Statistics Theory

Mohsen Hariri, Michael Hinczewski, Jing Ma et al.

This work addresses the underexplored issue of ranking models in test-time scaling for reasoning LLMs, providing practical tools for researchers and practitioners, though it is incremental in improving existing ranking methodologies.

Bruno Trentini, Dejan Stancevic, Michael M. Bronstein et al.

For practitioners of flow-based generative models with limited inference compute, this work provides a principled, training-free scheduling method that consistently outperforms heuristic grids.

Jingda Wu, Changxiao Cai

Provides the first rigorous theoretical justification for diffusion models' ability to adapt to low-dimensional structure and multi-modality, addressing a key gap in understanding their empirical success.

Wenxuan Zou, Galen Reeves

This work provides a theoretical understanding of free energy universality for tensor estimation problems, extending prior results from matrix settings to a broader class of models and scaling regimes.

Tor Lattimore

This provides theoretical insights into the performance of policy gradient in bandit settings, which is incremental as it builds on existing methods with new analysis.

Charlie Cowen-Breen, Alekh Agarwal, Stephen Bates et al.

Provides a general framework for resource-constrained statistical estimation, improving efficiency for practitioners using multiple proxies.

Steve Hanneke, Anay Mehrotra, Grigoris Velegkas et al.

Provides a theoretical characterization and first algorithm for a classic but understudied learning model, clarifying the role of membership queries.

Shi Chen, Zhengjiang Lin, Kaizhao Liu et al.

Provides rigorous statistical guarantees for transformer performance as context length grows, addressing a key theoretical gap for practitioners scaling models.

Krishnakumar Balasubramanian, Shiva Prasad Kasiviswanathan

This provides a principled understanding of forgetting for researchers and practitioners in machine learning, though it is incremental as it builds on existing theoretical abstractions.

Yujia Zheng, Zijian Li, Shunxing Fan et al.

For practitioners in unsupervised learning, this provides a principled way to recover partial latent structure without unverifiable assumptions, though the results are theoretical and domain-agnostic.

Danru Xu, Sébastien Lachapelle, Sara Magliacane

For researchers in causal representation learning, this work extends identifiability guarantees to degenerate Gaussian mixtures, addressing a challenging setting where standard density-based methods fail.

Giuseppe Bruno, Shi Chen, Zhengjiang Lin et al.

For theorists studying transformer scaling, this provides precise phase transition boundaries and limiting laws, but the analysis is restricted to i.i.d. keys and fixed queries, limiting direct applicability.

Begoña García Malaxechebarría, Courtney Paquette, Maryam Fazel et al.

Provides a rigorous theoretical framework for understanding SGD dynamics in a simplified neural network setting, offering explicit non-asymptotic convergence guarantees.

Matthew S. Zhang, Jason M. Altschuler, Sinho Chewi

This resolves the computational bottleneck of finding warm starts for HMC, which is crucial for practitioners in statistics, engineering, and sciences who rely on HMC for high-dimensional sampling, though it is incremental as it builds on prior theoretical work.

Libin Zhu, Damek Davis, Dmitriy Drusvyatskiy et al.

This provides a theoretical explanation for the sample efficiency of iterative kernel methods like Recursive Feature Machines in learning low-dimensional structure from high-dimensional data.

Daniil Dmitriev, Zhihan Huang, Yuting Wei

This work provides a theoretical advancement in the efficiency of discrete diffusion models, which is important for researchers and practitioners working with high-dimensional data generation.

Zhiyang Xun, Eric Price

Provides fundamental limits for diffusion sampling acceleration, relevant to researchers designing faster sampling algorithms.

Felix Zhou, Anay Mehrotra, Quanquan C. Liu

For practitioners seeking to improve reasoning in language models without additional training, this work offers a practical sampling method that outperforms prior approaches and RL-trained models.

Kelvin Kan, Xingjian Li, Benjamin J. Zhang et al.

This work provides the first convergence theory for discrete diffusion models that scales to large vocabularies (e.g., hundreds of thousands of tokens) by removing the state-space-size dependence that made prior bounds vacuous for modern language tasks.

Florian Kalinke, Zoltán Szabó, Bharath K. Sriperumbudur

For practitioners needing scalable goodness-of-fit tests on large datasets, this work provides a theoretically grounded acceleration of KSD-based testing without sacrificing statistical performance.