Scaling up Test-Time Compute with Latent Reasoning: A Recurrent Depth Approach
Open accessThe DOI matches its Crossref record; 134 of the 134 references it cites were found in a registry or an index.
Abstract
Depth-recurrent transformers scale test-time compute by iterating a core block in latent space before emitting tokens. The geometry of these latent trajectories is a named open problem in the latent reasoning literature (survey arXiv:2604.02029, §6.3). We report a pre-registered experiment on Huginn-0125 (arXiv:2502.05171), a 3.5B-parameter depth-recurrent transformer, measuring the per-head attention power-law exponent Δ of the recurrent core at recurrence counts r ∈ {1, 2, 4, 8, 16, 32}. For natural-text inputs the median exponent decreases monotonically with recurrence (Spearman ρ = −0.94), from 0.29 at r=1 to 0.239 at r=32 — converging onto the Sachdev–Ye–Kitaev (SYK) q=4 conformal value Δ = 1/4 — while the count of SYK-near heads grows monotonically (ρ = +0.77). Random-token inputs show an architecture-driven convergence with a transient at Δ ≈ 1/2 (the q=2 prethermal plateau previously observed in training time) but no monotone SYK-near growth. A randomized-weights control freezes completely: Δ_med constant to the sixth decimal across all r, zero SYK-near heads. Inference-time recurrence therefore acts as renormalization-group flow toward the conformal fixed point, and the flow is a property of the trained model, not the iteration procedure. Together with prior results on architectural depth (10.5281/zenodo.19225996) and training time, this completes a three-axis triangulation: the conformal fixed point is an attractor of iterative attending. The published orbit/spiral phenomenology of recurrent-depth latent trajectories acquires a candidate theory — the spiral is flow near an infrared fixed point — and the geometric benefit of additional recurrence saturates when the flow arrives, consistent with convergence-based early-exit criteria. All code, pre-registration (commit 702efd95), per-head data, and experiment notes with recorded protocol deviations are in the public repository 3ld0n/attention-geometry.
Overview
The claims and statistics of this paper haven’t been checked yet.
Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
The claims and statistics of this paper haven’t been checked yet.
Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
The cited source says what the claim attributes to it.
In the citing paper
“latent CoT—recurrent reasoning in the hidden state, without token traces [64,65,66]—may soften the feedforward bottleneck, though it still iterates over activations rather than weight-level self-models.”
From Self-Referential Introspection in Large Language Models: The Critical Threshold for Recursive Self-Improvement (2026) · cites this paper as [65]
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
The self-check needs the paper’s full text.
Citations
Cites
The publisher hasn’t deposited this paper’s reference list.
Research works printed but not linked: 36
- Brandon Amos and J. Zico Kolter. 2017. OptNet: Differentiable Optimization as a Layer in Neural Networks. In International Conference on Machine Learning , pages 136-145.
- Zhangir Azerbayev, Hailey Schoelkopf, Keiran Paster, Marco Dos Santos, Stephen Marcus McAleer, Albert Q. Jiang, Jia Deng, Stella Biderman, and Sean Welleck. 2023. Llemma: An Open Language Model for Mathematics. In The Twelfth International Conference on Learning Representations .
- Shaojie Bai, J. Zico Kolter, and Vladlen Koltun. 2019. Deep Equilibrium Models. In Advances in Neural Information Processing Systems , volume 32. Curran Associates, Inc.
- Shaojie Bai, Vladlen Koltun, and J. Zico Kolter. 2022. Neural Deep Equilibrium Solvers. In International Conference on Learning Representations .
- Andrea Banino, Jan Balaguer, and Charles Blundell. 2021. PonderNet: Learning to Ponder. In 8th ICML Workshop on Automated Machine Learning (AutoML) .
- Valentino Braitenberg. 1986. Vehicles: Experiments in Synthetic Psychology . MIT press.
- Tianle Cai, Yuhong Li, Zhengyang Geng, Hongwu Peng, Jason D. Lee, Deming Chen, and Tri Dao. 2024. Medusa: Simple LLM Inference Acceleration Framework with Multiple Decoding Heads. In Forty-First International Conference on Machine Learning . character.ai. 2024. Optimizing AI Inference at Character.AI. https://research.character.ai/optimizing-inference/
- Hantian Ding, Zijian Wang, Giovanni Paolini, Varun Kumar, Anoop Deoras, Dan Roth, and Stefano Soatto. 2024. Fewer Truncations Improve Language Modeling. In Forty-First International Conference on Machine Learning .
- Maha Elbayad, Jiatao Gu, Edouard Grave, and Michael Auli. 2019. Depth-Adaptive Transformer. In International Conference on Learning Representations .
