Proposes Actor-Critic-based improper RL algorithms for combining multiple controllers to optimize unknown MDPs, with proven convergence rates.

Mohammadi Zaki, Avinash Mohan, Aditya Gopalan et al.

This work reveals SGD leverages Fourier spectral gaps to gradually amplify sparse features, approaching the computational limit in learning k-sparse parity problems.

Boaz Barak, Benjamin L. Edelman, Surbhi Goel et al.

Proposes a unified 2D-3D molecular pretraining model, boosting property prediction accuracy by 8.3%.

Jinhua Zhu, Yingce Xia, Lijun Wu et al.

Proposes CausalAgents benchmark, using removal of non-causal agents to evaluate motion prediction robustness, with 25-38% minADE change observed.

Rebecca Roelofs, Liting Sun, Ben Caine et al.

This paper introduces TokenGT, a pure Transformer for graphs using node and edge tokens, theoretically matching or surpassing 2-IGN in expressive power.

Jinwoo Kim, Tien Dat Nguyen, Seonwoo Min et al.

Introduces Autocast dataset with news retrieval, using large models (T5, GPT-2) for future event forecasting; accuracy 65% vs. 92% human baseline.

Andy Zou, Tristan Xiao, Ryan Jia et al.

This paper proves that neural networks trained via gradient descent can efficiently learn high-degree polynomials depending on few features, with sample complexity much lower than kernel methods.

Alex Damian, Jason D. Lee, Mahdi Soltanolkotabi

Proposes data pruning to surpass power-law neural scaling, achieving exponential error reduction; validated on ResNet with CIFAR-10, SVHN, ImageNet.

Ben Sorscher, Robert Geirhos, Shashank Shekhar et al.

Proposed Structural Entropy Guided Pooling (SEP), avoiding local structure damage and improving graph and node classification accuracy.

Junran Wu, Xueyuan Chen, Ke Xu et al.

VPT uses a small labeled dataset to train an inverse dynamics model, then leverages large-scale unlabeled videos for pretraining, achieving near-human performance in Minecraft.

Bowen Baker, Ilge Akkaya, Peter Zhokhov et al.

Proposes Behavior Transformer (BeT), combining action discretization and multi-task residual correction, to model multi-modal behaviors with high accuracy.

Nur Muhammad Mahi Shafiullah, Zichen Jeff Cui, Ariuntuya Altanzaya et al.

This study uses a synthetic dataset to compare FID, IS, and classical f-divergences, revealing that divergence-based metrics are more stable for model evaluation.

Eyal Betzalel, Coby Penso, Aviv Navon et al.

Proposes Discrete Langevin Proposal (DLP), a gradient-based high-dimensional discrete sampler with zero asymptotic bias, outperforming Gibbs in efficiency.

Ruqi Zhang, Xingchao Liu, Qiang Liu

Proposes Nested Exponential Weights (NEW) algorithm leveraging hierarchical similarity structures, achieving regret bounds of O(√neff log n · T).

Matthieu Martin, Panayotis Mertikopoulos, Thibaud Rahier et al.

Proposes a novel algorithm that adaptively learns stochastic and adversarial bandits with general graph feedback, achieving poly log T regret in stochastic and O(T^{2/3}) in adversarial environments.

Fang Kong, Yichi Zhou, Shuai Li

Proposed RFedSVRG algorithm achieves O(1/ε²) convergence for federated optimization on Riemannian manifolds, handling non-convex constraints effectively.

Jiaxiang Li, Shiqian Ma

Meta OT leverages meta-learning to predict optimal transport maps, significantly accelerating multiple similar OT problem solutions with minimal computation.

Brandon Amos, Samuel Cohen, Giulia Luise et al.

ROI-Constrained Bidding via Curriculum-Guided Bayesian Reinforcement Learning enhances learning efficiency and stability.

Haozhe Wang, Chao Du, Panyan Fang et al.

Concept Relevance Propagation (CRP) integrates local and global XAI, providing human-understandable explanations with concept-level insights.

Reduan Achtibat, Maximilian Dreyer, Ilona Eisenbraun et al.

Introduces neuron partition analysis, proving early-stage rapid convergence and global convergence of mildly parameterized neural networks, surpassing NTK over-parameterization limits.

Mingze Wang, Chao Ma