute, such as TTV and ATV.
  • Because the router is much smaller than GPT-3.5-Turbo, the full assistant can run locally on modest hardware, preserving privacy for sensitive inputs like medical documents.
  • The system's modular expert agents can be swapped or extended as better open-source conversion models appear, without changing the router.

Reading between the lines

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

  • The reported gap may shrink on real user queries, since the synthetic dataset is likely easier to classify than natural, noisy requests; a field trial with actual accessibility users would test this.
  • The same 'small router + specialized converters' pattern could be applied to other routing problems, such as choosing between translation, summarization, or question-answering for a user query.
  • The ModConTT labels include TTV and ATV that MATE cannot currently execute, so the dataset anticipates adding video-generation experts without retraining the router.
  • The paper does not report inter-annotator agreement on the human verification step; measuring that would indicate how reliable the benchmark labels are.
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 / 5 minor

Summary. The paper introduces MATE, an open-source multi-agent system that routes accessibility requests to modality-conversion expert agents (TTS, TTI, STT, ITT, ATI, ITA, VTT) after an interpreter agent classifies the user's request. The authors construct the ModConTT dataset of AI-generated, human-verified prompts, compare several LLM-based interpreter agents and classical classifiers, and introduce ModCon-Task-Identifier, a fine-tuned BERT model. On the second ModConTT version, ModCon-Task-Identifier reports accuracy 0.917 and F1-score 0.916, outperforming GPT-3.5-Turbo (0.750 accuracy) and classical classifiers. The paper also discusses limitations and future directions, including integration with healthcare and education.

Significance. If the ModConTT benchmark faithfully represents real accessibility requests, the paper's contribution is useful: an open-source, locally runnable accessibility framework with a small task-classification model would address privacy and customization concerns, and the released code, data, and Hugging Face model support reproducibility. The architecture is straightforward and the experimental comparison covers a reasonable set of baselines. However, the significance is contingent on dataset validity and on the fairness of the model comparison; the current evidence does not establish that the reported 0.917 accuracy transfers to real users, so the central claim should be read as a promising in-distribution result rather than a demonstrated state of the art for accessibility task recognition.

major comments (4)
  1. [3.2, ModConTT Dataset Creation] The dataset's fidelity to real accessibility requests is load-bearing but unverified. The paper says the dataset was 'AI-generated and human-verified', but it does not report which LLM(s) generated the prompts, the generation template or instructions, the human verification criteria, or inter-annotator agreement. Because the fine-tuned model is trained and evaluated on this same synthetic distribution, high accuracy may reflect formulaic phrasing rather than robust understanding. The sensitivity of results to the synthetic distribution is visible in the two dataset versions: GPT-3.5-Turbo scores 0.865 in Table 2 but only 0.750 in Table 3. Please provide dataset-generation details, verification statistics, and an evaluation on an external or real-user sample, or explicitly restrict the claims to the ModConTT distribution.
  2. [4.2, Evaluation of Task Classification Models] The comparison that supports the central claim is asymmetric. ModCon-Task-Identifier is fine-tuned on a training portion of the same ModConTT version used for testing, whereas GPT-3.5-Turbo (and the other LLMs) are evaluated zero-shot on that test set. A fine-tuned model usually has an inherent advantage in-distribution, so Table 3 does not support the statement that the model 'significantly outperforms' LLMs in a task-agnostic sense. Please compare against fine-tuned or few-shot LLM baselines, or clearly frame the result as trained versus zero-shot and discuss the distribution caveat.
  3. [4.2, Table 3] All classifier results are reported on a single 90/10 stratified split with no confidence intervals, significance tests, or repeated cross-validation. With only 60 test points (10% of 600), the difference between 0.917 and 0.783 is not enough to establish statistical significance without variance estimates. Please report confidence intervals, per-class results, or repeated-split results, and avoid the word 'significant' without a statistical test.
  4. [5.1, Limitations of the Study] The paper claims MATE is a useful accessibility assistant, but the system is never evaluated end-to-end. There are no user studies, no latency measurements, no evaluation of the expert agents' output quality, and no test of the full pipeline on real user requests. The limitations section acknowledges external model errors but does not quantify their impact. At minimum, add a clear statement that the system-level claims are architectural proposals rather than validated outcomes, and provide a small demonstration or pilot evaluation to support the accessibility claim.
minor comments (5)
  1. [4.1, Evaluation of Interpreter Agents] The text says 'The highest failure rate was observed for the UNK class (32%)' but Figure 2(d) shows the distribution of failures by class, not per-class failure rates. Per-class failure rates should be computed as failures in class k divided by total examples in class k, and the text should be reworded accordingly.
  2. [3.3, Model Training and Evaluation] The description of fine-tuning hyperparameters is incomplete: the learning rate, batch size, optimizer, and evaluation metric for early stopping are not reported. Please provide these details for reproducibility.
  3. [1, Introduction] The phrase 'numerical accessibility' in Section 2.1 appears to be a typo or nonstandard term; consider replacing it with a clearer phrase.
  4. [2.3, Modality Conversion] The subsection describes 'Audio-to-Video (ATV)' as decomposing into 'SST and Text-to-Video', but the earlier list uses 'STT' for speech-to-text. The shorthand 'SST' is not defined and conflicts with the paper's notation.
  5. [Abstract and 1, Introduction] The statement 'This is the first open-source and lightweight MAS built specifically for modality adaptation tasks' is a strong novelty claim that would benefit from a more systematic related-work comparison; the cited open-source MAS works (AutoGen, LangChain, OpenManus) are general frameworks rather than accessibility-specific baselines.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the reported 0.917 accuracy is an empirical result on held-out ModConTT data, not a construction or self-citation artifact.

