% ============================================================================
% papers.bib -- UQO Working Paper 31 (PRISME), literature review
% Compiled 2026-10-29 by verifying every entry against primary sources:
%   * arXiv Atom API (https://export.arxiv.org/api/query?id_list=...) for
%     title, author list, first-submission date, latest version, comments/DOI;
%   * GitHub REST API (api.github.com/repos/<owner>/<repo>) for code + licence;
%   * JMLR / PMLR / project pages where cited.
% Convention: `year` = year of first arXiv submission unless a venue is given.
% Venues are stated ONLY when verified from arXiv metadata (comment/DOI field),
% a PMLR/JMLR page, or an official repository description. Where a venue is
% widely known but was NOT verified here, it is flagged in `note`.
% `note` also records code availability and licence (SPDX id from GitHub API).
% ============================================================================

% ---------------------------------------------------------------------------
% 1. Core references requested (12)
% ---------------------------------------------------------------------------

@inproceedings{dehghani2019universal,
  title     = {Universal Transformers},
  author    = {Dehghani, Mostafa and Gouws, Stephan and Vinyals, Oriol and Uszkoreit, Jakob and Kaiser, {\L}ukasz},
  booktitle = {International Conference on Learning Representations (ICLR)},
  year      = {2019},
  eprint    = {1807.03819},
  archivePrefix = {arXiv},
  url       = {https://arxiv.org/abs/1807.03819},
  note      = {arXiv v1 2018-07-10, v3 2019-03-05; ICLR 2019 per arXiv comment. Mechanism: depth-shared transformer block applied recurrently with ACT-style per-position dynamic halting. Code: tensorflow/tensor2tensor (Apache-2.0).}
}

@inproceedings{banino2021pondernet,
  title     = {PonderNet: Learning to Ponder},
  author    = {Banino, Andrea and Balaguer, Jan and Blundell, Charles},
  booktitle = {8th ICML Workshop on Automated Machine Learning (AutoML)},
  year      = {2021},
  eprint    = {2107.05407},
  archivePrefix = {arXiv},
  url       = {https://arxiv.org/abs/2107.05407},
  note      = {arXiv v1 2021-07-12, v2 2021-09-02; venue per arXiv comment. Mechanism: probabilistic halting (geometric prior, KL-regularised) over recurrent steps; expected-loss training. No official code release verified.}
}

@inproceedings{hawthorne2022perceiverar,
  title     = {General-purpose, long-context autoregressive modeling with {Perceiver AR}},
  author    = {Hawthorne, Curtis and Jaegle, Andrew and Cangea, C{\u{a}}t{\u{a}}lina and Borgeaud, Sebastian and Nash, Charlie and Malinowski, Mateusz and Dieleman, Sander and Vinyals, Oriol and Botvinick, Matthew and Simon, Ian and Sheahan, Hannah and Zeghidour, Neil and Alayrac, Jean-Baptiste and Carreira, Jo{\~a}o and Engel, Jesse},
  booktitle = {International Conference on Machine Learning (ICML)},
  year      = {2022},
  eprint    = {2202.07765},
  archivePrefix = {arXiv},
  url       = {https://arxiv.org/abs/2202.07765},
  note      = {arXiv v1 2022-02-15, v2 2022-06-14; ICML 2022 per arXiv comment. Mechanism: small set of latents (last positions) cross-attend causally to a long input; latents re-created per forward pass. Code: google-research/perceiver-ar (Apache-2.0).}
}

@article{goyal2019rims,
  title     = {Recurrent Independent Mechanisms},
  author    = {Goyal, Anirudh and Lamb, Alex and Hoffmann, Jordan and Sodhani, Shagun and Levine, Sergey and Bengio, Yoshua and Sch{\"o}lkopf, Bernhard},
  journal   = {arXiv preprint arXiv:1909.10893},
  year      = {2019},
  url       = {https://arxiv.org/abs/1909.10893},
  note      = {arXiv v1 2019-09-24, v6 2020-11-17. Venue (ICLR 2021) NOT verified from arXiv metadata. Mechanism: K recurrent modules, top-k activated per step by input attention, sparse communication; only selected modules update. Code: anirudh9119/RIMs (no licence file in repo).}
}

@article{raposo2024mixtureofdepths,
  title     = {Mixture-of-Depths: Dynamically allocating compute in transformer-based language models},
  author    = {Raposo, David and Ritter, Sam and Richards, Blake and Lillicrap, Timothy and Humphreys, Peter Conway and Santoro, Adam},
  journal   = {arXiv preprint arXiv:2404.02258},
  year      = {2024},
  url       = {https://arxiv.org/abs/2404.02258},
  note      = {arXiv v1 2024-04-02 (single version). Mechanism: per-layer top-k token routing with fixed capacity; non-routed tokens bypass the block through the residual (exact skip). No official code release found.}
}

@article{geiping2025recurrentdepth,
  title     = {Scaling up Test-Time Compute with Latent Reasoning: A Recurrent Depth Approach},
  author    = {Geiping, Jonas and McLeish, Sean and Jain, Neel and Kirchenbauer, John and Singh, Siddharth and Bartoldson, Brian R. and Kailkhura, Bhavya and Bhatele, Abhinav and Goldstein, Tom},
  journal   = {arXiv preprint arXiv:2502.05171},
  year      = {2025},
  url       = {https://arxiv.org/abs/2502.05171},
  note      = {Huginn. arXiv v1 2025-02-07, v2 2025-02-17. Mechanism: prelude / recurrent core / coda; core iterated r times with random r at training; zero-shot per-token exit via KL threshold between iterations. Code: seal-rg/recurrent-pretraining (Apache-2.0); model huggingface.co/tomg-group-umd/huginn-0125.}
}

@inproceedings{bae2025mixtureofrecursions,
  title     = {Mixture-of-Recursions: Learning Dynamic Recursive Depths for Adaptive Token-Level Computation},
  author    = {Bae, Sangmin and Kim, Yujin and Bayat, Reza and Kim, Sungnyun and Ha, Jiyoun and Schuster, Tal and Fisch, Adam and Harutyunyan, Hrayr and Ji, Ziwei and Courville, Aaron and Yun, Se-Young},
  booktitle = {Advances in Neural Information Processing Systems (NeurIPS)},
  year      = {2025},
  eprint    = {2507.10524},
  archivePrefix = {arXiv},
  url       = {https://arxiv.org/abs/2507.10524},
  note      = {arXiv v1 2025-07-14, v3 2025-10-25. NeurIPS 2025 per official repository description. Mechanism: shared recursive block + lightweight routers (expert-choice / token-choice) assigning a recursion depth per token; recursion-wise KV caching. Code: raymin0223/mixture\_of\_recursions (Apache-2.0).}
}

