Establishes non-asymptotic convergence bounds for three discrete diffusion models using Euler discretization, applicable to finite and infinite state spaces.

Giovanni Conforti, Alain Durmus, Le-Tuyet-Nhi Pham et al.

APRM improves mathematical reasoning accuracy by 3.4 percentage points through adversarial training between a generator and reward model.

Gurusha Juneja, Deepak Nathani, William Yang Wang

S2KAN integrates symbolic primitives with differentiable sparsity, achieving compact, interpretable models with high accuracy.

James Bagrow, Josh Bongard

EvilGenie benchmarks coding-agent reward hacking with holdouts, file monitoring, and LLM judges; GPT-5 had one false positive on unambiguous cases.

Jonathan Gabor, Jayson Lynch, Jonathan Rosenfeld

Combines linear prediction, nearest neighbor classification, and conformal calibration for real-time flight safety risk warning with statistical guarantees.

Aaron O. Feldman, D. Isaiah Harp, Joseph Duncan et al.

Nemotron-Flash optimizes depth-width ratios and operator choices to significantly enhance latency and throughput of small language models.

Yonggan Fu, Xin Dong, Shizhe Diao et al.

SAMBA model leverages spatial embedding and differential Mamba for long-context EEG modeling, significantly improving accuracy.

Jiazhen Hong, Geoffrey Mackellar, Soheila Ghane

ReBaPL integrates cyclical SGHMC and representation-space repulsion to enhance multi-modal prompt Bayesian inference, capturing multi-peak posteriors for better generalization.

Yassir Bendou, Omar Ezzahir, Eduardo Fernandes Montesuma et al.

GESC introduces a projection-based interference cancellation in complex space, achieving state-of-the-art accuracy of 84.9% on node classification benchmarks.

Yoonhyuk Choi, Jiho Choi, Jiwoo Kang

EntroPIC employs proportional-integral control to stabilize entropy, enhancing exploration in large language model training.

Kai Yang, Xin Xu, Yangkun Chen et al.

Introduced soft-label training, reducing KL divergence by 32% on datasets like ChaosNLI, enhancing model uncertainty expression.

Agamdeep Singh, Ashish Tiwari, Hosein Hasanbeig et al.

VISAGNN introduces a dynamic staleness-aware mechanism, significantly improving large-scale GNN training efficiency and accuracy.

Rui Xue

Energy sparse grids, Glynn symmetrization, and ReLU networks achieve O(m^-1) approximation with only polynomial dimension prefactors.

Yulong Lu, Tong Mao, Jinchao Xu et al.

HardFlow reformulates hard-constrained sampling as trajectory optimization using optimal control, ensuring terminal constraints efficiently.

Zeyang Li, Kaveh Alim, Navid Azizan

IGNS integrates Hamiltonian dynamics into graph neural simulators, greatly improving long-range interaction accuracy and stability.

Tai Hoang, Alessandro Trenta, Alessio Gravina et al.

Stuart-Landau Oscillatory Graph Neural Network (SLGNN) dynamically adjusts amplitudes to address oversmoothing in GNNs.

Kaicheng Zhang, David N. Reynolds, Piero Deidda et al.

Introduces Complexity as Advantage (CAA) framework, defining complexity via predictive regret.

Oshri Naparstek

Lookahead Unmasking (LookUM) improves diffusion language decoding by path search, reducing errors with only 2-3 candidate paths, outperforming greedy methods.

Sanghyun Lee, Seungryong Kim, Jongho Park et al.

Proposes a structured world model framework based on Hidden Markov Models (HMM) and switching linear dynamical systems (sLDS), supporting passive prediction and active control.

Lancelot Da Costa, Sanjeev Namjoshi, Mohammed Abbas Ansari et al.

DeepThinkVLA integrates hybrid attention and two-stage training, achieving 97.0% success by satisfying decoding and causal alignment conditions.

Cheng Yin, Yankai Lin, Wang Xu et al.