full rationale

The paper's central empirical claim is explicitly scoped to its custom ModConTT dataset. Section 3.3 states: 'The data set was divided into train and test (90% and 10%, respectively)' and the fine-tuned ModCon-Task-Identifier is evaluated on the held-out test split, while LLM and classical baselines are scored on the same test prompts (Table 3). No equation in the paper defines accuracy or F1 in terms of the model's construction; the reported numbers are empirical outcomes of that held-out evaluation. The dataset was author-created and LLM-generated, but this is a data-validity concern, not circularity, because the test labels are not used to fit the model. The paper contains no load-bearing self-citations, no uniqueness theorem imported from the authors' prior work, and no fitted parameter renamed as a prediction. The limitations section acknowledges reliance on external models and missing video generation, which further indicates that the claims are not forced by definition.

Assumptions & free parameters 3 free parameters · 3 assumptions · 0 invented entities

The central claims rest on a synthetic dataset whose fidelity to real user requests is unverified, on external model reliability, and on a comparison design that favors the fine-tuned classifier. No new physical or theoretical entities are introduced.

free parameters (3)
  • Train/test split ratio = 90/10
    The classification results are based on a single random 90/10 split of the ModConTT dataset, with no cross-validation or multiple seeds reported.
  • BERT fine-tuning epochs = 6
    The fine-tuned BERT model was trained for up to 6 epochs with early stopping at 5 rounds; these hyperparameters affect the reported accuracy and were chosen by hand.
  • Class distribution in ModConTT = 50 per task class; 150 for UNK
    The dataset was constructed with 50 examples for each of the nine task types and 150 for the UNK class, a hand-chosen balance that affects both training and evaluation.
assumptions (3)
  • domain assumption The ModConTT dataset labels are correct and complete
    The dataset was AI-generated and human-verified, but the verification process is not described; the entire evaluation depends on label correctness.
  • domain assumption The underlying modality conversion models are reliable components
    MATE's outputs depend on external models such as Whisper, BLIP, Tacotron 2, and Stable Diffusion, which are assumed to work adequately in the integrated system.
  • domain assumption Zero-shot LLM evaluation is a meaningful baseline
    The paper compares a fine-tuned model against zero-shot LLMs; this choice affects the claimed outperformance.

how reviews work

0 comments
Cite this review

Pith. "Pith review of MATE: LLM-Powered Multi-Agent Translation Environment for Accessibility Applications." pith.science (2026). https://pith.science/paper/QBTP6NUQ

@misc{pith2026250619502,
  author       = {Pith},
  title        = {Pith review of: MATE: LLM-Powered Multi-Agent Translation Environment for Accessibility Applications},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/QBTP6NUQ}},
  note         = {Machine review of arXiv:2506.19502}
}
read the original abstract