@article{zhu2025ouro,
  title     = {Scaling Latent Reasoning via Looped Language Models},
  author    = {Zhu, Rui-Jie and Wang, Zixuan and Hua, Kai and Zhang, Tianyu and Li, Ziniu and Que, Haoran and Wei, Boyi and Wen, Zixin and Yin, Fan and Xing, He and Li, Lu and Shi, Jiajun and Ma, Kaijing and Li, Shanda and Kergan, Taylor and Smith, Andrew and Qu, Xingwei and Hui, Mude and Wu, Bohong and Min, Qiyang and Huang, Hongzhi and Zhou, Xun and Ye, Wei and Liu, Jiaheng and Yang, Jian and Shi, Yunfeng and Lin, Chenghua and Zhao, Enduo and Cai, Tianle and Zhang, Ge and Huang, Wenhao and Bengio, Yoshua and Eshraghian, Jason},
  journal   = {arXiv preprint arXiv:2510.25741},
  year      = {2025},
  url       = {https://arxiv.org/abs/2510.25741},
  note      = {Ouro / LoopLM. arXiv v1 2025-10-29, v6 2026-10-11. Mechanism: weight-tied stack looped up to 4 times; per-step exit gate lambda\_t = sigma(Linear(h\_t)) trained (Stage I) with entropy-regularised expected loss and (Stage II, LM frozen) with BCE against labels w = sigma(k(max(0, L\_{t-1}-L\_t) - gamma)) derived from observed per-token loss improvements. Weights: huggingface.co/ByteDance/Ouro-1.4B, Ouro-2.6B (+ -Thinking) under Apache-2.0 (research-use notice). No official training code found; community reimplementation rkstgr/LoopLM (no licence). No venue listed.}
}

@article{wang2026recurtrace,
  title     = {RecurTrace: Adaptive Latent Reasoning with Loop-Time Memory},
  author    = {Wang, Yuxiang and Feng, Kunyu and Shen, Yingda and Xu, Haoning and Wang, Junyu and Wu, Zhizheng},
  journal   = {arXiv preprint arXiv:2609.03379},
  year      = {2026},
  url       = {https://arxiv.org/abs/2609.03379},
  note      = {arXiv v1 2026-09-03, v3 2026-10-19. Mechanism: Loop Memory Attention (each looped layer attends over a window W of its own outputs from earlier loops at the same position; empty at loop 1 so the term is zero) + sequence-level halting head p\_t = sigma(w^T z\_t + b) trained with BCE against oracle labels 1[min\_{t'>t} L\_{t'} < L\_t - delta]. Code: JensenYX/RecurTrace (Apache-2.0, repo created 2026-10-17). No venue listed.}
}

@article{ladhak2026elastic,
  title     = {One Model to Loop Them All: Elastic Latent Recurrence for Language Models},
  author    = {Ladhak, Faisal and Acharya, Shantanu and Akle Serano, Santiago and Sun, Simeng and Ginsburg, Boris and Puvvada, Krishna C.},
  journal   = {arXiv preprint arXiv:2610.12582},
  year      = {2026},
  url       = {https://arxiv.org/abs/2610.12582},
  note      = {ELR. arXiv v1 2026-10-08. Mechanism: training recipe removing the "one-loop tax": gated state update with learned step size, mixed loop counts per optimizer update, loop-index/budget conditioning (zero-initialised when added late), last-loop-only KV cache. Reports Koi-4B checkpoint. No code or checkpoint link verified (paper HTML links only to the nvidia nemotron-post-training-v3 HF collection); third-party claims of a public "Dory1.0-4B" release were NOT verified.}
}

@article{behrouz2024titans,
  title     = {Titans: Learning to Memorize at Test Time},
  author    = {Behrouz, Ali and Zhong, Peilin and Mirrokni, Vahab},
  journal   = {arXiv preprint arXiv:2501.00663},
  year      = {2024},
  url       = {https://arxiv.org/abs/2501.00663},
  note      = {arXiv v1 2024-12-31 (single version). Mechanism: neural long-term memory module updated at test time by gradient (surprise) with forgetting; persistent memory tokens; three integration variants (MAC/MAG/MAL). No official code; unofficial lucidrains/titans-pytorch (MIT).}
}

@inproceedings{elhoushi2024layerskip,
  title     = {LayerSkip: Enabling Early Exit Inference and Self-Speculative Decoding},
  author    = {Elhoushi, Mostafa and Shrivastava, Akshat and Liskovich, Diana and Hosmer, Basil and Wasti, Bram and Lai, Liangzhen and Mahmoud, Anas and Acun, Bilge and Agarwal, Saurabh and Roman, Ahmed and Aly, Ahmed A. and Chen, Beidi and Wu, Carole-Jean},
  booktitle = {Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (ACL)},
  year      = {2024},
  doi       = {10.18653/v1/2024.acl-long.681},
  eprint    = {2404.16710},
  archivePrefix = {arXiv},
  url       = {https://arxiv.org/abs/2404.16710},
  note      = {arXiv v1 2024-04-25, v4 2024-10-18. Mechanism: layer dropout + shared early-exit loss; self-speculative decoding (draft with early layers, verify with the rest). Code: facebookresearch/LayerSkip (CC BY-NC 4.0).}
}

% ---------------------------------------------------------------------------
% 2. Adaptive computation, looped / recurrent-depth, latent reasoning
% ---------------------------------------------------------------------------

@article{graves2016act,
  title     = {Adaptive Computation Time for Recurrent Neural Networks},
  author    = {Graves, Alex},
  journal   = {arXiv preprint arXiv:1603.08983},
  year      = {2016},
  url       = {https://arxiv.org/abs/1603.08983},
  note      = {arXiv v1 2016-03-29, v6 2017-02-21. Mechanism: sigmoidal halting unit accumulated to a threshold; ponder cost penalty. No official code.}
}

@inproceedings{hao2024coconut,
  title     = {Training Large Language Models to Reason in a Continuous Latent Space},
  author    = {Hao, Shibo and Sukhbaatar, Sainbayar and Su, DiJia and Li, Xian and Hu, Zhiting and Weston, Jason and Tian, Yuandong},
  booktitle = {Conference on Language Modeling (COLM)},
  year      = {2025},
  eprint    = {2412.06769},
  archivePrefix = {arXiv},
  url       = {https://arxiv.org/abs/2412.06769},
  note      = {Coconut. arXiv v1 2024-12-09, v4 2026-08-23; COLM 2025 per arXiv comment. Mechanism: last hidden state fed back as next input embedding ("continuous thought") for a fixed number of latent steps, curriculum from CoT. Code: facebookresearch/coconut (MIT).}
}

@article{giannou2023looped,
  title     = {Looped Transformers as Programmable Computers},
  author    = {Giannou, Angeliki and Rajput, Shashank and Sohn, Jy-yong and Lee, Kangwook and Lee, Jason D. and Papailiopoulos, Dimitris},
  journal   = {arXiv preprint arXiv:2301.13196},
  year      = {2023},
  url       = {https://arxiv.org/abs/2301.13196},
  note      = {arXiv v1 2023-01-30. Venue (ICML 2023) NOT verified from arXiv metadata. Constructive result: a looped transformer with a scratchpad can emulate a programmable computer.}
}

