mlsanityData to 7 Oct 2026

Entropy-Gated Latent Recursion

arXiv 2606.16620 ↗First posted 15 Jun 2026cs.LG cs.AIcited by 0

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

Abstract (arXiv)

Inference-time scaling has become the dominant lever for improving language-model reasoning, but existing methods derive rollout diversity from a single source: stochastic token-level sampling. We argue that this single-axis sampling space is fundamentally limiting, and identify a second, fully deterministic and complementary axis: the layer span L at which a frozen model's top decoder layers are recursively re-applied at high-uncertainty tokens. Different choices of L produce distinct rollouts that solve different subsets of problems, with no stochasticity. We instantiate this axis through Entropy-Gated Latent Recursion (EGLR), a training-free decoding procedure that re-applies the top-L layers for at most Kmax iterations until the next-token distribution converges. Combined with T temperature samples, EGLR turns a single-axis stochastic rollout pool into an L× T Cartesian sampling space at almost the same per-rollout cost. We characterize this space across 8 instruction-tuned models and 6 math reasoning benchmarks, and show that the L-axis is genuinely complementary to temperature: on MATH-500 with Qwen2.5-3B-Instruct, the joint L× T oracle reaches 91.6%, +8.2 percentage points beyond the temperature-only oracle (83.4%) and +10.4 points beyond the layer-only oracle (81.2%), confirming that the two axes capture genuinely complementary problems. The expanded rollout pool provides richer per-prompt candidates for any downstream procedure that consumes rollouts, including self-consistency, best-of-N with verifiers, and group-relative RL training (GRPO), opening a new direction for inference-time scaling that does not rely on stochastic noise.

Key facts. No citing papers in the map yet (data through 7 Oct 2026). Placed in Test-time compute and efficient thinking, one of 17 directions in LLM reasoning. Builds on 21 of its 25 references that are in the map. First posted on arXiv 15 Jun 2026.

Builds on

21 of 25 references.

Closest ancestors · 4

Same direction, newest first

  1. 2025Hierarchical Reasoning Model26 Jun
  2. 2025Entropy-Gated Branching for Efficient Test-Time Reasoning27 Mar
  3. 2024Scaling LLM Test-Time Compute Optimally can be More Effective than Scaling Model Parameters6 Aug
  4. 2023Tree of Thoughts: Deliberate Problem Solving with Large Language Models17 May

Other references · 17

  1. 2024The Llama 3 Herd of Modelscited by 14,192
  2. 2021Training Verifiers to Solve Math Word Problemscited by 9,743
  3. 2024DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Modelscited by 8,851
  4. 2022Self-Consistency Improves Chain of Thought Reasoning in Language Modelscited by 5,601
  5. 2021Measuring Mathematical Problem Solving With the MATH Datasetcited by 5,203
  6. 2024Qwen2.5 Technical Reportcited by 4,713
  7. 2023Let's Verify Step by Stepcited by 3,598
  8. 2022Competition-Level Code Generation with AlphaCodecited by 1,427
  9. 2022Solving Quantitative Reasoning Problems with Language Modelscited by 1,359
  10. 2024Qwen2.5-Math Technical Report: Toward Mathematical Expert Model via Self-Improvementcited by 1,057
  11. 2024Training Large Language Models to Reason in a Continuous Latent Spacecited by 763
  12. 2022Contrastive Decoding: Open-ended Text Generation as Optimizationcited by 598
5 more
  1. 2023DoLa: Decoding by Contrasting Layers Improves Factuality in Large Language Modelscited by 367
  2. 2020DeeBERT: Dynamic Early Exiting for Accelerating BERT Inferencecited by 322
  3. 2022Confident Adaptive Language Modelingcited by 297
  4. 2024Quiet-STaR: Language Models Can Teach Themselves to Think Before Speakingcited by 292
  5. 2023Looped Transformers as Programmable Computerscited by 224

Built on by

Not cited yet.

Similar, not cited

Close in wording, no citation either way

  1. 2026Sample Where You Struggle: Sharpening Base Model Reasoning via Entropy-Guided Power Sampling
  2. 2026Think Twice Before You Write -- an Entropy-based Decoding Strategy to Enhance LLM Reasoning
  3. 2025The Sequential Edge: Inverse-Entropy Voting Beats Parallel Self-Consistency at Matched Compute
  4. 2026Beyond the Frontier: Stochastic Backtracking for Efficient Test-Time Scaling
  5. 2025Rollout Roulette: A Probabilistic Inference Approach to Inference-Time Scaling of LLMs using Particle-Based Monte Carlo Methods
  6. 2025Encode, Think, Decode: Scaling test-time reasoning with recursive latent thoughts
  7. 2025Entropy Adaptive Decoding: Dynamic Model Switching for Efficient Inference
  8. 2026Entropic-Time Inference: Self-Organizing Large Language Model Decoding Beyond Attention