Accessibility remains a critical concern in today's society, as many technologies are not developed to support the full range of user needs. Existing multi-agent systems (MAS) often cannot provide comprehensive assistance for users in need due to the lack of customization stemming from closed-source designs. Consequently, individuals with disabilities frequently encounter significant barriers when attempting to interact with digital environments. We introduce MATE, a multimodal accessibility MAS, which performs the modality conversions based on the user's needs. The system is useful for assisting people with disabilities by ensuring that data will be converted to an understandable format. For instance, if the user cannot see well and receives an image, the system converts this image to its audio description. MATE can be applied to a wide range of domains, industries, and areas, such as healthcare, and can become a useful assistant for various groups of users. The system supports multiple types of models, ranging from LLM API calling to using custom machine learning (ML) classifiers. This flexibility ensures that the system can be adapted to various needs and is compatible with a wide variety of hardware. Since the system is expected to run locally, it ensures the privacy and security of sensitive information. In addition, the framework can be effectively integrated with institutional technologies (e.g., digital healthcare service) for real-time user assistance. Furthermore, we introduce ModCon-Task-Identifier, a model that is capable of extracting the precise modality conversion task from the user input. Numerous experiments show that ModCon-Task-Identifier consistently outperforms other LLMs and statistical models on our custom data. Our code and data are publicly available at https://github.com/AlgazinovAleksandr/Multi-Agent-MATE.

Figures

Figures reproduced from arXiv: 2506.19502 by the authors.

Figure 1
Figure 1. MATE Workflow [48], and Video-LLaMA [52]. • Audio-to-Image (ATI). In this scenario, audio is converted to text, and then an image is created from the textual de￾scription. One of the effective ways to solve the Text-to￾Image (TTI) problem is to apply stable diffusion [35] models. While the separate conversion problems (for example, TTS) do not require the involvement of agents, a more complex system can benefit from… view at source ↗
Figure 2
Figure 2. Comparison of Model Performance and Failure Distribution Across Task Types [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

55 extracted references · 29 canonical work pages

  1. [1]

    Inclusion AI, Biao Gong, Cheng Zou, Chuanyang Zheng, Chunluan Zhou, Canx- iang Yan, Chunxiang Jin, Chunjie Shen, Dandan Zheng, Fudong Wang, Furong Xu, GuangMing Yao, Jun Zhou, Jingdong Chen, Jianxin Sun, Jiajia Liu, Jianjiang Zhu, Jun Peng, Kaixiang Ji, Kaiyou Song, Kaimeng Ren, Libin Wang, Lixiang Ru, Lele Xie, Longhua Tan, Lyuxin Xue, Lan Wang, Mochen B...

    arXiv 2025

  2. [2]

    Vignesh Amarnath and Research Pub. 2025. AI-DRIVEN MULTIMODAL COG- NITIVE SUPPORT: ADVANCING DIGITAL ACCESSIBILITY THROUGH ADAP- TIVE TECHNOLOGIES. INTERNATIONAL JOURNAL OF COMPUTER ENGI- NEERING & TECHNOLOGY 16 (02 2025), 2380–2393. https://doi.org/10.34218/ IJCET_16_01_170

    work page 2025

  3. [3]

    Alexei Baevski, Henry Zhou, Abdelrahman Mohamed, and Michael Auli. 2020. wav2vec 2.0: a framework for self-supervised learning of speech representations (NIPS ’20). Curran Associates Inc., Red Hook, NY, USA, Article 1044, 12 pages

    work page 2020

  4. [4]

    Christoph Bartneck, Christoph Lütge, Alan Wagner, and Sean Welsh. 2021. Au- tonomous Vehicles. Springer International Publishing, Cham, 83–92. https: //doi.org/10.1007/978-3-030-51110-4_10

    work page doi:10.1007/978-3-030-51110-4_10 2021

  5. [5]

    Edresson Casanova, Kelly Davis, Eren Gölge, Görkem Göknar, Iulian Gulea, Logan Hart, Aya Aljafari, Joshua Meyer, Reuben Morais, Samuel Olayemi, and Julian Weber. 2024. XTTS: a Massively Multilingual Zero-Shot Text-to-Speech Model. 4978–4982. https://doi.org/10.21437/Interspeech.2024-2016

    doi:10.21437/interspeech.2024-2016 2024

  6. [6]

    Harrison Chase. 2022. LangChain. https://github.com/langchain-ai/langchain

    2022

  7. [7]

    Khansa Chemnad and Achraf Othman. 2024. Digital accessibility in the era of artificial intelligence—Bibliometric analysis and systematic review. Frontiers in Artificial Intelligence 7 (2024), 1349668. https://doi.org/10.3389/frai.2024.1349668

    work page arXiv 2024

  8. [8]

    Hao Cui and Taha Yasseri. 2024. AI-enhanced collective intelligence. Patterns 5, 11 (Nov. 2024), 101074. https://doi.org/10.1016/j.patter.2024.101074

    arXiv 2024

Show all 55 references
  1. [9]

    Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. 2019. 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...