@inproceedings{saunshi2025reasoning,
  title     = {Reasoning with Latent Thoughts: On the Power of Looped Transformers},
  author    = {Saunshi, Nikunj and Dikkala, Nishanth and Li, Zhiyuan and Kumar, Sanjiv and Reddi, Sashank J.},
  booktitle = {International Conference on Learning Representations (ICLR)},
  year      = {2025},
  eprint    = {2502.17416},
  archivePrefix = {arXiv},
  url       = {https://arxiv.org/abs/2502.17416},
  note      = {arXiv v1 2025-02-24; ICLR 2025 per arXiv comment. Theory + experiments: k-layer block looped L times matches kL-layer models on reasoning; looping-inspired regularisation.}
}

@inproceedings{fan2025loopedlength,
  title     = {Looped Transformers for Length Generalization},
  author    = {Fan, Ying and Du, Yilun and Ramchandran, Kannan and Lee, Kangwook},
  booktitle = {International Conference on Learning Representations (ICLR)},
  year      = {2025},
  eprint    = {2409.15647},
  archivePrefix = {arXiv},
  url       = {https://arxiv.org/abs/2409.15647},
  note      = {arXiv v1 2024-09-24, v5 2025-05-12; ICLR 2025 per arXiv comment. Adaptive number of loops tied to problem length for algorithmic tasks.}
}

@inproceedings{bae2024relaxed,
  title     = {Relaxed Recursive Transformers: Effective Parameter Sharing with Layer-wise LoRA},
  author    = {Bae, Sangmin and Fisch, Adam and Harutyunyan, Hrayr and Ji, Ziwei and Kim, Seungyeon and Schuster, Tal},
  booktitle = {International Conference on Learning Representations (ICLR)},
  year      = {2025},
  eprint    = {2410.20672},
  archivePrefix = {arXiv},
  url       = {https://arxiv.org/abs/2410.20672},
  note      = {arXiv v1 2024-10-28, v3 2025-02-28; ICLR 2025 per arXiv comment. Recursive (looped) transformers with per-loop LoRA; continuous depth-wise batching for early exit. Predecessor of Mixture-of-Recursions.}
}

@inproceedings{goyal2023pause,
  title     = {Think before you speak: Training Language Models With Pause Tokens},
  author    = {Goyal, Sachin and Ji, Ziwei and Rawat, Ankit Singh and Menon, Aditya Krishna and Kumar, Sanjiv and Nagarajan, Vaishnavh},
  booktitle = {International Conference on Learning Representations (ICLR)},
  year      = {2024},
  eprint    = {2310.02226},
  archivePrefix = {arXiv},
  url       = {https://arxiv.org/abs/2310.02226},
  note      = {arXiv v1 2023-10-03, v3 2024-04-21; ICLR 2024 per arXiv comment. Mechanism: learnable dummy tokens appended before the answer to buy extra (dense, uniform) computation whose outputs are discarded.}
}

@article{sun2025latenttokens,
  title     = {Enhancing Latent Computation in Transformers with Latent Tokens},
  author    = {Sun, Yuchang and Chen, Yanxi and Li, Yaliang and Ding, Bolin},
  journal   = {arXiv preprint arXiv:2505.12629},
  year      = {2025},
  url       = {https://arxiv.org/abs/2505.12629},
  note      = {arXiv v1 2025-05-19. Mechanism: dummy non-linguistic tokens inserted before verbal tokens, parameter-efficient training, loss terms on them dropped.}
}

@inproceedings{zeng2025ponderlm,
  title     = {PonderLM: Pretraining Language Models to Ponder in Continuous Space},
  author    = {Zeng, Boyi and Song, Shixiang and Huang, Siyuan and Wang, Yixuan and Li, He and He, Ziwei and Wang, Xinbing and Li, Zhiyu and Lin, Zhouhan},
  booktitle = {International Conference on Learning Representations (ICLR)},
  year      = {2026},
  eprint    = {2505.20674},
  archivePrefix = {arXiv},
  url       = {https://arxiv.org/abs/2505.20674},
  note      = {arXiv v1 2025-05-27, v3 2026-02-20; ICLR 2026 per arXiv comment. Mechanism: within one token step, feed back the probability-weighted sum of token embeddings as the next input for additional forward passes (self-supervised). Code: LUMIA-Group/PonderingLM (Apache-2.0).}
}

@article{chen2025innerthinking,
  title     = {Inner Thinking Transformer: Leveraging Dynamic Depth Scaling to Foster Adaptive Internal Thinking},
  author    = {Chen, Yilong and Shang, Junyuan and Zhang, Zhenyu and Xie, Yanxi and Sheng, Jiawei and Liu, Tingwen and Wang, Shuohuan and Sun, Yu and Wu, Hua and Wang, Haifeng},
  journal   = {arXiv preprint arXiv:2502.13842},
  year      = {2025},
  url       = {https://arxiv.org/abs/2502.13842},
  note      = {arXiv v1 2025-02-19, v2 2025-02-23. Mechanism: Adaptive Token Routing selects critical tokens for extra inner iterations of a layer; residual thinking connections; thinking-step encoding.}
}

@article{koishekenov2025etd,
  title     = {Encode, Think, Decode: Scaling test-time reasoning with recursive latent thoughts},
  author    = {Koishekenov, Yeskendir and Lipani, Aldo and Cancedda, Nicola},
  journal   = {arXiv preprint arXiv:2510.07358},
  year      = {2025},
  url       = {https://arxiv.org/abs/2510.07358},
  note      = {arXiv v1 2025-10-08. Mechanism: mid-training to iterate a small subset of reasoning-relevant layers; explores adaptive per-token depth.}
}

@article{wu2025parallelloop,
  title     = {Parallel Loop Transformer for Efficient Test-Time Computation Scaling},
  author    = {Wu, Bohong and Chen, Mengzhao and Luo, Xiang and Yan, Shen and Yu, Qifan and Xia, Fan and Zhang, Tianqi and Zhan, Hongrui and Zhong, Zheng and Zhou, Xun and Qiao, Siyuan and Bin, Xingyan},
  journal   = {arXiv preprint arXiv:2510.24824},
  year      = {2025},
  url       = {https://arxiv.org/abs/2510.24824},
  note      = {arXiv v1 2025-10-28. Mechanism: cross-loop parallelism (different loops for different tokens in one pass), first-loop KV sharing, gated sliding-window attention.}
}

@inproceedings{fu2026thinkathard,
  title     = {Think-at-Hard: Dynamic Looped Transformers for Improved Reasoning},
  author    = {Fu, Tianyu and You, Yichen and Chen, Zekai and Dai, Guohao and Yang, Huazhong and Wang, Yu},
  booktitle = {Proceedings of the 43rd International Conference on Machine Learning (ICML), PMLR 306},
  pages     = {32129--32154},
  year      = {2026},
  eprint    = {2511.08577},
  archivePrefix = {arXiv},
  url       = {https://proceedings.mlr.press/v306/fu26t.html},
  note      = {TaH. arXiv v1 2025-11-11 (earlier title "Selective Latent Iterations to Improve Reasoning Language Models"), v4 2026-08-28; ICML 2026 verified on PMLR. Mechanism: lightweight neural decider triggers a second latent iteration only at tokens likely wrong after the first pass (oracle = token mismatch); depth-aware LoRA; duo-causal attention across iteration depth. Code: thu-nics/TaH (Apache-2.0).}
}

