BRAID leverages conservative reward modeling to fine-tune diffusion models, outperforming offline data with 15-20% reward gains while avoiding invalid designs.

Masatoshi Uehara, Yulai Zhao, Ehsan Hajiramezanali et al.

SparseDrive achieves end-to-end autonomous driving via sparse scene representation, reducing collision rate by 71.4%.

Wenchao Sun, Xuewu Lin, Yining Shi et al.

VideoTree constructs adaptive hierarchical video representations, achieving 61.1% accuracy for long video reasoning without training.

Ziyang Wang, Shoubin Yu, Elias Stengel-Eskin et al.

PathReasoner models reasoning paths with equivalent extension, excelling in logical QA.

Fangzhi Xu, Qika Lin, Tianzhe Zhao et al.

LatProtRL employs reinforcement learning in latent space to optimize protein fitness, outperforming baseline methods in key benchmarks.

Minji Lee, Luiz Felipe Vecchietti, Hyunkyu Jung et al.

Proposed QUIRE algorithm enhances CoT effectiveness and faithfulness by recalling extra information from questions; experiments show 2.4% effectiveness and 5.6% faithfulness improvement.

Jiachun Li, Pengfei Cao, Yubo Chen et al.

Proposes a language generation method using strict proper scoring rules (Brier, spherical), improving BLEU by 1.5-2.0 points on translation tasks.

Chenze Shao, Fandong Meng, Yijin Liu et al.

Proposes ICTM, an efficient MAP approximation for high-dimensional linear inverse problems using flow models, outperforming existing flow matching methods.

Yasi Zhang, Peiyu Yu, Yaxuan Zhu et al.

T2V-Turbo breaks video consistency model quality bottleneck with mixed reward feedback, achieving superior 4-step generation over 50-step DDIM.

Jiachen Li, Weixi Feng, Tsu-Jui Fu et al.

Phased Consistency Models (PCM) outperform LCM in 1-16 step generation, applicable to high-res images and videos, enabling state-of-the-art few-step text-to-video synthesis.

Fu-Yun Wang, Zhaoyang Huang, Alexander William Bergman et al.

EffoVPR leverages DINOv2 self-attention features for zero-shot and fine-tuned VPR, achieving state-of-the-art results with compact 128D features.

Issar Tzachor, Boaz Lerner, Matan Levy et al.

Survey on tool learning with large language models, analyzing benefits and implementation methods.

Changle Qu, Sunhao Dai, Xiaochi Wei et al.

MoSca: a 4D dynamic scene reconstruction system using Gaussian fusion from casual monocular videos, leveraging a novel motion scaffold and physics-inspired regularization, achieving state-of-the-art results.

Jiahui Lei, Yijia Weng, Adam Harley et al.

Using Abacus Embeddings, Transformers achieve 99% accuracy on 100-digit addition.

Sean McLeish, Arpit Bansal, Alex Stein et al.

Zamba is a 7B hybrid model combining Mamba backbone and shared attention, outperforming comparable transformers in speed and memory efficiency.

Paolo Glorioso, Quentin Anthony, Yury Tokpanov et al.

OptAgent uses directional feedback for text optimization; across four functions and 10-step trials, it is more stable and efficient.

Allen Nie, Ching-An Cheng, Andrey Kolobov et al.

KG-FIT leverages LLM-guided hierarchy construction and multi-objective fine-tuning to improve link prediction by 14.4% on FB15K-237.

Pengcheng Jiang, Lang Cao, Cao Xiao et al.

LLM agents with introspection show a 3.5% success rate boost and 45% efficiency gain in WebArena.

Haoyu Wang, Tao Li, Zhiwei Deng et al.

MambaLLIE uses global-then-local state space design to achieve PSNR of 30.12 in low-light image enhancement.

Jiangwei Weng, Zhiqiang Yan, Ying Tai et al.

COLT improves completeness-oriented tool retrieval; BERT-mini (11M) beats BERT-large (340M).

Changle Qu, Sunhao Dai, Xiaochi Wei et al.