# Efficient Post-Training Refinement of Latent Reasoning in Large Language Models

arXiv: https://arxiv.org/abs/2506.08552 · HTML: https://mlsanity.com/paper/2506.08552/

## Key facts

- Cited by 7 papers in the map (data through 14 Oct 2026).
- Placed in Test-time compute and efficient thinking, one of 17 directions in LLM reasoning.
- Builds on 27 of its 65 references that are in the map.
- First posted on arXiv 10 Jun 2025.

## Placed in

- LLM reasoning / [Test-time compute and efficient thinking](https://mlsanity.com/lr/test-time-compute-and-efficient-thinking/)

## Builds on (most cited first, 27 of 27)

- [Training Verifiers to Solve Math Word Problems](https://mlsanity.com/paper/2110.14168/) (2021)
- [Self-Consistency Improves Chain of Thought Reasoning in Language Models](https://mlsanity.com/paper/2203.11171/) (2022)
- [Emergent Abilities of Large Language Models](https://mlsanity.com/paper/2206.07682/) (2022)
- [Least-to-Most Prompting Enables Complex Reasoning in Large Language Models](https://mlsanity.com/paper/2205.10625/) (2022)
- [Program of Thoughts Prompting: Disentangling Computation from Reasoning for Numerical Reasoning Tasks](https://mlsanity.com/paper/2211.12588/) (2022)
- [Training Large Language Models to Reason in a Continuous Latent Space](https://mlsanity.com/paper/2412.06769/) (2024)
- [Automatic Chain of Thought Prompting in Large Language Models](https://mlsanity.com/paper/2210.03493/) (2022)
- [MathQA: Towards Interpretable Math Word Problem Solving with Operation-Based Formalisms](https://mlsanity.com/paper/1905.13319/) (2019)
- [Program Induction by Rationale Generation : Learning to Solve and Explain Algebraic Word Problems](https://mlsanity.com/paper/1705.04146/) (2017)
- [Complexity-Based Prompting for Multi-Step Reasoning](https://mlsanity.com/paper/2210.00720/) (2022)
- [Selection-Inference: Exploiting Large Language Models for Interpretable Logical Reasoning](https://mlsanity.com/paper/2205.09712/) (2022)
- [Reasoning with Latent Thoughts: On the Power of Looped Transformers](https://mlsanity.com/paper/2502.17416/) (2025)
- [MixLLM: Dynamic Routing in Mixed Large Language Models](https://mlsanity.com/paper/2502.18482/) (2025)
- [Language Models are Hidden Reasoners: Unlocking Latent Reasoning Capabilities via Self-Rewarding](https://mlsanity.com/paper/2411.04282/) (2024)
- [Think Before Recommend: Unleashing the Latent Reasoning Power for Sequential Recommendation](https://mlsanity.com/paper/2503.22675/) (2025)
- [Reasoning to Learn from Latent Thoughts](https://mlsanity.com/paper/2503.18866/) (2025)
- [LLM-Enhanced User-Item Interactions: Leveraging Edge Information for Optimized Recommendations](https://mlsanity.com/paper/2402.09617/) (2024)
- [Knockoff-Guided Feature Selection via A Single Pre-trained Reinforced Agent](https://mlsanity.com/paper/2403.04015/) (2024)
- [A Survey on Data-Centric AI: Tabular Learning from Reinforcement Learning and Generative AI Perspective](https://mlsanity.com/paper/2502.08828/) (2025)
- [Sculpting Features from Noise: Reward-Guided Hierarchical Diffusion for Task-Optimal Feature Transformation](https://mlsanity.com/paper/2505.15152/) (2025)
- [Topology-aware Reinforcement Feature Space Reconstruction for Graph Data](https://mlsanity.com/paper/2411.05742/) (2024)
- [Unsupervised Feature Transformation via In-context Generation, Generator-critic LLM Agents, and Duet-play Teaming](https://mlsanity.com/paper/2504.21304/) (2025)
- [Agentic Feature Augmentation: Unifying Selection and Generation with Teaming, Planning, and Memories](https://mlsanity.com/paper/2505.15076/) (2025)
- [Bridging the Domain Gap in Equation Distillation with Reinforcement Feedback](https://mlsanity.com/paper/2505.15572/) (2025)
- [Brownian Bridge Augmented Surrogate Simulation and Injection Planning for Geological CO2 Storage](https://mlsanity.com/paper/2505.18204/) (2025)
- [Towards Data-Centric AI: A Comprehensive Survey of Traditional, Reinforcement, and Generative Approaches for Tabular Data Transformation](https://mlsanity.com/paper/2501.10555/) (2025)
- [Building a Chinese Medical Dialogue System: Integrating Large-scale Corpora and Novel Models](https://mlsanity.com/paper/2410.03521/) (2024)

## Built on by (most cited first, 7 of 7)

- [SwiReasoning: Switch-Thinking in Latent and Explicit for Pareto-Superior Reasoning LLMs](https://mlsanity.com/paper/2510.05069/) (2025)
- [Teaching Time Series to See and Speak: Forecasting with Aligned Visual and Textual Perspectives](https://mlsanity.com/paper/2506.24124/) (2025)
- [StorageXTuner: An LLM Agent-Driven Automatic Tuning Framework for Heterogeneous Storage Systems](https://mlsanity.com/paper/2510.25017/) (2025)
- [Why Limit the Residual Stream to Layers and Not Tokens? Persistent Memory for Continuous Latent Reasoning](https://mlsanity.com/paper/2606.07720/) (2026)
- [Invariant Reasoning Directions in Latent Trajectories of Language Models](https://mlsanity.com/paper/2606.29164/) (2026)
- [Right Direction, Wrong Step: Geometric Analysis of Finite-Step Failure in Looped Transformers](https://mlsanity.com/paper/2609.16665/) (2026)
- [Multi-Agent Procedural Graph Extraction with Structural and Logical Refinement](https://mlsanity.com/paper/2601.19170/) (2026)

## Abstract (arXiv)

Reasoning is a key component of language understanding in Large Language Models. While Chain-of-Thought prompting enhances performance via explicit intermediate steps, it suffers from sufficient token overhead and a fixed reasoning trajectory, preventing step-wise refinement. Recent advances in latent reasoning address these limitations by refining internal reasoning processes directly in the model's latent space, without producing explicit outputs. However, a key challenge remains: how to effectively update reasoning embeddings during post-training to guide the model toward more accurate solutions. To overcome this challenge, we propose a lightweight post-training framework that refines latent reasoning trajectories using two novel strategies: 1) Contrastive reasoning feedback, which compares reasoning embeddings against strong and weak baselines to infer effective update directions via embedding enhancement; 2) Residual embedding refinement, which stabilizes updates by progressively integrating current and historical gradients, enabling fast yet controlled convergence. Extensive experiments and case studies are conducted on five reasoning benchmarks to demonstrate the effectiveness of the proposed framework. Notably, a 5% accuracy gain on MathQA without additional training.