@article{you2026tah2,
  title     = {Improving Test-Time Scaling with Adaptive Looped Transformers},
  author    = {You, Yichen and Fu, Tianyu and Feng, Aosong and Lv, Xingtai and Ning, Xuefei and Ding, Ning and Wang, Yu},
  journal   = {arXiv preprint arXiv:2609.35748},
  year      = {2026},
  url       = {https://arxiv.org/abs/2609.35748},
  note      = {TaH2. arXiv v1 2026-09-28. Mechanism: joint post-training of backbone and per-token iteration decider with "lookahead depth supervision": online per-token labels from the loss gain L\_m - L\_{m+1} (no-gradient lookahead branch for tokens that already stopped), coverage filtering (rho=0.99), cost-sensitive BCE; binary continue/stop, absorbing stop; no propensity logging. Code stated as github.com/thu-nics/TaH (shared with TaH).}
}

@inproceedings{jeddi2026loopformer,
  title     = {LoopFormer: Elastic-Depth Looped Transformers for Latent Reasoning via Shortcut Modulation},
  author    = {Jeddi, Ahmadreza and Ciccone, Marco and Taati, Babak},
  booktitle = {International Conference on Learning Representations (ICLR)},
  year      = {2026},
  eprint    = {2602.11451},
  archivePrefix = {arXiv},
  url       = {https://arxiv.org/abs/2602.11451},
  note      = {arXiv v1 2026-02-11; ICLR 2026 per arXiv comment. Mechanism: shortcut-consistency training on variable-length loop trajectories conditioned on time and step size (budget-conditioned depth).}
}

@article{song2026adaponderlm,
  title     = {AdaPonderLM: Gated Pondering Language Models with Token-Wise Adaptive Depth},
  author    = {Song, Shixiang and Li, He and Wang, Zitong and Zeng, Boyi and Song, Feichen and Wang, Yixuan and Xu, Zhiqin John and He, Ziwei and Lin, Zhouhan},
  journal   = {arXiv preprint arXiv:2603.01914},
  year      = {2026},
  url       = {https://arxiv.org/abs/2603.01914},
  note      = {arXiv v1 2026-03-02, v2 2026-03-11. Mechanism: iteration-specific MLP gates with monotonic halting mask learned self-supervised during pretraining; KV reuse for halted tokens.}
}

@article{noci2026lookahead,
  title     = {Thinking into the Future: Latent Lookahead Training for Transformers},
  author    = {Noci, Lorenzo and Bachmann, Gregor and Moosavi-Dezfooli, Seyed-Mohsen and Nabi, Moin},
  journal   = {arXiv preprint arXiv:2603.20219},
  year      = {2026},
  url       = {https://arxiv.org/abs/2603.20219},
  note      = {arXiv v1 2026-03-03. Mechanism: at selected positions, recursively feed hidden states back for tau latent steps supervised against the next tau tokens.}
}

@inproceedings{frey2026adaptiveloops,
  title     = {Adaptive Loops and Memory in Transformers: Think Harder or Know More?},
  author    = {Frey, Markus and Shomali, Behzad and Bashir, Ali Hamza and Berghaus, David and Koehler, Joachim and Ali, Mehdi},
  booktitle = {Latent \& Implicit Thinking Workshop @ ICLR 2026},
  year      = {2026},
  eprint    = {2603.08391},
  archivePrefix = {arXiv},
  url       = {https://arxiv.org/abs/2603.08391},
  note      = {arXiv v1 2026-03-09, v3 2026-03-11; workshop per arXiv comment. Mechanism: per-layer adaptive looping with learned halting + gated memory banks.}
}

@article{park2026loopus,
  title     = {LoopUS: Recasting Pretrained LLMs into Looped Latent Refinement Models},
  author    = {Park, Taekhyun and Lee, Yongjae and Kim, Dohee and Bae, Hyerim},
  journal   = {arXiv preprint arXiv:2605.11011},
  year      = {2026},
  url       = {https://arxiv.org/abs/2605.11011},
  note      = {arXiv v1 2026-05-10. Mechanism: encoder / looped block / decoder recast of a pretrained LLM; input-dependent selective gate; random deep supervision; confidence head for early exit.}
}

@article{huang2026lrt,
  title     = {Latent Recurrent Transformer: Architecture Exploration, Training Strategies, and Scaling Behavior},
  author    = {Huang, Zeyi and He, Xuehai and Ren, LiLiang and Wang, Yiping and Peng, Baolin and Cheng, Hao and Wang, Shuohang and He, Pengcheng and Gao, Jianfeng and Lee, Yong Jae and Shen, Yelong},
  journal   = {arXiv preprint arXiv:2605.26797},
  year      = {2026},
  url       = {https://arxiv.org/abs/2605.26797},
  note      = {arXiv v1 2026-05-26, v2 2026-09-01. Mechanism: reuse a high-level source-layer hidden state of the previous token as recurrent memory for the next token; interleaved parallel training; one forward per token.}
}

@article{li2026later,
  title     = {LaTER: Efficient Test-Time Reasoning via Latent Exploration and Explicit Verification},
  author    = {Li, Xuan and Wang, Yining and Liu, Yuchen and Liu, Guanjun and Qiu, Delai and Liu, Shengping and Liang, Jiaen and Huang, Wei and Yu, Jun and Zhu, Junnan},
  journal   = {arXiv preprint arXiv:2605.07315},
  year      = {2026},
  url       = {https://arxiv.org/abs/2605.07315},
  note      = {arXiv v1 2026-05-08. Mechanism: bounded latent exploration then explicit CoT; halting supervision from a constructed corpus. Code: TioeAre/LaTER (Apache-2.0).}
}

@article{sharma2026readout,
  title     = {Dense Supervision Is Not Enough: The Readout Blind Spot in Looped Language Models},
  author    = {Sharma, Rituraj and Vu, Tu},
  journal   = {arXiv preprint arXiv:2606.24898},
  year      = {2026},
  url       = {https://arxiv.org/abs/2606.24898},
  note      = {arXiv v1 2026-06-12. Finding: per-loop cross-entropy through scale-invariant readouts does not control recurrent hidden-state scale; relevant to readout design in looped/edited states.}
}

@article{zeng2026depthattention,
  title     = {Depth-Attention: Cross-Layer Value Mixing for Language Models},
  author    = {Zeng, Boyi and Hao, Yiqin and Wang, Zitong and Song, Shixiang and Li, He and Song, Feichen and Liu, Yifan and He, Ziwei and Wang, Xinbing and Lin, Zhouhan},
  journal   = {arXiv preprint arXiv:2606.05014},
  year      = {2026},
  url       = {https://arxiv.org/abs/2606.05014},
  note      = {arXiv v1 2026-06-03, v2 2026-06-13. Mechanism: query attends over keys of earlier layers at the same position inside attention; no extra persistent state.}
}

