English(EN)Aligning Few-Step Generative Models by Amortizing Sample-based Variational Inference
新研究推动生成模型在效率和评估方面取得进展
作者PulseAugur 编辑部·[13 个来源]·
几篇最新的研究论文探讨了生成模型的进展，重点是提高其效率、可评估性和对齐性。其中一篇论文提出了一种使用基于分数的生成模型进行加权采样的新框架，实现了显著的加速。另一个理论框架解决了生成模型的统计可评估性问题，区分了可以从有限样本中可靠估计的指标和不能的指标。其他研究介绍了参数高效的生成建模方法、将模型校准到分布约束以及使用基于样本的变分推理来对齐少样本生成模型的方法。
AI
arXiv:2502.04646v2 Announce Type: replace-cross Abstract: Weighted sampling -- sampling from a probability density function (PDF) proportional to the product of a base PDF and a weight function -- is a fundamental technique with wide-ranging applications in variance reduction, bi…
arXiv cs.LG
TIER_1English(EN)·Shashaank Aiyer, Yishay Mansour, Shay Moran, Han Shao·
arXiv:2604.05324v2 Announce Type: replace Abstract: Statistical evaluation aims to estimate the generalization performance of a model using held-out i.i.d. test data sampled from the ground-truth distribution. In supervised learning settings such as classification, performance me…
arXiv:2503.07154v3 Announce Type: replace-cross Abstract: Generative pre-training is often framed through a false dichotomy between autoregressive models for discrete signals and diffusion models for continuous signals. We argue that the dichotomy is false because it conflates mo…
arXiv cs.LG
TIER_1English(EN)·Sascha Diefenbacher, Sofia Palacios Schweitzer, Gregor Kasieczka·
arXiv:2605.30453v1 Announce Type: cross Abstract: Generative machine learning has become an essential tool in theoretical and experimental physics, especially in the context of fast surrogates and density estimators. In this work, we first introduce the underlying framework of mo…
arXiv cs.LG
TIER_1English(EN)·Daniil Shlenskii, Nikita Gushchin, Lev Novitskiy, Dmitry V. Dylov, Alexander Korotin·
arXiv:2605.29920v1 Announce Type: new Abstract: We introduce Midpoint Generative Models (MGM), a principled framework for training one-step generative models. MGM is based on a simple symmetry of Flow Matching with linear interpolation: when the two endpoint distributions coincid…
We introduce a continuous-time generative modeling framework, motivated by the Chow-Rashevskii theorem, that builds expressive flows from a small set of fixed vector fields and learned scalar controls. Instead of learning an unconstrained high-dimensional vector field, our framew…
arXiv cs.AI
TIER_1English(EN)·Jaewoo Lee, Hyeongyu Kang, Dohyun Kim, Kyuil Sim, Woocheol Shin, Minsu Kim, Taeyoung Yun, Jeongjae Lee, Sanghyeok Choi, Tabitha Edith Lee, Jongchul Ye, Jinkyoo Park·
arXiv:2605.26552v1 Announce Type: cross Abstract: Aligning a few-step generative model is challenging, since existing alignment frameworks typically rely on restrictive assumptions: a tractable likelihood, a specific ODE/SDE solver, or a particular model family. We introduce FAV,…
arXiv stat.ML
TIER_1English(EN)·Meshi Bashari, Yonghoon Lee, Roy Maor Lotan, Edgar Dobriban, Yaniv Romano·
arXiv:2509.20345v3 Announce Type: replace-cross Abstract: The rapid proliferation of high-quality synthetic data -- generated by advanced AI models or collected as auxiliary data from related tasks -- presents both opportunities and challenges for statistical inference. This pape…
arXiv stat.ML
TIER_1English(EN)·Stanislas Strasman, Gabriel Cardoso, Sylvain Le Corff, Vincent Lemaire, Antonio Ocello·
arXiv:2601.21868v2 Announce Type: replace Abstract: Understanding the stability and long-time behavior of generative models is a fundamental problem in modern machine learning. This paper provides quantitative bounds on the sampling error of score-based generative models by lever…
arXiv stat.ML
TIER_1English(EN)·Cheng Lu, Mengxin Wang, Dennis J. Zhang, Heng Zhang·
arXiv:2604.14575v2 Announce Type: replace-cross Abstract: Large language models enable inexpensive AI-generated annotations, but using them reliably for causal inference remains challenging. Naively pooling AI and human data induces bias, while existing methods such as Prediction…
arXiv:2606.00425v1 Announce Type: cross Abstract: Moment conditions are widely used to identify parameters in models where the full likelihood is either unknown or intentionally left unspecified. Empirical likelihood methods address this problem by assigning probability weights t…
Moment conditions are widely used to identify parameters in models where the full likelihood is either unknown or intentionally left unspecified. Empirical likelihood methods address this problem by assigning probability weights to the observed data so that the sample moment cond…
arXiv stat.ML
TIER_1English(EN)·Henry D. Smith, Nathaniel L. Diamant, Brian L. Trippe·
arXiv:2510.10020v4 Announce Type: replace Abstract: Generative models frequently suffer miscalibration, wherein statistics of the sampling distribution, such as the fraction of generations in a given class, deviate from desired values. We frame calibration as a constrained optimi…
