OmniISR unifies centralized and federated learning via intermediate supervision, reducing performance gap by 22.60%.

Wei-Bin Kou, Guangxu Zhu, Ming Tang et al.

HINT-SD employs full-trajectory hindsight analysis for targeted self-distillation, boosting long-horizon RL agent performance by up to 13.60 percentage points with 2.26× training efficiency.

Woongyeong Yeo, Yumin Choi, Taekyung Ki et al.

WinQ accelerates low-bit quantization-aware training by 4× using periodic weight interpolation and noise regularization, boosting efficiency and performance.

Dongyue Li, Zechun Liu, Kai Yi et al.

Developed geometric theory of projection heads in self-supervised learning, revealing their role and limitations.

Faris Chaudhry

SE-GA integrates hierarchical memory and self-evolution to boost GUI task success to 89.0%, enabling long-term planning and continuous learning.

Shilong Jin, Lanjun Wang, Zhuosheng Zhang

This study compares MRL-trained and non-MRL models under various truncation levels, showing non-MRL models outperform MRL ones below 80% truncation.

Sotaro Takeshita, Yurina Takeshita, Simone Paolo Ponzetto et al.

MetaEns uses unsupervised meta-learning to select outlier detection ensembles, improving precision and reducing model count across 39 datasets.

Hong-Phuc Phan, Tuan-Anh Vu, Tung Kieu et al.

MO-CAPO introduces a multi-objective, cost-aware prompt optimization algorithm, improving Pareto front diversity and efficiency.

Jan Büssing, Moritz Schlager, Timo Heiß et al.

Derives transformer-like inference architecture via optimal control, covering nonlinear discrete and linear Gaussian models.

Aditya Kudre, Heng-Sheng Chang, Prashant G. Mehta

Proposes Ghosted Layers, a training-free method deriving a closed-form linear operator to recover pruned layer activations, improving accuracy and perplexity.

Vincent-Daniel Yun, Junhyuk Jo, Sai Praneeth Karimireddy et al.

EvoLib constructs an evolving knowledge library via self-supervised abstraction extraction, boosting performance in mathematical reasoning and code generation without parameter updates.

Weijia Xu, Alessandro Sordoni, Chandan Singh et al.

Proposes a neighborhood-based Neural ODE training framework combining Jacobian accuracy and statistical properties, improving chaotic system modeling.

Joon-Hyuk Ko, Andrus Giraldo, Deok-Sun Lee

Introduces Riemannian Hodge Message Passing (RHMP), achieving top performance across seven physical benchmarks.

Dongzhe Zheng, Christine Allen-Blanchette

Full Jacobian eigendecomposition reveals spectral evolution from non-normal to symmetric layers, with training inducing low-rank bottlenecks and topology-spectral coupling.

Jesseba Fernando, Grigori Guitchounts

Proposes accuracy drop as a layer relevance metric, outperforming cosine similarity in large language models.

Cristian Hinostroza, Rodrigo Toro Icarte, Christ Devia et al.

Introduces scale-sensitive shattering, proving equivalence between uniform convergence, learnability, and fat-shattering dimension at optimal scales.

Shashaank Aiyer, Yishay Mansour, Shay Moran et al.

Extends neural posterior estimation (NPE) to support mixed discrete and continuous parameters using a factorized inference network with autoregressive classification and generative models.

Jan Boelts, Cornelius Schröder, Jonas Beck et al.

N-vium achieves 57.9% inference speedup with mixture-of-exits transformer, no perplexity loss.

Aleksander Lorenc, Frédéric Berdoz, Joël Mathys et al.

Exact Linear Attention (ELA) achieves O(L) complexity via kernel decomposition, eliminating approximation errors, and enhances long-sequence and vision tasks.

Weinuo Ou

Proposes Multi-Rollout On-Policy Distillation (MOPD), leveraging peer successes and failures to enhance model reasoning by 20% on math tasks.

Weichen Yu, Xiaomin Li, Yizhou Zhao et al.