    2019 doi

  2. [10]

    Amanda Cardoso Duarte, Shruti Palaskar, Lucas Ventura, Deepti Ghadiyaram, Kenneth DeHaan, Florian Metze, Jordi Torres, and Xavier Giro -i-Nieto. 2021. How2Sign: A Large-Scale Multimodal Dataset for Continuous American Sign Language. In Proceedings of the IEEE/CVF Conference on...

    2021

  3. [11]

    Sen Fang, Chunyu Sui, Yanghao Zhou, Xuedong Zhang, Hongbin Zhong, Yapeng Tian, and Chen Chen. 2025. SignDiff: Diffusion Model for American Sign Lan- guage Production. arXiv:2308.16082 [cs.CV] https://arxiv.org/abs/2308.16082

    2025 arXiv

  4. [12]

    Daniele Giansanti and Antonia Pirrera. 2025. Integrating AI and Assistive Tech- nologies in Healthcare: Insights from a Narrative Review of Reviews. Healthcare 13, 5 (2025). https://doi.org/10.3390/healthcare13050556

    2025 doi

  5. [13]

    Team GLM, :, Aohan Zeng, Bin Xu, Bowen Wang, Chenhui Zhang, Da Yin, Dan Zhang, Diego Rojas, Guanyu Feng, Hanlin Zhao, Hanyu Lai, Hao Yu, Hongning Wang, Jiadai Sun, Jiajie Zhang, Jiale Cheng, Jiayi Gui, Jie Tang, Jing Zhang, Jingyu Sun, Juanzi Li, Lei Zhao, Lindong Wu, Lucen Zh...

    2024 arXiv

  6. [14]

    Yu Hao, Fan Yang, Hao Huang, Shuaihang Yuan, Sundeep Rangan, John-Ross Rizzo, Yao Wang, and Yi Fang. 2024. A Multi-Modal Foundation Model to Assist People with Blindness and Low Vision in Environmental Interaction. Journal of Imaging 10, 5 (2024). https://doi.org/10.3390/jimag...

    2024 doi

  7. [15]

    Mengkang Hu, Yuhang Zhou, Wendong Fan, Yuzhou Nie, Bowei Xia, Tao Sun, Ziyu Ye, Zhaoxuan Jin, Yingru Li, Qiguang Chen, et al. 2025. Owl: Optimized workforce learning for general multi-agent assistance in real-world task automa- tion. arXiv preprint arXiv:2505.23885 (2025)

    2025 arXiv

  8. [16]

    Antonia Karamolegkou, Malvina Nikandrou, Georgios Pantazopoulos, Danae Sanchez Villegas, Phillip Rust, Ruchira Dhar, Daniel Hershcovich, and Anders Søgaard. 2025. Evaluating Multimodal Language Models as Visual Assistants for Visually Impaired Users. arXiv:2503.22610 [cs.HC] h...

    2025 arXiv

  9. [17]

    Laxminarayana Korada, Vijay Sikha, and Dayakar Siramgari. 2024. AI & Ac- cessibility: A Conceptual Framework for Inclusive Technology. 12 (08 2024). https://doi.org/10.5281/zenodo.14279774 Aleksandr Algazinov, Matt Laing∗, and Paul Laban∗

    2024 doi

  10. [18]

    Yoach Lacombe, Vaibhav Srivastav, and Sanchit Gandhi. 2024. Parler-TTS. https: //github.com/huggingface/parler-tts

    2024

  11. [19]

    Junnan Li, Dongxu Li, Caiming Xiong, and Steven Hoi. 2022. BLIP: Bootstrapping Language-Image Pre-training for Unified Vision-Language Understanding and Generation. In Proceedings of the 39th International Conference on Machine Learn- ing (Proceedings of Machine Learning Resea...