@article{fu2026discoloop,
  title     = {DiscoLoop: Looping Discrete Embeddings and Continuous Hidden States for Multi-hop Reasoning},
  author    = {Fu, Hengyu and Guo, Tianyu and Wang, Zixuan and Zhu, Hanlin and Lee, Jason D. and Jiao, Jiantao and Russell, Stuart and Mei, Song},
  journal   = {arXiv preprint arXiv:2607.00341},
  year      = {2026},
  url       = {https://arxiv.org/abs/2607.00341},
  note      = {arXiv v1 2026-07-01, v2 2026-07-27.}
}

@article{popescu2026adaptivedepth,
  title     = {Adaptive Depth in Looped Transformers: Diagnosing Learned Halting Gates and Trajectory Readouts},
  author    = {Popescu, Andrei Cristian and S{\'a}ez de Oc{\'a}riz Borde, Haitz and Li{\`o}, Pietro},
  journal   = {arXiv preprint arXiv:2607.20519},
  year      = {2026},
  url       = {https://arxiv.org/abs/2607.20519},
  note      = {arXiv v1 2026-07-08. Finding: jointly trained halting gates entangle exit selection and trajectory formation; post-hoc confidence readouts on fixed-prior-supervised trajectories often match learned gates (incl. on Ouro checkpoints).}
}

@article{logan2026fixedpoint,
  title     = {Per-Token Fixed-Point Convergence in Depth-Recurrent Transformers},
  author    = {Logan, Joe},
  journal   = {arXiv preprint arXiv:2607.14427},
  year      = {2026},
  url       = {https://arxiv.org/abs/2607.14427},
  note      = {arXiv v1 2026-07-15. Finding: training-free convergence-based exit allocates depth better than a trained router. Code: jlognn/depth-recurrent-convergence (MIT).}
}

@article{dau2026beyonddepth,
  title     = {Beyond Depth Truncation: Controlled Evaluation of Depth Utilization in Recursive Language Models},
  author    = {Dau, Ha Van and Khuat, Thanh Tung and Dung, Nguyen Thanh},
  journal   = {arXiv preprint arXiv:2609.19934},
  year      = {2026},
  url       = {https://arxiv.org/abs/2609.19934},
  note      = {arXiv v1 2026-09-17.}
}

@article{huang2026trading,
  title     = {Trading Depth for Time in Recurrent Transformers},
  author    = {Huang, Zeyi and He, Xuehai and Shen, Yelong and Lee, Yong Jae},
  journal   = {arXiv preprint arXiv:2609.21605},
  year      = {2026},
  url       = {https://arxiv.org/abs/2609.21605},
  note      = {arXiv v1 2026-09-18, v2 2026-10-13.}
}

@article{ye2026reflux,
  title     = {Write Back the $\Delta$: Revisiting the Same Tokens with Fresh Representations},
  author    = {Ye, Wencheng and Hu, Anning and Zhang, Xiangdong and Wang, Tianyi and Li, Yikang and Jin, Hengyu and Li, Bing and Yan, Junchi},
  journal   = {arXiv preprint arXiv:2609.32457},
  year      = {2026},
  url       = {https://arxiv.org/abs/2609.32457},
  note      = {ReFlux. arXiv v1 2026-09-26. Mechanism: learnable feedback graph over a frozen backbone; router (per block, shared across tokens) selects a sparse set of deep-to-shallow edges carrying the depth increment Delta = z\_s - z\_d, written additively; trained with next-token loss + L1/L2 route regularisation (Gumbel-sigmoid); no counterfactual labels. With no active edges the backbone is unchanged. Code: gooogleshanghai/reflux (MIT).}
}

@article{olszowiec2026monitoring,
  title     = {Monitoring latent reasoning in looped language models},
  author    = {Olszowiec, Pawel and Tomaszewska, Paulina and {\.Z}elaszczyk, Maciej and Byra, Michal and Chung, Neo Christopher},
  journal   = {arXiv preprint arXiv:2610.18127},
  year      = {2026},
  url       = {https://arxiv.org/abs/2610.18127},
  note      = {arXiv v1 2026-10-13.}
}

% ---------------------------------------------------------------------------
% 3. Slots, workspaces, memory-augmented networks
% ---------------------------------------------------------------------------

@inproceedings{locatello2020slot,
  title     = {Object-Centric Learning with Slot Attention},
  author    = {Locatello, Francesco and Weissenborn, Dirk and Unterthiner, Thomas and Mahendran, Aravindh and Heigold, Georg and Uszkoreit, Jakob and Dosovitskiy, Alexey and Kipf, Thomas},
  booktitle = {Advances in Neural Information Processing Systems (NeurIPS)},
  year      = {2020},
  eprint    = {2006.15055},
  archivePrefix = {arXiv},
  url       = {https://arxiv.org/abs/2006.15055},
  note      = {arXiv v1 2020-06-26; NeurIPS 2020 per arXiv comment. Mechanism: K slots sampled per input, iteratively refined by competitive (softmax over slots) cross-attention + GRU; all slots updated every iteration. Code: google-research/google-research/slot\_attention (Apache-2.0).}
}

@inproceedings{goyal2022workspace,
  title     = {Coordination Among Neural Modules Through a Shared Global Workspace},
  author    = {Goyal, Anirudh and Didolkar, Aniket and Lamb, Alex and Badola, Kartikeya and Ke, Nan Rosemary and Rahaman, Nasim and Binas, Jonathan and Blundell, Charles and Mozer, Michael and Bengio, Yoshua},
  booktitle = {International Conference on Learning Representations (ICLR)},
  year      = {2022},
  eprint    = {2103.01197},
  archivePrefix = {arXiv},
  url       = {https://arxiv.org/abs/2103.01197},
  note      = {arXiv v1 2021-03-01, v2 2022-03-22; ICLR 2022 per arXiv comment. Mechanism: bandwidth-limited shared workspace (few slots); specialist modules compete (top-k) to write; broadcast back. Workspace persists across time steps.}
}

@inproceedings{jaegle2021perceiver,
  title     = {Perceiver: General Perception with Iterative Attention},
  author    = {Jaegle, Andrew and Gimeno, Felix and Brock, Andrew and Zisserman, Andrew and Vinyals, Oriol and Carreira, Joao},
  booktitle = {International Conference on Machine Learning (ICML)},
  year      = {2021},
  eprint    = {2103.03206},
  archivePrefix = {arXiv},
  url       = {https://arxiv.org/abs/2103.03206},
  note      = {arXiv v1 2021-03-04; ICML 2021 per arXiv comment. Mechanism: learned latent array cross-attends to inputs, then latent self-attention; latents re-initialised per forward pass; weight sharing across iterations.}
}

@article{graves2014ntm,
  title     = {Neural Turing Machines},
  author    = {Graves, Alex and Wayne, Greg and Danihelka, Ivo},
  journal   = {arXiv preprint arXiv:1410.5401},
  year      = {2014},
  url       = {https://arxiv.org/abs/1410.5401},
  note      = {arXiv v1 2014-10-20, v2 2014-12-10. Mechanism: differentiable content/location addressing, soft (dense) erase/add writes to persistent memory.}
}

