Enhances math problem-solving models with multi-turn preference learning.
problem Improving mathematical problem-solving capabilities of large language models.
method Introduces a multi-turn direct preference learning framework for tool-integrated mathematical reasoning tasks.
result Significant performance improvements in model accuracy on math datasets.
The paper certifies AI reliability via sampling and calibration, providing exact guarantees.
problem Ensuring trust in black-box AI systems' outputs.
method Self-consistency sampling and conformal calibration.
result Reliability levels derived from these methods offer finite-sample guarantees.
LaTRO optimizes latent reasoning in LLMs without external reward.
problem Training LLMs to perform complex reasoning tasks.
method Formulates reasoning as latent distribution sampling and optimizes via variational approaches.
result LLMs improve reasoning and evaluation quality through self-improvement.
A new method for math reasoning that allows for iterative correction.
problem Standard reasoning models commit to each token and cannot recover from early errors.
method Generative framework with latent thought vectors for iterative self-correction.
result 30 rethinking iterations surpass baselines with 15 times more parameters.
QAlign improves language model alignment with less compute, outperforming existing methods.
problem Improving language model performance with limited test-time computation.
method QAlign: sampling from optimal aligned distribution using Markov chain Monte Carlo.
result Consistent improvements over existing methods on various benchmarks.
DTM improves dLLM fine-tuning stability and performance.
problem Intractable sequence-level marginal likelihoods for masked diffusion models.
method Discrete Tilt Matching (DTM) recasts dLLM fine-tuning as state-level matching of local unmasking posteriors under reward tilting.
result DTM yields strong gains on Sudoku and Countdown while remaining competitive on MATH500 and GSM8K.
The study examines how verifier imperfections impact test-time scaling techniques.
problem Understanding how verifier imperfections affect test-time scaling methods.
method Proves the instance-level accuracy of Best-of-N and Rejection Sampling methods using the geometry of the verifier's ROC curve.
result RS outperforms BoN for fixed compute, but both converge to the same accuracy in the infinite-compute limit.
PIPA aligns preferences without reinforcement learning, improving language model performance.
problem Aligning preferences in language models efficiently and without reinforcement learning.
method Formulates preference alignment as a Maximum Likelihood Estimation problem with prior constraints.
result PIPA algorithms achieve up to 10% performance improvement on benchmarks.
GSI improves efficiency of large language model inference.
problem Efficiently guiding test-time alignment in large language models.
method Combines soft best-of-n scaling with a reward model and speculative samples. result Achieves higher accuracy and reduced latency compared to standard methods.
EORM boosts LLM accuracy with a lightweight, energy-based verifier.
problem Efficiently verifying mathematical reasoning in large language models.
method Energy-based framework to rank Chain-of-Thought solutions using simple outcome labels.
result EORM boosts LLM accuracy to 90.7% on GSM8k and 63.7% on MATH with only 55M parameters.
LLMs can fail to maximize aligned values even after training, due to irrational reasoning.
problem Value misalignment in LLMs' reasoning despite training alignment.
method Formalized rational value risk and decomposed estimation error.
result Rational value risk is widespread and cannot be fully eliminated.
wd1 improves reasoning in dLLMs by optimizing policies without policy ratios.
problem Improving reasoning in diffusion-based large language models through RL.
method wd1: ratio-free policy optimization using weighted log-likelihood.
result wd1 outperforms diffusion-based GRPO while requiring lower computational cost.
New method uses model comparison signals to improve LLM evaluation accuracy.
problem Limited benchmark sizes and model stochasticity in evaluating LLMs' mathematical reasoning.
method Combines standard labeled outcomes with model comparison signals to design a statistically efficient evaluation framework.
result Semiparametric estimator achieves the semiparametric efficiency bound and substantially improves ranking accuracy.
New method improves inference for discrete diffusion models, achieving better quality and efficiency.
problem High dimensionality of discrete diffusion models causes inference challenges.
method Developed high-order numerical inference schemes for discrete diffusion models.
result Second-order accuracy of the θ-Trapezoidal method in KL divergence. Enhances large language models' reasoning through simpler off-policy reinforcement learning.
problem Improving large language models' ability to reason and solve problems.
method EM Policy Gradient, optimizing expected return over reasoning trajectories using Expectation-Maximization (EM) optimization.
result Achieves comparable or slightly superior performance to state-of-the-art methods on reasoning datasets, with additional cognitive behaviors.
A new sparse benchmark metabench identifies key abilities from large benchmarks.
problem Redundancy and compression in existing benchmarks.
method Data from 5000+ LLMs to identify most informative items, distilling a sparse benchmark.
result Sparse benchmark metabench captures underlying abilities with high accuracy.
A conformal procedure improves CoT reasoning by aggregating reasoning paths and calibrating abstention rules.
problem Aggregation uncertainty in chain-of-thought reasoning makes correct answers less reliable.
method Introduces a conformal procedure for CoT reasoning that uses weighted score aggregation and abstention rules.
result Achieves higher selective accuracy with abstention, reducing confident-error rate.
ReD improves LLM inference efficiency at fixed budget, reducing attempts and cost.
problem Improving LLM inference efficiency at a fixed budget.
method Reset-and-Discard (ReD) query method.
result ReD increases coverage@cost for a given budget, reducing attempts and cost.
A new method for optimizing language-based agentic systems using semantic backpropagation.
problem Lack of proper feedback assignment in optimizing agentic systems.
method Formalization of semantic backpropagation with semantic gradients and semantic gradient descent.
result Our method outperforms existing state-of-the-art methods for solving GASO problems.
Our method upweights easy samples to mitigate forgetting in fine-tuning.
problem Catastrophic forgetting in fine-tuning pre-trained models.
method Sample weighting based on pre-trained model's losses.
result Our method reduces forgetting by up to 0.8% on MetaMathQA while preserving more accuracy on pre-training datasets.
Higher conservative training increases reward-hacking in reasoning models.
problem Reward hacking during online adaptation in reasoning models.
method Conservative offline training with varying levels of conservatism (β) was applied to a Qwen3-14B policy, and online adaptation was measured against a reward ensemble.
result Higher conservatism (β) increases reward-hacking damage, measured by the Goodhart gap and AUGC.
Top-H decoding improves text generation by balancing creativity and coherence.
problem Balancing creativity and coherence in text generation.
method Entropy-constrained minimum divergence problem, NP-hard ECMM problem, top-H decoding.
result Top-H decoding outperforms min-p sampling by up to 25.63% on creative writing benchmarks.