- Ying Fan, Yilun Du, Kannan Ramchandran, and Kangwook Lee. 2025. Looped Transformers for Length Generalization. In The Thirteenth International Conference on Learning Representations .
- Jonas Geiping and Tom Goldstein. 2023. Cramming: Training a Language Model on a single GPU in one day. In Proceedings of the 40th International Conference on Machine Learning , pages 11117-11143. PMLR.
- Angeliki Giannou, Shashank Rajput, Jy-Yong Sohn, Kangwook Lee, Jason D. Lee, and Dimitris Papailiopoulos. 2023. Looped Transformers as Programmable Computers. In Proceedings of the 40th International Conference on Machine Learning , pages 11398-11442. PMLR.
- Tamir David Hay and Lior Wolf. 2023. Dynamic Layer Tying for Parameter-Efficient Transformers. In The Twelfth International Conference on Learning Representations .
- Dan Hendrycks, Collin Burns, Steven Basart, Andy Zou, Mantas Mazeika, Dawn Song, and Jacob Steinhardt. 2021a. Measuring massive multitask language understanding. Proceedings of the International Conference on Learning Representations (ICLR) .
- Dan Hendrycks, Collin Burns, Steven Basart, Andy Zou, Mantas Mazeika, Dawn Song, and Jacob Steinhardt. 2021b. Measuring Massive Multitask Language Understanding. In International Conference on Learning Representations .
- Pavel Izmailov, Dmitrii Podoprikhin, Timur Garipov, Dmitry Vetrov, and Andrew Gordon Wilson. 2018. Averaging weights leads to wider optima and better generalization: 34th Conference on Uncertainty in Artificial Intelligence 2018, UAI 2018. 34th Conference on Uncertainty in Artificial Intelligence 2018, UAI 2018 , pages 876-885.
- Angelos Katharopoulos, Apoorv Vyas, Nikolaos Pappas, and François Fleuret. 2020. Transformers are RNNs: Fast Autoregressive Transformers with Linear Attention. In Proceedings of the 37th International Conference on Machine Learning , pages 5156-5165. PMLR.
- Diederik P. Kingma and Jimmy Ba. 2015. Adam: A Method for Stochastic Optimization. In International Conference on Learning Representations (ICLR) , San Diego.
- Zhenzhong Lan, Mingda Chen, Sebastian Goodman, Kevin Gimpel, Piyush Sharma, and Radu Soricut. 2019. ALBERT: A Lite BERT for Self-supervised Learning of Language Representations. In International Conference on Learning Representations .
- Yann LeCun. 2022. A Path Towards Autonomous Machine Intelligence. Preprint , Version 0.9.2:62.
- Yann LeCun and Fu Jie Huang. 2005. Loss functions for dis- criminative training of energy-based models. In AISTATS 2005 https://nyuscholars.nyu.edu/en/publications/loss-functions-for-discriminative-training-of-energy-based-models
- Yaniv Leviathan, Matan Kalman, and Yossi Matias. 2023. Fast Inference from Transformers via Speculative Decoding. In Proceedings of the 40th International Conference on Machine Learning , pages 19274-19286. PMLR.
- Mrinal Mathur, Barak A. Pearlmutter, and Sergey M. Plis. 2024. MIND over Body: Adaptive Thinking using Dynamic Computation. In The Thirteenth International Conference on Learning Representations .
- Neel Nanda, Lawrence Chan, Tom Lieberum, Jess Smith, and Jacob Steinhardt. 2022. Progress measures for grokking via mechanistic interpretability. In The Eleventh International Conference on Learning Representations .
- Keiran Paster, Marco Dos Santos, Zhangir Azerbayev, and Jimmy Ba. 2023. OpenWebMath: An Open Dataset of High-Quality Mathematical Web Text. In The Twelfth International Conference on Learning Representations .
- Victor Sanh, Albert Webson, Colin Raffel, Stephen Bach, Lintang Sutawika, Zaid Alyafeai, Antoine Chaffin, Arnaud Stiegler, Arun Raja, Manan Dey, M. Saiful Bari, Canwen Xu, Urmish Thakker, Shanya Sharma Sharma, Eliza Szczechla, Taewoon Kim, Gunjan Chhablani, Nihal Nayak, Debajyoti Datta, and 21 others. 2021. Multitask Prompted Training Enables Zero-Shot Task Generalization. In International Conference on Learning Representations .
- Sunny Sanyal, Atula Tejaswi Neerkaje, Jean Kaddour, Abhishek Kumar, and sujay sanghavi. 2024. Early weight averaging meets high learning rates for LLM pre-training. In First Conference on Language Modeling .
