YAQA, a Hessian-structured adaptive rounding algorithm, reduces quantization error by ~30%, with theoretical end-to-end error bounds.

Albert Tseng, Zhaofeng Sun, Christopher De Sa

Proposes GenICL, using LLM feedback for direct demonstration selection optimization, significantly improving ICL performance.

Zheng Zhang, Shaocheng Lan, Lei Song et al.

Proposes density-ratio-free doubly robust kernel estimators for proxy causal learning, effective in high-dimensional continuous treatments.

Bariscan Bozkurt, Houssam Zenati, Dimitri Meunier et al.

Self-distillation framework for flow map learning accelerates consistency models, achieving 10-100× inference speedup with high stability.

Nicholas M. Boffi, Michael S. Albergo, Eric Vanden-Eijnden

CapBencher reduces Bayes accuracy by randomizing answers to detect LLM test-set overfitting.

Takashi Ishida, Thanawat Lodkaew, Ikko Yamane

Introduces Semi-Simplicial Neural Networks (SSNs) for directed higher-order relation modeling, achieving state-of-the-art brain activity decoding accuracy.

Manuel Lecha, Andrea Cavallo, Francesca Dominici et al.

Trinity-RFT offers a unified, scalable framework for reinforcement fine-tuning of large language models, supporting multi-mode, multi-device training with high efficiency.

Xuchen Pan, Yanxi Chen, Yushuo Chen et al.

Proposes T-MEX score within measurement model framework to evaluate causal representations in CRL.

Dingling Yao, Shimeng Huang, Riccardo Cadei et al.

RIPT-VLA fine-tunes pretrained VLA models via reinforcement learning, boosting success rate to 97.5% with only one demonstration.

Shuhan Tan, Kairan Dou, Yue Zhao et al.

Capped Squared Loss learns contextual value distributions with O(dξ²cmax⁴/ε⁸δ²) samples.

Anna Heuser, Thomas Kesselheim

Think-RM models internal reasoning to enable long-horizon inference, outperforming BT RM and scaled GenRM by 8% on RM-Bench.

Ilgee Hong, Changlong Yu, Liang Qiu et al.

Efficiently scaling diffusion Transformers via $μ$P, PixArt-$α$ surpasses baseline at 0.61B parameters.

Chenyu Zheng, Xinyu Zhang, Rongzhen Wang et al.

ADASAP combines randomized Nyström approximation and sketch-and-project for scalable large-scale GP inference, outperforming state-of-the-art methods.

Pratik Rathore, Zachary Frangella, Sachin Garg et al.

MeanFlow introduces average velocity for one-step generation, achieving FID 3.43, outperforming prior methods.

Zhengyang Geng, Mingyang Deng, Xingjian Bai et al.

Fractured Sampling method reduces token usage while improving Pass@k accuracy.

Baohao Liao, Hanze Dong, Yuhui Xu et al.

Semi-supervised model alignment using conditional flow matching with inter-modal bridge cost achieves effective cross-modal distribution mapping with limited paired data.

Ali Gholamzadeh, Noor Sajid

Proposes dLLM-Cache, a training-free adaptive caching method combining long-interval prompt caching and partial response updates, reducing FLOPs by up to 9.1× and inference latency.

Zhiyuan Liu, Yicun Yang, Yaojie Zhang et al.

211 student-made applied-math problems benchmark LLMs on asymptotics and approximation.

James V. Roggeveen, Erik Y. Wang, Will Flintoft et al.

Ready2Unlearn optimizes models during training for future unlearning readiness.

Hanyu Duan, Yi Yang, Ahmed Abbasi et al.

L2T framework optimizes LLM reasoning efficiency and effectiveness using information-theoretic reinforcement learning, reducing unnecessary token usage.

Jingyao Wang, Wenwen Qiang, Zeen Song et al.