@article{weston2014memory,
  title     = {Memory Networks},
  author    = {Weston, Jason and Chopra, Sumit and Bordes, Antoine},
  journal   = {arXiv preprint arXiv:1410.3916},
  year      = {2014},
  url       = {https://arxiv.org/abs/1410.3916},
  note      = {arXiv v1 2014-10-15, v11 2015-11-29. Venue (ICLR 2015) NOT verified from arXiv metadata.}
}

@inproceedings{rae2016sparse,
  title     = {Scaling Memory-Augmented Neural Networks with Sparse Reads and Writes},
  author    = {Rae, Jack W. and Hunt, Jonathan J. and Harley, Tim and Danihelka, Ivo and Senior, Andrew and Wayne, Greg and Graves, Alex and Lillicrap, Timothy P.},
  booktitle = {Advances in Neural Information Processing Systems (NIPS)},
  year      = {2016},
  eprint    = {1610.09027},
  archivePrefix = {arXiv},
  url       = {https://arxiv.org/abs/1610.09027},
  note      = {Sparse Access Memory (SAM). arXiv v1 2016-10-27; NIPS 2016 per arXiv comment. Mechanism: sparse (top-k) differentiable reads and writes to a large persistent memory for efficiency.}
}

@article{wu2020memformer,
  title     = {Memformer: A Memory-Augmented Transformer for Sequence Modeling},
  author    = {Wu, Qingyang and Lan, Zhenzhong and Qian, Kun and Gu, Jing and Geramifard, Alborz and Yu, Zhou},
  journal   = {arXiv preprint arXiv:2010.06891},
  year      = {2020},
  url       = {https://arxiv.org/abs/2010.06891},
  note      = {arXiv v1 2020-10-14, v2 2022-04-12. Mechanism: fixed-size external memory slots read by cross-attention and written each segment (memory writer with slot-wise attention); persistent across segments. Official code not verified.}
}

@inproceedings{bulatov2022rmt,
  title     = {Recurrent Memory Transformer},
  author    = {Bulatov, Aydar and Kuratov, Yuri and Burtsev, Mikhail S.},
  booktitle = {Advances in Neural Information Processing Systems (NeurIPS)},
  year      = {2022},
  eprint    = {2207.06881},
  archivePrefix = {arXiv},
  url       = {https://arxiv.org/abs/2207.06881},
  note      = {arXiv v1 2022-07-14; NeurIPS 2022 per arXiv comment. Mechanism: memory tokens prepended/appended to each segment, carried across segments. Code: booydar/recurrent-memory-transformer (no licence file).}
}

@inproceedings{ryoo2023ttm,
  title     = {Token Turing Machines},
  author    = {Ryoo, Michael S. and Gopalakrishnan, Keerthana and Kahatapitiya, Kumara and Xiao, Ted and Rao, Kanishka and Stone, Austin and Lu, Yao and Ibarz, Julian and Arnab, Anurag},
  booktitle = {IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
  year      = {2023},
  eprint    = {2211.09119},
  archivePrefix = {arXiv},
  url       = {https://arxiv.org/abs/2211.09119},
  note      = {arXiv v1 2022-11-16; CVPR 2023 per arXiv journal-ref. Mechanism: token-set memory read/written by a transformer controller via token summarisation; persistent across steps. Code: google-research/scenic (Apache-2.0).}
}

@article{berges2024memorylayers,
  title     = {Memory Layers at Scale},
  author    = {Berges, Vincent-Pierre and O{\u{g}}uz, Barlas and Haziza, Daniel and Yih, Wen-tau and Zettlemoyer, Luke and Ghosh, Gargi},
  journal   = {arXiv preprint arXiv:2412.09764},
  year      = {2024},
  url       = {https://arxiv.org/abs/2412.09764},
  note      = {arXiv v1 2024-12-12, v2 2024-12-20. Mechanism: sparsely activated trainable key-value memory layers (product keys) replacing FFNs.}
}

@article{darcet2023registers,
  title     = {Vision Transformers Need Registers},
  author    = {Darcet, Timoth{\'e}e and Oquab, Maxime and Mairal, Julien and Bojanowski, Piotr},
  journal   = {arXiv preprint arXiv:2309.16588},
  year      = {2023},
  url       = {https://arxiv.org/abs/2309.16588},
  note      = {arXiv v1 2023-09-28, v2 2024-04-12. Venue (ICLR 2024) NOT verified from arXiv metadata. Mechanism: extra learnable register tokens providing scratch capacity, discarded at output.}
}

% ---------------------------------------------------------------------------
% 4. Long-context / recurrence / linear-time baselines
% ---------------------------------------------------------------------------

@inproceedings{dai2019transformerxl,
  title     = {Transformer-XL: Attentive Language Models Beyond a Fixed-Length Context},
  author    = {Dai, Zihang and Yang, Zhilin and Yang, Yiming and Carbonell, Jaime and Le, Quoc V. and Salakhutdinov, Ruslan},
  booktitle = {Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics (ACL)},
  year      = {2019},
  eprint    = {1901.02860},
  archivePrefix = {arXiv},
  url       = {https://arxiv.org/abs/1901.02860},
  note      = {arXiv v1 2019-01-09; ACL 2019 per arXiv comment. Code: kimiyoung/transformer-xl (Apache-2.0).}
}

@article{rae2019compressive,
  title     = {Compressive Transformers for Long-Range Sequence Modelling},
  author    = {Rae, Jack W. and Potapenko, Anna and Jayakumar, Siddhant M. and Lillicrap, Timothy P.},
  journal   = {arXiv preprint arXiv:1911.05507},
  year      = {2019},
  url       = {https://arxiv.org/abs/1911.05507},
  note      = {arXiv v1 2019-11-13. Venue (ICLR 2020) NOT verified from arXiv metadata. Mechanism: compress old memories into a secondary compressed memory.}
}

@article{gu2023mamba,
  title     = {Mamba: Linear-Time Sequence Modeling with Selective State Spaces},
  author    = {Gu, Albert and Dao, Tri},
  journal   = {arXiv preprint arXiv:2312.00752},
  year      = {2023},
  url       = {https://arxiv.org/abs/2312.00752},
  note      = {arXiv v1 2023-12-01, v2 2024-05-31. Code: state-spaces/mamba (Apache-2.0). Included only as the canonical fixed-state recurrent baseline.}
}

% ---------------------------------------------------------------------------
% 5. Sparse routing / mixture-of-experts
% ---------------------------------------------------------------------------

@article{shazeer2017moe,
  title     = {Outrageously Large Neural Networks: The Sparsely-Gated Mixture-of-Experts Layer},
  author    = {Shazeer, Noam and Mirhoseini, Azalia and Maziarz, Krzysztof and Davis, Andy and Le, Quoc and Hinton, Geoffrey and Dean, Jeff},
  journal   = {arXiv preprint arXiv:1701.06538},
  year      = {2017},
  url       = {https://arxiv.org/abs/1701.06538},
  note      = {arXiv v1 2017-01-23. Venue (ICLR 2017) NOT verified from arXiv metadata. Mechanism: noisy top-k gating over experts; load-balancing losses.}
}