- Avi Schwarzschild. 2023. Deep Thinking Systems: Logical Extrapolation with Recurrent Neural Networks . Ph.D. thesis, University of Maryland, College Park, College Park.
- Avi Schwarzschild, Eitan Borgnia, Arjun Gupta, Furong Huang, Uzi Vishkin, Micah Goldblum, and Tom Goldstein. 2021b. Can You Learn an Algorithm? Generalizing from Easy to Hard Problems with Recurrent Networks. In Advances in Neural Information Processing Systems , volume 34, pages 6695-6706. Curran Associates, Inc.
- Yu Sun, Xiaolong Wang, Zhuang Liu, John Miller, Alexei Efros, and Moritz Hardt. 2020. Test-Time Training with SelfSupervision for Generalization under Distribution Shifts. In Proceedings of the 37th International Conference on Machine Learning , pages 9229-9248. PMLR.
- Ilya Sutskever, Geoffrey E Hinton, and Graham W Taylor. 2008. The Recurrent Temporal Restricted Boltzmann Machine. In Advances in Neural Information Processing Systems , volume 21. Curran Associates, Inc.
- Mitchell Wortsman, Tim Dettmers, Luke Zettlemoyer, Ari S. Morcos, Ali Farhadi, and Ludwig Schmidt. 2023b. Stable and low-precision training for large-scale vision-language models. In Thirty-Seventh Conference on Neural Information Processing Systems .
- Yuhuai Wu, Markus Norman Rabe, DeLesley Hutchins, and Christian Szegedy. 2022. Memorizing Transformers. In International Conference on Learning Representations .
- Liu Yang, Kangwook Lee, Robert D. Nowak, and Dimitris Papailiopoulos. 2024a. Looped Transformers are Better at Learning Learning Algorithms. In The Twelfth International Conference on Learning Representations .
- Longhui Yu, Weisen Jiang, Han Shi, Jincheng Yu, Zhengying Liu, Yu Zhang, James Kwok, Zhenguo Li, Adrian Weller, and Weiyang Liu. 2023. MetaMath: Bootstrap Your Own Mathematical Questions for Large Language Models. In The Twelfth International Conference on Learning Representations .
- Yifan Zhang, Yifan Luo, Yang Yuan, and Andrew C. Yao. 2024c. Autonomous Data Selection with Language Models for Mathematical Texts. In ICLR 2024 Workshop on Navigating and Addressing Data Problems for Foundation Models .
Cited by
Items 101–200 of 361
Deep Thinking by Markov Chain of Continuous ThoughtsPreprint
Liu, J., Huang, Z., Yang, X., Ji, T., Sims, A., Xu, H., Chen, E., Teh, Y. W., & Miao, N. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2509.25020 · Cited by 12
Bridging Search and Recommendation through Latent Cross ReasoningPreprint
Shi, T., Qin, W., Yu, W., Zhang, X. Y., He, M., Jianping Fan, B., & Xu, J. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2508.04152 · Cited by 8
Chain-of-Experts: Unlocking the Communication Power of Mixture-of-Experts ModelsPreprint
Wang, Z., Pan, R., Yao, J., Csordás, R., Li, L., Yin, L., Wu, J., Zhang, T., Li, M., & Liu, S. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2506.18945 · Cited by 13
Mixture of Universal Experts: Scaling Virtual Width via Depth-Width TransformationPreprint
Chen, Y., Gu, N., Shang, J., Zhang, Z., Feng, Y., Sheng, J., Liu, T., Wang, S., Sun, Y., Wu, H., et al. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2603.04971 · Cited by 9
Representational Collapse in Multi-Agent LLM Committees: Measurement and Diversity-Aware ConsensusPreprint
Patel, D. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2604.03809 · Cited by 10
LASAR: Latent Adaptive Semantic Aligned Reasoning for Generative RecommendationPreprint
Chen, Y., Zhang, F., Chen, Z., Li, H., Xu, P., Liu, H., Liu, H., Pei, X., Zhuang, F., & Zhang, Z. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2605.10207 · Cited by 8
Shapiro, M. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2608.11233 · Cited by 9
Gated Recurrent Transformers: Expressive Depth through Recurrent ModulationPreprint
Hegazy, A., Alanwar, A., & Elhoushi, M. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2608.15062 · Cited by 8
Think Shallow, Solve Deep: Controlling Recurrent Dynamics for Reliable Test-Time DepthPreprint
Viakhirev, I., Borodin, K., Almutairi, A., Barannikov, S., Abramov, M. V., & Mkrtchian, G. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2608.18222 · Cited by 8
Towards Looped Models Done Right, Part II: Rethinking at Fixed PointsPreprint
