mlsanityData to 13 Oct 2026

Uncovering Latent Reasoning Strategies in Language Models

arXiv 2607.17674 ↗First posted 20 Jul 2026cs.LG cs.AIcited by 0

On the map: LLM reasoning → Test-time compute and efficient thinking

Abstract (arXiv)

A language model pθ(y mid x) trained on reasoning tasks learns to solve problems via multiple distinct strategies, yet these strategies are implicit and entangled within the model's response distribution. We study the problem of decomposing the response distribution of a given pretrained language model into a structured, strategy-conditioned representation. Specifically, we learn a latent-variable factorization pθ(y mid x) leadsto (rφ(z mid x), gφ(y mid x,z)), where a router r maps each input to a distribution over latent strategies z and a generator g produces the response conditioned on that strategy. A key challenge is that the generator, initialized from the base model, already represents pθ(y mid x) without using z. Standard variational inference therefore gives the model no incentive to route information through z and can yield a severe form of posterior collapse. To address this, we propose a variational objective that measures fractional information gain relative to the base model's response loss and concentrates reconstruction pressure on tokens with high base model surprisal, encouraging z to encode strategy-relevant response variation. We introduce a benchmark of multi-strategy algorithmic tasks and show that this objective recovers latent codes aligned with distinct reference strategies while preserving the base model's response distribution.

Key facts. No citing papers in the map yet (data through 13 Oct 2026). Placed in Test-time compute and efficient thinking, one of 17 directions in LLM reasoning. Builds on 32 of its 40 references that are in the map. First posted on arXiv 20 Jul 2026.

Builds on

32 of 40 references.

Closest ancestors · 2

Same direction, newest first

  1. 2025Latent Thought Models with Variational Bayes Inference-Time Computation3 Feb
  2. 2023Tree of Thoughts: Deliberate Problem Solving with Large Language Models17 May

Other references · 30

  1. 2021LoRA: Low-Rank Adaptation of Large Language Modelscited by 14,231
  2. 2022Self-Consistency Improves Chain of Thought Reasoning in Language Modelscited by 5,657
  3. 2024Qwen2.5 Technical Reportcited by 4,840
  4. 2021Prefix-Tuning: Optimizing Continuous Prompts for Generationcited by 3,898
  5. 2017Outrageously Large Neural Networks: The Sparsely-Gated Mixture-of-Experts Layercited by 3,463
  6. 2023Sparse Autoencoders Find Highly Interpretable Features in Language Modelscited by 1,391
  7. 2020Concept Bottleneck Modelscited by 1,027
  8. 2025Does Reinforcement Learning Really Incentivize Reasoning Capacity in LLMs Beyond the Base Model?cited by 964
  9. 2016InfoGAN: Interpretable Representation Learning by Information Maximizing Generative Adversarial Netscited by 860
  10. 2019CTRL: A Conditional Transformer Language Model for Controllable Generationcited by 757
  11. 2018Disentangling by Factorisingcited by 614
  12. 2015Generating Sentences from a Continuous Spacecited by 609
18 more
  1. 2019Variational Autoencoders and Nonlinear ICA: A Unifying Frameworkcited by 498
  2. 2018Understanding disentangling in β-VAEcited by 334
  3. 2018Nonlinear ICA Using Auxiliary Variables and Generalized Contrastive Learningcited by 279
  4. 2017Learning Discourse-level Diversity for Neural Dialog Models using Conditional Variational Autoencoderscited by 222
  5. 2019Cyclical Annealing Schedule: A Simple Approach to Mitigating KL Vanishingcited by 191
  6. 2016A Hierarchical Latent Variable Encoder-Decoder Model for Generating Dialoguescited by 174
  7. 2019Lagging Inference Networks and Posterior Collapse in Variational Autoencoderscited by 124
  8. 2020Optimus: Organizing Sentences via Pre-trained Modeling of a Latent Spacecited by 114
  9. 2025The Invisible Leash: Why RLVR May or May Not Escape Its Origincited by 95
  10. 2019Generating Sentences from Disentangled Syntactic and Semantic Spacescited by 59
  11. 2025I Have Covered All the Bases Here: Interpreting Reasoning Features in Large Language Models via Sparse Autoencoderscited by 43
  12. 2023Interpreting Pretrained Language Models via Concept Bottleneckscited by 33
  13. 2022Controllable Text Generation via Probability Density Estimation in the Latent Spacecited by 27
  14. 2021Disentangling Generative Factors in Natural Language with Discrete Variational Autoencoderscited by 19
  15. 2025Representation-Based Exploration for Language Models: From Test-Time to Post-Trainingcited by 16
  16. 2021Finetuning Pretrained Transformers into Variational Autoencoderscited by 14
  17. 2022Learning Disentangled Representations of Negation and Uncertaintycited by 13
  18. 2023Graph-Induced Syntactic-Semantic Spaces in Transformer-Based Variational AutoEncoderscited by 6

Built on by

Not cited yet.

Similar, not cited

Close in wording, no citation either way

  1. 2026Latent Thought Flow: Efficient Latent Reasoning in Large Language Models
  2. 2025Variational Reasoning for Language Models
  3. 2024Language Models are Hidden Reasoners: Unlocking Latent Reasoning Capabilities via Self-Rewarding
  4. 2025A Survey on Latent Reasoning
  5. 2026Emergent Search and Backtracking in Latent Reasoning Models
  6. 2025RAVR: Reference-Answer-guided Variational Reasoning for Large Language Models
  7. 2024Markovian Transformers for Informative Language Modeling
  8. 2025On the Failure of Latent State Persistence in Large Language Models