    2022

  12. [20]

    Xinbin Liang, Jinyu Xiang, Zhaoyang Yu, Jiayi Zhang, Sirui Hong, Sheng Fan, and Xiao Tang. 2025. OpenManus: An open-source framework for building general AI agents. https://doi.org/10.5281/zenodo.15186407

    2025 doi

  13. [21]

    Rui Liu, Yu Shen, Peng Gao, Pratap Tokekar, and Ming Lin. 2025. CAML: Collaborative Auxiliary Modality Learning for Multi-Agent Systems. arXiv:2502.17821 [cs.RO] https://arxiv.org/abs/2502.17821

    2025

  14. [22]

    Dan Lyth and Simon King. 2024. Natural language guidance of high-fidelity text-to-speech with synthetic annotations. arXiv:2402.01912 [cs.SD]

    2024 arXiv

  15. [23]

    Manduzio, Federico A

    Graziano A. Manduzio, Federico A. Galatolo, Mario G. C. A. Cimino, Enzo Pasquale Scilingo, and Lorenzo Cominelli. 2024. Improving Small- Scale Large Language Models Function Calling for Reasoning Tasks. arXiv:2410.18890 [cs.AI] https://arxiv.org/abs/2410.18890

    2024 arXiv

  16. [24]

    Meta AI. 2024. Introducing LLaMA 3.1: Our most capable models to date. https: //ai.meta.com/blog/meta-llama-3-1/

    2024

  17. [25]

    Amit Moryossef, Mathias Müller, Anne Göhring, Zifan Jiang, Yoav Goldberg, and Sarah Ebling. 2023. An Open-Source Gloss-Based Baseline for Spoken to Signed Language Translation. InProceedings of the Second International Workshop on Automatic Translation for Signed and Spoken La...

    2023

  18. [26]

    Nasim Nezamoddini and Amirhosein Gholami. 2022. A Survey of Adaptive Multi-Agent Networks and Their Applications in Smart Cities. Smart Cities 5, 1 (2022), 318–347. https://doi.org/10.3390/smartcities5010019

    2022 doi

  19. [27]

    Engelmann, Davi Kniest, Renata Vieira, and Rafael H

    Juliana Damasio Oliveira, Debora C. Engelmann, Davi Kniest, Renata Vieira, and Rafael H. Bordini. 2022. Multi-Agent Interaction to Assist Visually-Impaired and Elderly People. International Journal of Environmental Research and Public Health 19, 15 (2022), 8945. https://doi.or...

    2022 doi

  20. [28]

    Gonzalez, Matei Zaharia, and Ion Stoica

    Melissa Z Pan, Mert Cemri, Lakshya A Agrawal, Shuyi Yang, Bhavya Chopra, Rishabh Tiwari, Kurt Keutzer, Aditya Parameswaran, Kannan Ramchandran, Dan Klein, Joseph E. Gonzalez, Matei Zaharia, and Ion Stoica. 2025. Why Do Multiagent Systems Fail?. In ICLR 2025 Workshop on Buildin...

    2025

  21. [29]

    Liudmila Prokhorenkova, Gleb Gusev, Aleksandr Vorobev, Anna Veronika Doro- gush, and Andrey Gulin. 2018. CatBoost: Unbiased Boosting with Categorical Features. In Advances in Neural Information Processing Systems 31 . Curran As- sociates, Inc., 6639–6649. http://papers.nips.cc...

    2018

  22. [30]

    Fan Qi, Yu Duan, Huaiwen Zhang, and Changsheng Xu. 2024. SignGen: End-to- End Sign Language Video Generation with Latent Diffusion. In Computer Vision – ECCV 2024: 18th European Conference, Milan, Italy, September 29–October 4, 2024, Proceedings, Part LIII (Milan, Italy)....

    2024 doi

  23. [31]

    Yujia Qin, Shihao Liang, Yining Ye, Kunlun Zhu, Lan Yan, Yaxi Lu, Yankai Lin, Xin Cong, Xiangru Tang, Bill Qian, Sihan Zhao, Lauren Hong, Runchu Tian, Ruobing Xie, Jie Zhou, Mark Gerstein, dahai li, Zhiyuan Liu, and Maosong Sun

  24. [32]

    Jorge Peña Queralta, Jussi Taipalmaa, Bilge Can Pullinen, Victor Kathan Sarker, Tuan Nguyen Gia, Hannu Tenhunen, Moncef Gabbouj, Jenni Raitoharju, and Tomi Westerlund. 2020. Collaborative Multi-Robot Search and Rescue: Planning, Coordination, Perception, and Active Vision.IEEE...