Huang, B., Shi, C.-F., Chen, J.-L., Wen, S.-C., Liu, Z.-Z., Xing, E., & Ma, X.-Z. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2610.06833 · Cited by 10
The Big Send-off: Scalable and Performant Collectives for Deep Learning
Singh, S., Pradeep, K., Singh, M., Wei, C., & Bhatele, A. S. (2026) · IEEE International Parallel and Distributed Processing Symposium (IPDPS) · doi:10.1109/ipdps65963.2026.00108 · Cited by 7
Self-citationSeLaR: Selective Latent Reasoning in Large Language Models
Fu, R., & Luo, G. (2026) · Proceedings of the conference - Association for Computational Linguistics. Meeting · doi:10.18653/v1/2026.acl-long.320 · Cited by 13
Factorized Latent Reasoning for LLM-based Recommendation
Gao, T., Huang, C., Wang, Z., Liu, C., Zeng, K., & Yao, L. (2026) · International ACM SIGIR Conference on Research and Development in Information Retrieval (SIGIR) · doi:10.1145/3805712.3809559 · Cited by 7
Artificial Kuramoto Oscillatory Neurons
Miyato, T. (2025) · The Brain & Neural Networks · doi:10.3902/jnns.32.127 · Cited by 10
Exploring System 1 and 2 communication for latent reasoning in LLMsPreprint
Coda-Forno, J., Zhao, Z., Zhang, Q., Tamboli, D., Li, W., Fan, X., Zhang, L., Schulz, E., & Tseng, H.-P. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2510.00494 · Cited by 7
Universal Redundancies in Time Series Foundation ModelsPreprint
Bao, A., Vattikuti, V. H., Lai, J., & Gilpin, W. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2602.01605 · Cited by 8
Multimodal Latent Reasoning via Hierarchical Visual Cues InjectionPreprint
Zhang, Y., Yan, Q., Jiang, B., & Han, K. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2602.05359 · Cited by 7
Towards Explainable Industrial Anomaly Detection via Knowledge-Guided Latent ReasoningPreprint
Chen, P., Wang, W., Cao, Y., Liu, C., Wang, W., Yang, M., Shen, L., Ren, W., Cao, X., & Huang, C. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2602.09850 · Cited by 8
Looped Diffusion Language ModelsPreprint
Lee, S., Hong, C., Kim, S., Lee, J., Park, J.-H., & Park, D. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2605.26106 · Cited by 7
Stabilizing Extrapolation in Looped Transformers via Learned Stochastic StoppingPreprint
Kuo, H.-Y., Chayti, E. M., Reizinger, P., Brendel, W., & Jaggi, M. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2606.29983 · Cited by 7
LoopSpec: Pipelined Self-Speculative Decoding for Looped TransformersPreprint
Cho, S., Cui, L.-Q., Kim, S., Han, D.-S., & Han, I. (2026) · arXiv.org · doi:10.48550/arxiv.2609.17184 · Cited by 10
Anthropocentric Bias in Language Model Evaluation
Millière, R., & Rathkopf, C. (2026) · Computational Linguistics · doi:10.1162/coli.a.582 · Cited by 6
Silence the Judge: Reinforcement Learning with Self-Verifier via Latent Geometric Clustering
Zhang, N., Ma, W., Ma, Z., Xu, J., Gao, J., Hao, J., He, R., & Xu, J. (2026) · Conference on Empirical Methods in Natural Language Processing (EMNLP) · doi:10.18653/2026.emnlp-main.2569 · Cited by 6
MeSH: Memory-as-State-Highways for Recursive Transformers
Yu, C., Shu, X., Wang, Y., Zhang, Y., Wu, H., Li, J., Long, R., Chen, Z., Xu, Y. C., Su, W., et al. (2025) · International Conference on Learning Representations · doi:10.48550/arxiv.2510.07739 · Cited by 6
JEPA-Reasoner: Decoupling Latent Reasoning from Token GenerationPreprint
Liu, B. K., Chen, Z. P., & Woodruff, D. P. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2512.19171 · Cited by 6
Inverse Depth Scaling From Most Layers Being SimilarPreprint
Liu, Y., Kangaslahti, S., Liu, Z.-M., & Gore, J. (2026) · arXiv.org · doi:10.48550/arxiv.2602.05970 · Cited by 12
Pretraining with Token-Level Adaptive Latent Chain-of-ThoughtPreprint
Zeng, B., Hao, Y., Li, H., Song, S., Song, F., Wang, Z., Huang, S., Xu, Y., He, Z., Wang, X., et al. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2602.08220 · Cited by 6
Tiny Recursive Reasoning with Mamba-2 Attention HybridPreprint
Wang, W.-L., & Reid, F. (2026) · arXiv.org · doi:10.48550/arxiv.2602.12078 · Cited by 7
Capps, C. A. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2606.01495 · Cited by 6
Pretraining Recurrent Networks without RecurrencePreprint
