Introduces Capped Implicit Exploration (CIX) to balance bias-variance tradeoff in NeuRD, improving stability in non-stationary environments.

Dustin Morrill, Esra'a Saleh, Michael Bowling et al.

DPM-Solver uses analytical diffusion ODE solutions, enabling high-quality sampling in 10-20 steps, with 4-16× speedup over prior methods.

Cheng Lu, Yuhao Zhou, Fan Bao et al.

DepthShrinker merges linear layers by removing unimportant activation functions, boosting hardware utilization and achieving 3.06% accuracy gain and 1.53× throughput over SOTA pruning.

Yonggan Fu, Haichuan Yang, Jiayi Yuan et al.

CoNSoLe uses double-convex deep Q-learning and Locally Convex Equation Learner for symbolic regression, improving accuracy.

Haoran Li, Yang Weng, Hanghang Tong

SymFormer uses a Transformer-based architecture for symbolic regression, achieving fast and accurate inference.

Martin Vastl, Jonáš Kulhánek, Jiří Kubalík et al.

Post-hoc concept bottleneck models (PCBM) enable interpretable, high-performance neural networks without dense concept annotations, utilizing multimodal and residual techniques.

Mert Yuksekgonul, Maggie Wang, James Zou

Balanced EFCE-OMD makes Φ-Hedge polynomial-time in EFGs and achieves near-optimal \tilde{O}(√XAT) bandit trigger regret.

Yu Bai, Chi Jin, Song Mei et al.

Large-step GD can escape local minima and outperform small-step GD.

Amirkeivan Mohtashami, Martin Jaggi, Sebastian Stich

Proposes Hilbert Curve Projection (HCP) distance for efficient high-dimensional distribution comparison, with theoretical guarantees and superior empirical performance.

Tao Li, Cheng Meng, Hongteng Xu et al.

FlashAttention: IO-aware exact attention reduces HBM accesses, boosting training speed by up to 15% on BERT-large.

Tri Dao, Daniel Y. Fu, Stefano Ermon et al.

Mathematical analysis reveals traditional RPE in Transformers limits universal approximation; proposes URPE to ensure model expressiveness.

Shengjie Luo, Shanda Li, Shuxin Zheng et al.

OptFormer, a Transformer-based universal hyperparameter optimizer, can imitate 7 algorithms, leveraging large-scale real tuning data to improve prediction and optimization.

Yutian Chen, Xingyou Song, Chansoo Lee et al.

Proposes Contextual Pandora’s Box with reservation value linking context and distribution, achieving sublinear regret.

Alexia Atsidakou, Constantine Caramanis, Evangelia Gergatsouli et al.

Using large language models for autoformalization, achieving 25.3% perfect translation of math problems into Isabelle/HOL.

Yuhuai Wu, Albert Q. Jiang, Wenda Li et al.

This study shows pre-trained models inherently regularize, making KD and label smoothing less effective during fine-tuning; performance differences are negligible.

Ivan Kobyzev, Aref Jafari, Mehdi Rezagholizadeh et al.

Multi-Head Online Learning model significantly improves delayed feedback CVR and VPC predictions.

Hui Gao, Yihan Yang

Proposes an optimal comparator adaptive online learning algorithm with switching costs, achieving the best regret bounds via continuous-time dual space scaling.

Zhiyu Zhang, Ashok Cutkosky, Ioannis Ch. Paschalidis

This survey reviews exploration techniques in deep reinforcement learning, including reward for novel states, diverse behaviors, goal-based, probabilistic, imitation, safe, and random methods.

Pawel Ladosz, Lilian Weng, Minwoo Kim et al.

DEIS employs exponential integrators for diffusion models, enabling high-fidelity sampling in as few as 10 steps, with FID 4.17 on CIFAR10 at 10 NFEs.

Qinsheng Zhang, Yongxin Chen

AGIC is an efficient approximate gradient inversion attack leveraging multi-epoch updates, achieving up to 50% PSNR improvement and 5x speed over SOTA.

Jin Xu, Chi Hong, Jiyue Huang et al.