@article{fedus2022switch,
  title     = {Switch Transformers: Scaling to Trillion Parameter Models with Simple and Efficient Sparsity},
  author    = {Fedus, William and Zoph, Barret and Shazeer, Noam},
  journal   = {Journal of Machine Learning Research},
  year      = {2022},
  eprint    = {2101.03961},
  archivePrefix = {arXiv},
  url       = {https://arxiv.org/abs/2101.03961},
  note      = {arXiv v1 2021-01-11, v3 2022-06-16; JMLR per arXiv comment. Mechanism: top-1 expert routing per token.}
}

% ---------------------------------------------------------------------------
% 6. Value learning, off-policy / counterfactual evaluation
% ---------------------------------------------------------------------------

@article{ernst2005tree,
  title     = {Tree-Based Batch Mode Reinforcement Learning},
  author    = {Ernst, Damien and Geurts, Pierre and Wehenkel, Louis},
  journal   = {Journal of Machine Learning Research},
  volume    = {6},
  pages     = {503--556},
  year      = {2005},
  url       = {https://www.jmlr.org/papers/v6/ernst05a.html},
  note      = {Fitted Q iteration from logged transitions (verified on JMLR page).}
}

@inproceedings{mnih2013atari,
  title     = {Playing Atari with Deep Reinforcement Learning},
  author    = {Mnih, Volodymyr and Kavukcuoglu, Koray and Silver, David and Graves, Alex and Antonoglou, Ioannis and Wierstra, Daan and Riedmiller, Martin},
  booktitle = {NIPS Deep Learning Workshop},
  year      = {2013},
  eprint    = {1312.5602},
  archivePrefix = {arXiv},
  url       = {https://arxiv.org/abs/1312.5602},
  note      = {arXiv v1 2013-12-19; workshop per arXiv comment. Deep Q-network for discrete action selection with greedy argmax over predicted action values.}
}

@inproceedings{dudik2011doublyrobust,
  title     = {Doubly Robust Policy Evaluation and Learning},
  author    = {Dud{\'i}k, Miroslav and Langford, John and Li, Lihong},
  booktitle = {International Conference on Machine Learning (ICML)},
  year      = {2011},
  eprint    = {1103.4601},
  archivePrefix = {arXiv},
  url       = {https://arxiv.org/abs/1103.4601},
  note      = {arXiv v1 2011-03-23; ICML 2011 per arXiv comment. Off-policy evaluation of contextual-bandit policies from logged data with propensities (DR estimator).}
}

@article{bottou2013counterfactual,
  title     = {Counterfactual Reasoning and Learning Systems},
  author    = {Bottou, L{\'e}on and Peters, Jonas and Qui{\~n}onero-Candela, Joaquin and Charles, Denis X. and Chickering, D. Max and Portugaly, Elon and Ray, Dipankar and Simard, Patrice and Snelson, Ed},
  journal   = {arXiv preprint arXiv:1209.2355},
  year      = {2012},
  url       = {https://arxiv.org/abs/1209.2355},
  note      = {arXiv v1 2012-09-11, v5 2013-07-27. Venue (JMLR 2013) NOT verified from arXiv metadata. Counterfactual estimation from logged randomised decisions.}
}

@article{swaminathan2015crm,
  title     = {Counterfactual Risk Minimization: Learning from Logged Bandit Feedback},
  author    = {Swaminathan, Adith and Joachims, Thorsten},
  journal   = {arXiv preprint arXiv:1502.02362},
  year      = {2015},
  url       = {https://arxiv.org/abs/1502.02362},
  note      = {arXiv v1 2015-02-09, v2 2015-05-20. Venue (ICML 2015) NOT verified from arXiv metadata. Learning a policy from logged data with recorded propensities via variance-regularised IPS.}
}

% ---------------------------------------------------------------------------
% 7. Early exit, speculative decoding, test-time compute
% ---------------------------------------------------------------------------

@inproceedings{schuster2022calm,
  title     = {Confident Adaptive Language Modeling},
  author    = {Schuster, Tal and Fisch, Adam and Gupta, Jai and Dehghani, Mostafa and Bahri, Dara and Tran, Vinh Q. and Tay, Yi and Metzler, Donald},
  booktitle = {Advances in Neural Information Processing Systems (NeurIPS)},
  year      = {2022},
  eprint    = {2207.07061},
  archivePrefix = {arXiv},
  url       = {https://arxiv.org/abs/2207.07061},
  note      = {CALM. arXiv v1 2022-07-14; NeurIPS 2022 (oral) per arXiv comment. Mechanism: per-token early exit with calibrated confidence thresholds (softmax diff / hidden-state saturation / learned classifier).}
}

@inproceedings{leviathan2023speculative,
  title     = {Fast Inference from Transformers via Speculative Decoding},
  author    = {Leviathan, Yaniv and Kalman, Matan and Matias, Yossi},
  booktitle = {International Conference on Machine Learning (ICML)},
  year      = {2023},
  eprint    = {2211.17192},
  archivePrefix = {arXiv},
  url       = {https://arxiv.org/abs/2211.17192},
  note      = {arXiv v1 2022-11-30; ICML 2023 oral per arXiv comment.}
}

@article{snell2024testtime,
  title     = {Scaling LLM Test-Time Compute Optimally can be More Effective than Scaling Model Parameters},
  author    = {Snell, Charlie and Lee, Jaehoon and Xu, Kelvin and Kumar, Aviral},
  journal   = {arXiv preprint arXiv:2408.03314},
  year      = {2024},
  url       = {https://arxiv.org/abs/2408.03314},
  note      = {arXiv v1 2024-08-06. Compute-optimal allocation of test-time compute (verifier search / revisions) as a function of prompt difficulty.}
}

@article{ye2024physics,
  title     = {Physics of Language Models: Part 2.1, Grade-School Math and the Hidden Reasoning Process},
  author    = {Ye, Tian and Xu, Zicheng and Li, Yuanzhi and Allen-Zhu, Zeyuan},
  journal   = {arXiv preprint arXiv:2407.20311},
  year      = {2024},
  url       = {https://arxiv.org/abs/2407.20311},
  note      = {arXiv v1 2024-07-29. Controlled synthetic study of reasoning; depth matters for multi-step reasoning more than width (cited as the "depth vs. reasoning" evidence).}
}

% ---------------------------------------------------------------------------
% 8. Zero-initialised gates / exact-identity insertion (for C3)
% ---------------------------------------------------------------------------

@article{bachlechner2020rezero,
  title     = {ReZero is All You Need: Fast Convergence at Large Depth},
  author    = {Bachlechner, Thomas and Majumder, Bodhisattwa Prasad and Mao, Huanru Henry and Cottrell, Garrison W. and McAuley, Julian},
  journal   = {arXiv preprint arXiv:2003.04887},
  year      = {2020},
  url       = {https://arxiv.org/abs/2003.04887},
  note      = {arXiv v1 2020-03-10, v2 2020-06-25. Mechanism: residual branch scaled by a scalar initialised at zero, so each block starts as the identity.}
}