Kumar, A., & Isola, P. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2606.06479 · Cited by 6
When More is Less: Understanding Chain-of-Thought Length in LLMs
Wu, Y., Wang, Y., Ye, Z., Du, T., Jegelka, S., & Wang, Y. (2025) · arXiv · doi:10.48550/arxiv.2502.07266 · Cited by 5
Transformers as Intrinsic Optimizers: Forward Inference through the Energy PrinciplePreprint
Ren, R., Ouyang, S., Tang, H., & Liu, Y. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2511.00907 · Cited by 5
Unsupervised decoding of encoded reasoning using language model interpretabilityPreprint
Fang, C., & Marks, S. (2025) · arXiv.org · doi:10.48550/arxiv.2512.01222 · Cited by 4
Unlocking Out-of-Distribution Generalization in Transformers via Recursive Latent Space ReasoningPreprint
Altabaa, A., Chen, S., Lafferty, J., & Yang, Z. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2510.14095 · Cited by 6
PonderLM-3: Adaptive Token-Wise Pondering with Differentiable MaskingPreprint
Li, H., Song, F., Zeng, B., Song, S., Xu, Z. J., He, Z., & Lin, Z. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2603.02023 · Cited by 5
Think Fast: Estimating No-CoT Task-Completion Time Horizons of Frontier AI ModelsPreprint
Gould, D., Ward, F. R., Woodruff, A. C., Arike, R., Hills, J., Serrano, A., Caspary, I., Brown, J. R., Jiao, J. J., Leask, P., et al. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2606.07157 · Cited by 5
Dense Supervision Is Not Enough: The Readout Blind Spot in Looped Language ModelsPreprint
Sharma, R., & Vu, T. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2606.24898 · Cited by 6
Hidden Decoding at Scale: Latent Computation Scaling for Large Language ModelsPreprint
Liu, A., Shi, C., Wu, C. E., Lei, C., Lu, D. H., He, D., Zhang, F., Kong, F., Zhang, F., Wang, G., et al. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2607.08186 · Cited by 5
Closing the Loop: Practical Training Recipes for Looped Language ModelsPreprint
Marchenko, A., Bezrukov, V., Kashurin, O., Fedorova, I., Bocharov, D., Shakhvalieva, Y., Tikhonova, M., & Ternovskii, V. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2610.00673 · Cited by 5
MoEs Are Stronger than You Think: Hyper-Parallel Inference Scaling with RoEs
Zibakhsh, S., Samragh, M., Nishu, K., Hannah, L. A., Kundu, A., & Cho, M. (2026) · Findings of the Association for Computational Linguistics: EMNLP (Findings of EMNLP) · doi:10.18653/2026.findings-emnlp.2143 · Cited by 4
Li, C., Liu, X., Zhang, Z., Liu, S., Ma, G., Lan, Y., Wang, C., & Shen, C. (2026) · Proceedings of the conference - Association for Computational Linguistics. Meeting · doi:10.18653/v1/2026.acl-long.1810 · Cited by 4
Recurrent Self-Attention Dynamics: An Energy-Agnostic Perspective from Jacobians
Tomihari, A., & Karakida, R. (2025) · neural information processing systems · doi:10.52202/085713-4675 · Cited by 8
SCOUT: Teaching Pre-trained Language Models to Enhance Reasoning via Flow Chain-of-Thought
Li, G., Jiang, W., Chen, M., Li, Y., Yu, H., Dong, S., Ren, T., Tang, M., & Yuan, C. (2025) · neural information processing systems · doi:10.52202/085713-3187 · Cited by 6
ReaCritic : Reasoning Transformer-Based DRL Critic-Model Scaling for Wireless Networks
You, F., & Du, H. (2026) · IEEE Transactions on Mobile Computing · doi:10.1109/tmc.2026.3716970 · Cited by 6
Srinivas, S. S., Das, A., Gupta, S., & Runkana, V. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2504.01281 · Cited by 5
Predictability Enables Parallelization of Nonlinear State Space ModelsPreprint
Gonzalez, X., Kozachkov, L., Zoltowski, D. M., Clarkson, K. L., & Linderman, S. W. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2508.16817 · Cited by 5
Step-resolved data attribution for looped transformersPreprint
Kaissis, G. A., Mildenberger, D., Gomez, J. F., Menten, M. J., & Triantafillou, E. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2602.10097 · Cited by 4
COLD-Steer: Steering Large Language Models via In-Context One-step Learning Dynamics
Sharma, K., & Trivedi, R. (2026) · International Conference on Learning Representations · doi:10.48550/arxiv.2603.06495 · Cited by 4
Sparse Growing Transformer: Training-Time Sparse Depth Allocation via Progressive Attention LoopingPreprint
