Proposes end-to-end high-res diffusion with noise schedule adjustment, partial architecture scaling, and downsampling, achieving SOTA on ImageNet.

Emiel Hoogeboom, Jonathan Heek, Tim Salimans

This paper introduces diffusion models for imitation learning, outperforming traditional methods in modeling complex, multimodal human behaviors in sequential tasks.

Tim Pearce, Tabish Rashid, Anssi Kanervisto et al.

LUMEN combines pre-computed and on-the-fly encoding, improving QA performance with reduced compute, especially at larger scales.

Michiel de Jong, Yury Zemlyanskiy, Nicholas FitzGerald et al.

Solving inverse physics problems using score matching, achieving high accuracy and temporal stability.

Benjamin J. Holzschuh, Simona Vegetti, Nils Thuerey

HexPlane employs six orthogonal spatial-temporal feature planes for explicit dynamic scene representation, achieving over 100× faster training without quality loss.

Ang Cao, Justin Johnson

Proposes refined Nyström approximation combined with statistical learning theory, providing error bounds for non-i.i.d. samples, improving kernel quadrature accuracy.

Satoshi Hayakawa, Harald Oberhauser, Terry Lyons

Proposes GD-WL framework incorporating distance info, achieving provable expressiveness for graph biconnectivity tasks.

Bohang Zhang, Shengjie Luo, Liwei Wang et al.

D-Adaptation achieves parameter-free optimal convergence via adaptive D-estimation, matching hand-tuned rates without line searches.

Aaron Defazio, Konstantin Mishchenko

Grounds language and cognition in neural mechanisms; evaluates LLMs' formal vs. functional competence, highlighting strengths and gaps.

Kyle Mahowald, Anna A. Ivanova, Idan A. Blank et al.

Symbolic expression generation via VAE, SEGVAE achieves 65% recovery rate on Nguyen dataset.

Sergei Popov, Mikhail Lazarev, Vladislav Belavin et al.

This survey presents a comprehensive taxonomy of graph-level learning methods, including traditional kernels, substructure mining, graph embeddings, GNNs, and pooling, with performance insights.

Zhenyu Yang, Ge Zhang, Jia Wu et al.

Mechanistic interpretability reveals continuous progress measures for grokking via Fourier transforms, showing gradual amplification of learned algorithms.

Neel Nanda, Lawrence Chan, Tom Lieberum et al.

SlideVQA introduces a multi-image document VQA dataset and a Seq2Seq model with multi-task learning, boosting numerical reasoning by 12%.

Ryota Tanaka, Kyosuke Nishida, Kosuke Nishida et al.

Physics-informed kernel embeddings incorporate prior system knowledge, improving sample efficiency and prediction accuracy in control tasks.

Adam J. Thorpe, Cyrus Neary, Franck Djeumou et al.

Improved outlier-aware segmentation using mask-level recognition, significantly reducing false positives at semantic borders.

Matej Grcić, Josip Šarić, Siniša Šegvić

Laser-based loop closure integrated into commercial DVL-INS improves underwater map consistency by 30%, using factor graph optimization without raw sensor data.

Thomas Hitchcox, James Richard Forbes

Anonymous-FP combines anonymous random walks with PV-DBOW, reaching 95.74% accuracy on NCI-1.

Xue Liu, Qian Cheng, Dan Sun et al.

The e-Posterior uses e-variables to provide frequentist risk bounds, outperforming Bayesian posteriors in robustness under model misspecification.

Peter Grünwald

Proposed the joint Caption Grounding and Generation (CGG) framework, achieving 6.8% mAP boost on novel classes in open vocabulary instance segmentation.

Jianzong Wu, Xiangtai Li, Henghui Ding et al.

Muse: a masked Transformer for text-to-image generation with FID 6.06, enabling fast, high-fidelity synthesis.

Huiwen Chang, Han Zhang, Jarred Barber et al.