@article{zhang2023controlnet,
  title     = {Adding Conditional Control to Text-to-Image Diffusion Models},
  author    = {Zhang, Lvmin and Rao, Anyi and Agrawala, Maneesh},
  journal   = {arXiv preprint arXiv:2302.05543},
  year      = {2023},
  url       = {https://arxiv.org/abs/2302.05543},
  note      = {ControlNet. arXiv v1 2023-02-10, v3 2023-11-26. Venue (ICCV 2023) NOT verified from arXiv metadata. Mechanism: "zero convolutions" so the added branch is exactly inactive at initialisation and the frozen backbone is reproduced. Code: lllyasviel/ControlNet.}
}

@inproceedings{alayrac2022flamingo,
  title     = {Flamingo: a Visual Language Model for Few-Shot Learning},
  author    = {Alayrac, Jean-Baptiste and Donahue, Jeff and Luc, Pauline and Miech, Antoine and Barr, Iain and Hasson, Yana and Lenc, Karel and Mensch, Arthur and Millican, Katie and Reynolds, Malcolm and Ring, Roman and Rutherford, Eliza and Cabi, Serkan and Han, Tengda and Gong, Zhitao and Samangooei, Sina and Monteiro, Marianne and Menick, Jacob and Borgeaud, Sebastian and Brock, Andrew and Nematzadeh, Aida and Sharifzadeh, Sahand and Binkowski, Mikolaj and Barreira, Ricardo and Vinyals, Oriol and Zisserman, Andrew and Simonyan, Karen},
  booktitle = {Advances in Neural Information Processing Systems (NeurIPS)},
  year      = {2022},
  eprint    = {2204.14198},
  archivePrefix = {arXiv},
  url       = {https://arxiv.org/abs/2204.14198},
  note      = {arXiv v1 2022-04-29; NeurIPS 2022 per arXiv comment. Mechanism: gated cross-attention layers with tanh gates initialised at zero so the frozen LM is reproduced exactly at initialisation.}
}

% ---------------------------------------------------------------------------
% 9. Interpretability-style sparse interventions (for C1/C2 context)
% ---------------------------------------------------------------------------

@article{ernst2026dsi,
  title     = {Distributed Sparse Interventions in Language Models},
  author    = {Ernst, Maximilian S. and Linhardt, Lorenz and Peikert, Aaron and Eberle, Oliver},
  journal   = {arXiv preprint arXiv:2607.07128},
  year      = {2026},
  url       = {https://arxiv.org/abs/2607.07128},
  note      = {arXiv v1 2026-07-08, v2 2026-10-22. Mechanism: steering via a sparse set (<0.1\%) of neurons chosen iteratively; interventions are static per task, not per-token learned actions.}
}

% ---------------------------------------------------------------------------
% 10. Backbone building blocks and scaling laws
% ---------------------------------------------------------------------------

@article{shazeer2020glu,
  title     = {{GLU} Variants Improve Transformer},
  author    = {Shazeer, Noam},
  journal   = {arXiv preprint arXiv:2002.05202},
  year      = {2020},
  url       = {https://arxiv.org/abs/2002.05202},
  note      = {arXiv v1 2020-02-12. SwiGLU feed-forward.}
}

@article{su2021roformer,
  title     = {RoFormer: Enhanced Transformer with Rotary Position Embedding},
  author    = {Su, Jianlin and Lu, Yu and Pan, Shengfeng and Murtadha, Ahmed and Wen, Bo and Liu, Yunfeng},
  journal   = {arXiv preprint arXiv:2104.09864},
  year      = {2021},
  url       = {https://arxiv.org/abs/2104.09864},
  note      = {arXiv v1 2021-04-20, v5 2023-11-08. Journal version (Neurocomputing 2024) NOT verified here. RoPE.}
}

@inproceedings{ainslie2023gqa,
  title     = {{GQA}: Training Generalized Multi-Query Transformer Models from Multi-Head Checkpoints},
  author    = {Ainslie, Joshua and Lee-Thorp, James and de Jong, Michiel and Zemlyanskiy, Yury and Lebr{\'o}n, Federico and Sanghai, Sumit},
  booktitle = {Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing (EMNLP)},
  year      = {2023},
  eprint    = {2305.13245},
  archivePrefix = {arXiv},
  url       = {https://arxiv.org/abs/2305.13245},
  note      = {arXiv v1 2023-05-22, v3 2023-12-23; EMNLP 2023 per arXiv comment.}
}

@inproceedings{zhang2019rmsnorm,
  title     = {Root Mean Square Layer Normalization},
  author    = {Zhang, Biao and Sennrich, Rico},
  booktitle = {Advances in Neural Information Processing Systems (NeurIPS)},
  year      = {2019},
  eprint    = {1910.07467},
  archivePrefix = {arXiv},
  url       = {https://arxiv.org/abs/1910.07467},
  note      = {arXiv v1 2019-10-16; NeurIPS 2019 per arXiv comment.}
}

@article{kaplan2020scaling,
  title     = {Scaling Laws for Neural Language Models},
  author    = {Kaplan, Jared and McCandlish, Sam and Henighan, Tom and Brown, Tom B. and Chess, Benjamin and Child, Rewon and Gray, Scott and Radford, Alec and Wu, Jeffrey and Amodei, Dario},
  journal   = {arXiv preprint arXiv:2001.08361},
  year      = {2020},
  url       = {https://arxiv.org/abs/2001.08361},
  note      = {arXiv v1 2020-01-23.}
}

@article{hoffmann2022chinchilla,
  title     = {Training Compute-Optimal Large Language Models},
  author    = {Hoffmann, Jordan and Borgeaud, Sebastian and Mensch, Arthur and Buchatskaya, Elena and Cai, Trevor and Rutherford, Eliza and de Las Casas, Diego and Hendricks, Lisa Anne and Welbl, Johannes and Clark, Aidan and Hennigan, Tom and Noland, Eric and Millican, Katie and van den Driessche, George and Damoc, Bogdan and Guy, Aurelia and Osindero, Simon and Simonyan, Karen and Elsen, Erich and Rae, Jack W. and Vinyals, Oriol and Sifre, Laurent},
  journal   = {arXiv preprint arXiv:2203.15556},
  year      = {2022},
  url       = {https://arxiv.org/abs/2203.15556},
  note      = {Chinchilla. arXiv v1 2022-03-29. Venue (NeurIPS 2022) NOT verified from arXiv metadata.}
}

@article{penedo2024fineweb,
  title = {The {FineWeb} Datasets: Decanting the Web for the Finest Text Data at Scale},
  author = {Penedo, Guilherme and Kydl{\'\i}{\v{c}}ek, Hynek and Ben allal, Loubna and Lozhkov, Anton and Mitchell, Margaret and Raffel, Colin and Von Werra, Leandro and Wolf, Thomas},
  year = {2024},
  journal = {arXiv preprint arXiv:2406.17557},
  url = {https://arxiv.org/abs/2406.17557},
  note = {Verified on the arXiv API 2026-10-30 (published 2024-06-25). FineWeb-Edu sample-10BT, ODC-By 1.0.}
}