Chen, Y., Chen, Y., Yang, Y., Shang, J., Zhang, Z., Zhang, Z., Nie, S., Wang, S., Sun, Y., Wu, H., et al. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2603.23998 · Cited by 5
One Step Forward and K Steps Back: Better Reasoning with Denoising Recursion ModelsPreprint
Cameron, C., Wang, W., Ivanov, N., Bhattacharyya, A., Chételat, D., & Zhang, Y. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2604.18839 · Cited by 6
Looped World ModelsPreprint
Lu, H. A., Wei, Z. L. V., Zhang, Q., Zeng, J., Cao, B., Meng, L., Li, M., Wang, Z., Yin, H., Xue, N., et al. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2606.18208 · Cited by 4
Per-Token Fixed-Point Convergence in Depth-Recurrent TransformersPreprint
Logan, J. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2607.14427 · Cited by 4
Allocating Recurrent Compute in Looped Language ModelsPreprint
Lin, R., Guo, Y., Zhu, R.-J., Ye, H., & Eshraghian, J. K. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2608.18230 · Cited by 4
Think Before Recommend: Unleashing the Latent Reasoning Power for Sequential Recommendation
Tang, J., Dai, S., Shi, T., Xu, J., Chen, X., Chen, W., Wu, J., & Jiang, Y. (2026) · IEEE Transactions on Knowledge and Data Engineering · doi:10.1109/tkde.2026.3694421 · Cited by 3
Unlocking the Black Box of Latent Reasoning: An Interpretability-Guided Approach to Intervention
Chang, S., Bai, T., Zhang, X., Ma, Q., Liu, Q., Liao, Z., Miao, Y., & Niu, L. (2026) · Proceedings of the conference - Association for Computational Linguistics. Meeting · doi:10.18653/v1/2026.acl-long.1568 · Cited by 8
RecRec: Latent Interests Recursive Reasoning for Sequential Recommendation
Deng, W., Fu, J., Du, H., Karatzoglou, A., Arapakis, I., Guo, H., Zheng, K., Ni, Y., & Jose, J. M. (2026) · ACM Conference on Recommender Systems (RecSys) · doi:10.1145/3773078.3831821 · Cited by 3
Can large language models correctly interpret equations with errors?
McGinness, L. P., & Baumgartner, P. (2025) · Physical Review Physics Education Research · doi:10.1103/v8f8-s11v · Cited by 4
Latent Flow TransformerPreprint
Wu, Y.-C., Liao, F.-T., Chen, M.-H., Ho, P.-C., Nabiei, F., & Shiu, D.-S. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2505.14513 · Cited by 3
Chipmunk: Training-Free Acceleration of Diffusion Transformers with Dynamic Column-Sparse DeltasPreprint
Silveria, A., Govande, S. V., & Fu, D. Y. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2506.03275 · Cited by 6
SpikingMamba: Towards Energy-Efficient Large Language Models via Knowledge Distillation from Mamba
Huang, Y., Tang, J., Wang, C., Wang, Z., Zhang, J., Lu, Z., Cheng, B., & Leng, L. (2025) · Transactions on machine learning research · doi:10.48550/arxiv.2510.04595 · Cited by 3
Controllably Efficient Language ModelsPreprint
Prakash, J., Puli, A., & Ranganath, R. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2511.05313 · Cited by 3
Emergent temporal abstractions in autoregressive models enable hierarchical reinforcement learningPreprint
Kobayashi, S., Schimpf, Y., Schlegel, M., Steger, A., Wołczyk, M., von Oswald, J., Scherrer, N., Maile, K., Lajoie, G., Richards, B. A., et al. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2512.20605 · Cited by 5
Stop-Think-AutoRegress: Language Modeling with Latent Diffusion PlanningPreprint
Lovelace, J., Belardi, C. K., Zalouk, S., Polavaram, A., Kundurthy, S., & Weinberger, K. Q. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2602.20528 · Cited by 5
$\nabla$-Reasoner: LLM Reasoning via Test-Time Gradient Descent in Latent Space
Wang, P., Cai, R., Wang, Z., Mei, H., Liu, Q., Li, P., & Wang, Z. (2026) · International Conference on Learning Representations · doi:10.48550/arxiv.2603.04948 · Cited by 7
MedLVR: Latent Visual Reasoning for Reliable Medical Visual Question AnsweringPreprint
Xi, S., Hu, S., Lai, Y., Dan, W., Liu, Y., Wang, S., & Yang, X. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2604.09757 · Cited by 5
Self-Consistent Latent Reasoning: Long Latent Sequence Reasoning for Vision-Language ModelPreprint
Wang, C., He, W., Zhu, X., Zhou, C., Li, Q., Yan, S., Zheng, Y., Yu, C., Lu, F., Zhai, W., et al. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2605.12163 · Cited by 3