    2020

  25. [33]

    Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, Gretchen Krueger, and Ilya Sutskever. 2021. Learning Transferable Visual Models From Natural Language Supervision. In Proceedings...

    2021

  26. [34]

    Alec Radford, Jong Wook Kim, Tao Xu, Greg Brockman, Christine McLeavey, and Ilya Sutskever. 2023. Robust speech recognition via large-scale weak supervision. In Proceedings of the 40th International Conference on Machine Learning(Honolulu, Hawaii, USA) (ICML’23). JMLR.org, Art...

    2023

  27. [35]

    Robin Rombach, Andreas Blattmann, Dominik Lorenz, Patrick Esser, and Bjorn Ommer. 2022. High-Resolution Image Synthesis with Latent Diffusion Models . In 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) . IEEE Computer Society, Los Alamitos, CA, USA, ...

    2022

  28. [36]

    Saini, Rajesh Kumar, Sujil A., Ramesh C

    Vikas K. Saini, Rajesh Kumar, Sujil A., Ramesh C. Bansal, Chaouki Ghenai, Maamar Bettayeb, Vladimir Terzija, Elena Gryazina, and Petr Vorobev. 2024. Multi agent framework for consumer demand response in electricity market: Applications and recent advancement. Sustainable Energ...

    2024

  29. [37]

    Weiss, Mike Schuster, Navdeep Jaitly, Zongheng Yang, Zhifeng Chen, Yu Zhang, Yuxuan Wang, Rj Skerrv-Ryan, Rif A

    Jonathan Shen, Ruoming Pang, Ron J. Weiss, Mike Schuster, Navdeep Jaitly, Zongheng Yang, Zhifeng Chen, Yu Zhang, Yuxuan Wang, Rj Skerrv-Ryan, Rif A. Saurous, Yannis Agiomvrgiannakis, and Yonghui Wu. 2018. Natural TTS Syn- thesis by Conditioning Wavenet on MEL Spectrogram Predi...

    2018

  30. [38]

    Karthik Sreedhar and Lydia Chilton. 2024. Simulating Strategic Reasoning: Comparing the Ability of Single LLMs and Multi-Agent Systems to Replicate Human Behavior. arXiv:2402.08189 [cs.CL] https://arxiv.org/pdf/2402.08189

    2024 arXiv

  31. [39]

    Stephanie Stoll, Necati Camgoz, Simon Hadfield, and Richard Bowden. 2018. Sign Language Production using Neural Machine Translation and Generative Adversarial Networks

    2018

  32. [40]

    Lijun Sun, Yijun Yang, Qiqi Duan, Yuhui Shi, Chao Lyu, Yu-Cheng Chang, Chin- Teng Lin, and Yang Shen. 2025. Multi-Agent Coordination across Diverse Appli- cations: A Survey. arXiv:2502.14743 [cs.MA] https://arxiv.org/abs/2502.14743

    2025 arXiv

  33. [41]

    Shengeng Tang, Jiayi He, Dan Guo, Yanyan Wei, Feng Li, and Richang Hong. 2025. Sign-IDD: Iconicity Disentangled Diffusion for Sign Language Production. In Proceedings of the AAAI Conference on Artificial Intelligence , Vol. 39. 7266–7274

    2025

  34. [42]

    Khanh-Tung Tran, Dung Dao, Minh-Duong Nguyen, Quoc-Viet Pham, Barry O’Sullivan, and Hoang D. Nguyen. 2025. Multi-Agent Collaboration Mechanisms: A Survey of LLMs. arXiv:2501.06322 [cs.AI] https://arxiv.org/abs/2501.06322

    2025 arXiv

  35. [43]

    Jiuniu Wang, Hangjie Yuan, Dayou Chen, Yingya Zhang, Xiang Wang, and Shiwei Zhang. 2023. ModelScope Text-to-Video Technical Report. arXiv:2308.06571 [cs.CV] https://arxiv.org/abs/2308.06571

    2023 arXiv

  36. [44]

    Lei Wang, Chen Ma, Xueyang Feng, Zeyu Zhang, Hao Yang, Jingsen Zhang, Zhiyuan Chen, Jiakai Tang, Xu Chen, Yankai Lin, Wayne Xin Zhao, Zhewei Wei, and Jirong Wen. 2024. A survey on large language model based autonomous agents. Frontiers of Computer Science 18, 6 (March 2024). h...

