arXiv:2606. 16222v1 Announce Type: new Abstract: Large Language Models (LLMs) increasingly rely on intermediate reasoning, yet explicit Chain-of-Thought (CoT) suffers from a linguistic space bottleneck: each thought must be decoded into tokens, causing high inference overhead.
The paper introduces Scalable Power Sampling, a training‑free method that sharpens a large language model’s generative distribution by approximating the autoregressive power distribution with a token‑level scaled low‑temperature distribution. This approach eliminates the need for iterative Markov chain Monte Carlo, achieving performance comparable to or better than one‑shot GRPO on math, QA, and code tasks while cutting inference latency by more than tenfold. The method relies solely on the base model’s outputs, requiring no external rewards or verifiers.
The paper introduces Decision-Flow Sampling (DF‑Sample), a training‑free, data‑free inference framework that builds a hierarchical reasoning tree, evaluates entire trajectories, and back‑propagates utilities to guide branching decisions. Unlike local step‑wise sampling, DF‑Sample explicitly assesses global paths, enabling it to recover high‑quality, low‑probability reasoning chains that standard decoding misses. On the GPQA benchmark, DF‑Sample attains 45.6% accuracy, outperforming power sampling (38.9%) and GRPO (39.9%) and consistently surpassing baselines across multiple models and benchmarks, demonstrating significant latent reasoning potential in pretrained LLMs.
Generative Recursive Reasoning (GRAM) extends Recursive Reasoning Models by introducing probabilistic multi‑trajectory computation, allowing neural reasoning systems to explore multiple latent trajectories instead of a single deterministic path. GRAM treats reasoning as a stochastic latent trajectory, enabling multiple hypotheses, alternative solution strategies, and scalable inference through recursive depth and parallel sampling. Trained via amortized variational inference, GRAM outperforms deterministic recurrent and recursive baselines on structured reasoning and multi‑solution constraint satisfaction tasks, and also supports unconditional generation.
arXiv:2310.02423v3 Announce Type: replace Abstract: We present a new algorithm for amortized inference in sparse probabilistic graphical models (PGMs), which we call $\Delta$-amortized inference ($\D...
arXiv:2609.39525v1 Announce Type: new Abstract: Casting Bayesian inference as a neural network optimization problem targeting an amortized posterior is attractive, as it extends to otherwise intracta...
arXiv:2606. 19264v1 Announce Type: new Abstract: The knowledge encoded in large language models (LLMs) can serve as a substrate for structured reasoning over variables describing a complex world, but accessing this knowledge in a probabilistically coherent manner poses a difficult inference problem.
The paper introduces LB-MCTS, a trajectory-structured optimization framework that combines Monte Carlo Tree Search, Bayesian Optimization, and Large Language Models to address the CASH problem in AutoML. It uses a shared tree state for algorithm selection, hyperparameter refinement, and synergy between BO and LLM proposals, with a reliability-aware policy that shifts from LLM-driven to BO-driven suggestions as the surrogate model improves. Experiments on 104 AMLB datasets show that LB-MCTS consistently outperforms existing BO-based, LLM-based, and hybrid baselines.
arXiv:2602. 14169v2 Announce Type: replace-cross Abstract: Effective exploration is a key challenge in reinforcement learning for large language models: discovering high-quality trajectories within a limited sampling budget from the vast natural language sequence space.
arXiv:2604. 17892v4 Announce Type: replace-cross Abstract: Recently, latent reasoning has been introduced into large language models (LLMs) to leverage rich information within a continuous space.
arXiv:2606. 19750v1 Announce Type: cross Abstract: Reinforcement learning (RL) is a central approach for improving reasoning capabilities in large language models (LLMs), where training efficiency depends critically on how problems are sampled during optimization.
The paper introduces POISE, a reinforcement learning algorithm that uses a model’s internal states as a value estimator to reduce variance in reinforcement learning with verifiable rewards (RLVR). By employing a lightweight probe that reads internal signals during the forward pass, POISE predicts baselines online and uses a cross‑rollout construction to keep gradients unbiased. Experiments on Qwen3‑4B and OLMo3‑7B‑Instruct‑DPO across six domains show POISE outperforms existing RLVR baselines, offering more stable training and a value model that generalizes across tasks and scales with the policy.