TTE-Flash: Accelerating Reasoning-based Multimodal Representations via Think-Then-Embed TokensPreprint
Cheng, J., Wu, X., Zhang, J., Bao, W., Ahuja, C., Mishra, S. K., Cui, X., Yu, H., Gao, Y., Xia, F., et al. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2605.16638 · Cited by 3
Tying the Loop -- Tied Expert Layers in Mixture-of-Experts Language ModelsPreprint
Jaggi, M. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2606.16825 · Cited by 3
Recursive Scaling in Masked Diffusion ModelsPreprint
Carballo-Castro, A., Piskorz, J., Rauba, P., van der Schaar, M., & Frossard, P. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2606.18022 · Cited by 5
Don't Commit Alone: Joint Token Commitment in Diffusion Language ModelsPreprint
Lin, Y. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2607.04469 · Cited by 3
Spectral Margins and Memory Routing in Looped TransformersPreprint
Buehlmaier, M. M. M. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2607.10681 · Cited by 3
Looped Language Models Improve Compositional Tool CallingPreprint
Popescu, A. C., Borde, H. S. D. O., & Lió, P. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2608.18171 · Cited by 3
Dynamical phase selection controls compute scaling in looped transformersPreprint
Kim, G. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2608.26556 · Cited by 3
Quantizing Looped Transformers: Feedback Exposure and Calibration BlindnessPreprint
Li, N. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2609.30820 · Cited by 3
Improving Test-Time Scaling with Adaptive Looped TransformersPreprint
You, Y., Fu, T., Feng, A., Lv, X., Ning, X., Ding, N., & Wang, Y. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2609.35748 · Cited by 5
Towards Looped Models Done Right, Part I: Topology, Input Injection, Recurrent-State DesignPreprint
Huang, B., Shi, C., Chen, J., Wen, S., Liu, Z., Xing, E. P., & Ma, X. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2610.23036 · Cited by 11
Zhang, J., Yuan, B., & Zhang, Q. (2026) · Entropy · doi:10.3390/e28090951 · Cited by 2
LSRL: Process-Supervised GRPO on Latent Recurrent States Improves Mathematical Reasoning
Ren, H. (2025) · Findings of the Association for Computational Linguistics: EMNLP (Findings of EMNLP) · doi:10.18653/v1/2025.findings-emnlp.669 · Cited by 7
Social Genome: Grounded Social Reasoning Abilities of Multimodal Models
Mathur, L., Qian, M., Liang, P. P., & Morency, L. (2025) · Conference on Empirical Methods in Natural Language Processing (EMNLP) · doi:10.18653/v1/2025.emnlp-main.1264 · Cited by 23
FOL-Traces: Verified First-Order Logic Reasoning Traces at Scale
Lee, I., Liaw, S., & Yogatama, D. (2026) · Findings of the Association for Computational Linguistics: EACL · doi:10.18653/v1/2026.findings-eacl.115 · Cited by 3
Beyond Static Representation: Coarse-to-Fine Dynamic Latent Reasoning for Sequential Recommendation
Yang, X., Xu, Y., Wang, X., Wang, H., Jiang, Z., Chen, Z., & Xing, Z. (2026) · Lecture notes in computer science · doi:10.1007/978-981-92-0363-5_16 · Cited by 2
VeriThinker: Learning to Verify Makes Reasoning Model Efficient
Chen, Z., Ma, X., Fang, G., Yu, R., & Wang, X. (2025) · neural information processing systems · doi:10.52202/085713-4337 · Cited by 4
Intra-Layer Recurrence in Transformers for Language Modeling
Nguyen, A., & Lin, W. (2025) · doi:10.21428/594757db.834c24c6 · Cited by 5
Expressive Power of Implicit Models: Rich Equilibria and Test-Time ScalingPreprint
Liu, J., Ding, L., Osher, S., & Yin, W. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2510.03638 · Cited by 5
Alternatives To Next Token Prediction In Text Generation -- A SurveyPreprint
Wyatt, C., Joshi, A., & Salim, F. D. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2509.24435 · Cited by 2
Towards Distributed Neural ArchitecturesPreprint
Cowsik, A., He, T., & Gromov, A. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2506.22389 · Cited by 3
A Unifying Framework for Parallelizing Sequential Models with Linear Dynamical Systems
Gonzalez, X., Buchanan, E. K., Lee, H. D., Liu, J. W., Wang, K. A., Zoltowski, D. M., Kozachkov, L., Ré, C., & Linderman, S. W. (2026) · Transactions on machine learning research · doi:10.48550/arxiv.2509.21716 · Cited by 2