    2024

  37. [45]

    Qingyun Wu, Gagan Bansal, Jieyu Zhang, Yiran Wu, Beibin Li, Erkang Zhu, Li Jiang, Xiaoyun Zhang, Shaokun Zhang, Jiale Liu, Ahmed Hassan Awadallah, Ryen W White, Doug Burger, and Chi Wang. 2024. AutoGen: Enabling Next- Gen LLM Applications via Multi-Agent Conversations. In Firs...

    2024

  38. [46]

    Shengqiong Wu, Hao Fei, Leigang Qu, Wei Ji, and Tat-Seng Chua. 2024. NExT- GPT: any-to-any multimodal LLM. In Proceedings of the 41st International Con- ference on Machine Learning (Vienna, Austria) (ICML’24). JMLR.org, Article 2187, 32 pages

    2024

  39. [47]

    Tong Wu, Pan Zhou, Kai Liu, Yali Yuan, Xiumin Wang, Huawei Huang, and Dapeng Oliver Wu. 2020. Multi-Agent Deep Reinforcement Learning for Urban Traffic Light Control in Vehicular Networks. IEEE Transactions on Vehicular Technology 69, 8 (2020), 8243–8256. https://doi.org/10.11...

    2020

  40. [48]

    Zhuoyi Yang, Jiayan Teng, Wendi Zheng, Ming Ding, Shiyu Huang, Jiazheng Xu, Yuanming Yang, Wenyi Hong, Xiaohan Zhang, Guanyu Feng, Da Yin, Yuxuan Zhang, Weihan Wang, Yean Cheng, Xu Bin, Xiaotao Gu, Yuxiao Dong, and Jie Tang. 2025. CogVideoX: Text-to-Video Diffusion Models with...

    2025

  41. [49]

    Shunyu Yao, Jeffrey Zhao, Dian Yu, Nan Du, Izhak Shafran, Karthik Narasimhan, and Yuan Cao. 2023. ReAct: Synergizing Reasoning and Acting in Language Models. In International Conference on Learning Representations (ICLR)

    2023

  42. [50]

    Junjie Ye, Xuanting Chen, Nuo Xu, Can Zu, Zekai Shao, Shichun Liu, Yuhan Cui, Zeyang Zhou, Chao Gong, Yang Shen, Jie Zhou, Siming Chen, Tao Gui, Qi Zhang, and Xuanjing Huang. 2023. A Comprehensive Capability Analysis of GPT-3 and GPT-3.5 Series Models. arXiv:2303.10420 [cs.CL]...

    2023 arXiv

  43. [51]

    Biao Zhang, Mathias Müller, and Rico Sennrich. 2023. SLTUNET: A Simple Uni- fied Model for Sign Language Translation. In The Eleventh International Confer- ence on Learning Representations . https://openreview.net/forum?id=EBS4C77p_ 5S

    2023

  44. [52]

    Hang Zhang, Xin Li, and Lidong Bing. 2023. Video-LLaMA: An Instruction- tuned Audio-Visual Language Model for Video Understanding. In Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing: System Demonstrations, Yansong Feng and Els Lefever (Ed...

    2023 doi

  45. [53]

    Wenliang Zhao, Xumin Yu, and Zengyi Qin. 2023. MeloTTS: High-quality Multi- lingual Multi-accent Text-to-Speech. https://github.com/myshell-ai/MeloTTS

    2023

  46. [54]

    Bin Zhu, Munan Ning, Peng Jin, Bin Lin, Jinfa Huang, Qi Song, Junwu Zhang, Zhenyu Tang, Mingjun Pan, Xing Zhou, and Li Yuan. 2024. LLMBind: A Unified Modality-Task Integration Framework. arXiv:2402.14891 [cs.CL] https://arxiv. org/abs/2402.14891

    2024

  47. [2024]

    In The Twelfth International Conference on Learning Representations

    ToolLLM: Facilitating Large Language Models to Master 16000+ Real- world APIs. In The Twelfth International Conference on Learning Representations . https://openreview.net/forum?id=dHng2O0Jjr

Pith tools

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