Introduces QuALITY, a long-text (≈5000 words) multiple-choice QA dataset, with models achieving only 55.4% accuracy versus 93.5% by humans.

Richard Yuanzhe Pang, Alicia Parrish, Nitish Joshi et al.

Proposes an efficient geometry-aware 3D GAN with hybrid explicit-implicit architecture, enabling real-time high-res multi-view consistent image synthesis.

Eric R. Chan, Connor Z. Lin, Matthew A. Chan et al.

MS-ResNet achieves 482 layers on CIFAR-10 and 76.02% accuracy on ImageNet.

Yifan Hu, Lei Deng, Yujie Wu et al.

Scaling T5-based dual encoders with fixed embedding size significantly improves out-of-domain retrieval, outperforming SOTA on BEIR dataset.

Jianmo Ni, Chen Qu, Jing Lu et al.

Addressing the generative learning trilemma with Denoising Diffusion GANs, achieving 2000x faster sampling on CIFAR-10.

Zhisheng Xiao, Karsten Kreis, Arash Vahdat

LFGC combines a leader-follower game with MPC and achieves 97.5% merge success on NGSIM.

Kaiwen Liu, Nan Li, H. Eric Tseng et al.

Learning phase field mean curvature flows using neural networks, extending to non-orientable interfaces.

Elie Bretin, Roland Denis, Simon Masnou et al.

Introduced a score-based generative model using Critically-Damped Langevin Diffusion, improving image synthesis quality.

Tim Dockhorn, Arash Vahdat, Karsten Kreis

Proposes a PIT entropy-based recalibration method for epidemic forecasts, significantly improving calibration and accuracy across 27 influenza models.

Aaron Rumack, Ryan J. Tibshirani, Roni Rosenfeld

Uses Wasserstein distance to analyze human exploration-exploitation strategies in black-box optimization.

Antonio Candelieri, Andrea Ponti, Francesco Archetti

Orthogonal polynomial-based DPP reduces gradient variance in SGD, achieving faster convergence.

Remi Bardenet, Subhro Ghosh, Meixia Lin

AvatarMe++ uses differentiable rendering and a GAN to infer render-ready facial geometry and BRDF from one in-the-wild image.

Alexandros Lattas, Stylianos Moschoglou, Stylianos Ploumpis et al.

Using DINO-ViT deep features as dense descriptors enables unsupervised semantic segmentation and correspondence with competitive performance.

Shir Amir, Yossi Gandelsman, Shai Bagon et al.

VUT model achieves five UI tasks with multimodal input, improving accuracy and reducing model count.

Yang Li, Gang Li, Xin Zhou et al.

Proposes O(1) memory attention algorithm extended to O(log n), enabling scalable long-sequence modeling.

Markus N. Rabe, Charles Staats

Plenoxels uses sparse voxels and spherical harmonics to replace NeRF, training about 100× faster with no visible quality loss.

Alex Yu, Sara Fridovich-Keil, Matthew Tancik et al.

Gopher (280B parameters) uses Transformer architecture, trained on MassiveText, outperforming SOTA on 152 tasks, especially in knowledge and comprehension.

Jack W. Rae, Sebastian Borgeaud, Trevor Cai et al.

RETRO integrates large-scale retrieval with language models, achieving GPT-3 level performance with 25x fewer parameters by retrieving from a 2 trillion token database.

Sebastian Borgeaud, Arthur Mensch, Jordan Hoffmann et al.

Enhancing image restoration with Test-time Local Converter (TLC), improving single-image deblurring PSNR from 32.92 dB to 33.57 dB on GoPro dataset.

Xiaojie Chu, Liangyu Chen, Chengpeng Chen et al.

This study shows how Transformers with recurrent position encodings mimic spatial representations in the hippocampal formation.

James C. R. Whittington, Joseph Warren, Timothy E. J. Behrens