BigBang-Proton Technical Report: Next-Word-Prediction is Scientific Multitask LearnerPreprint
Wu, H., Liu, L., He, J., Wang, Q., Zhao, K., Hu, S., Fu, R., Liang, D., Zeng, L., Liu, B. X., et al. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2510.00129 · Cited by 2
A foundation model with multi-variate parallel attention to generate neuronal activityPreprint
Carzaniga, F. S., Hersche, M., Sebastian, A., Schindler, K. A., & Rahimi, A. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2506.20354 · Cited by 2
Inner Loop Inference for Pretrained Transformers: Unlocking Latent Capabilities Without TrainingPreprint
Lys, J., Gripon, V., Pasdeloup, B., Marmoret, A., Mauch, L., Cardinaux, F., & Hacene, G. B. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2602.14759 · Cited by 2
Turbo Connection: Reasoning as Information Flow from Higher to Lower LayersPreprint
Tang, M., & Lu, S. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2602.17993 · Cited by 2
Thinking into the Future: Latent Lookahead Training for TransformersPreprint
Noci, L., Bachmann, G., Moosavi-Dezfooli, S.-M., & Nabi, M. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2603.20219 · Cited by 2
LEPO: Latent Reasoning Policy Optimization for Large Language ModelsPreprint
Zhou, Y., Yu, J., Dong, H., Hao, Z., Wang, H., Zhang, J., & Lin, Q. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2604.17892 · Cited by 5
ReMoA: Recursive Modular Adaptation via Sparse Latent Refinement in Multimodal Diffusion ModelsPreprint
Sun, Y., Yao, Y., Li, H., Yuan, W., & Zhu, S. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2604.25299 · Cited by 2
Transformers Efficiently Perform In-Context Logistic Regression via Normalized Gradient DescentPreprint
Zhang, C., & Zhi-yuan, C. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2605.06609 · Cited by 2
Scratchpad Patching: Decoupling Compute from Patch Size in Byte-Level Language ModelsPreprint
Zheng, L., Bashlovkina, V., Dozat, T., Garrette, D., Rimell, L., & Maynez, J. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2605.09630 · Cited by 2
A Dual-Path Architecture for Scaling Compute and Capacity in LLMsPreprint
Frey, M., Shomali, B., Koehler, J., & Ali, M. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2605.30202 · Cited by 2
Anand, A., Ramesh, M., Mittal, A., Kumar, A., Vyalla, R. R., Cambria, E., Wang, Z., Liu, T., Ng, A. B., See, S., et al. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2606.11470 · Cited by 3
CHERRY: Compressed Hierarchical Experts with Recurrent Representational YieldPreprint
Kwon, D., & Park, Y. (2026) · arXiv.org · doi:10.48550/arxiv.2606.31796 · Cited by 2
Multimodal Continuous Reasoning via Asymmetric Mutual Variational LearningPreprint
Li, S., Gao, Y., Yang, S., Chen, T. X., Gan, C., He, Z., Zhao, Z., Guo, Y., Lin, W., & Yu, H. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2607.00461 · Cited by 2
Metadata
- Authors
- Geiping, J., McLeish, S., Jain, N., Kirchenbauer, J., Singh, S., Bartoldson, B., Kailkhura, B., Bhatele, A., & Goldstein, T.
- Affiliations
- Tübingen AI Center (DE)
- University of Maryland, College Park (US)
- Lawrence Livermore National Laboratory (US)
- ORCID
- Venue
- Advances in Neural Information Processing Systems 38
- Published
- 2025
- Pages
- 46242-46293
- ISSN
- 1049-5258 (ISSN-L)
- OpenAlex
- W4407310640
- Cited by
- 10
- References
- 0
- Citations by year
- 2026: 10
- FWCI
- 9.53
- Type
- conference paper
- Language
- en
- Open access
- green
- arxiv.org
- Locations
- neural information processing systems (publishedVersion)
- arXiv (Cornell University) (submittedVersion)
- arXiv (Cornell University)
- Topics
- Physical Sciences › Computer Science › Artificial Intelligence › Topic Modeling
- Physical Sciences › Computer Science › Artificial Intelligence › Explainable Artificial Intelligence (XAI)
- Physical Sciences › Computer Science › Artificial Intelligence › Natural Language Processing Techniques
- Keywords
- latent reasoning, test-time compute, Sachdev-Ye-Kitaev model, conformal fixed points, prethermalization, renormalization group flow, latent trajectories
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.