DisCo-Diff integrates discrete latent variables into diffusion models, reducing ODE curvature and improving image generation, achieving SOTA FID 1.65 on ImageNet-128.

Yilun Xu, Gabriele Corso, Tommi Jaakkola et al.

This study revisits classical NCA, integrating deep features and neighbor sampling, achieving top performance on 300 datasets comparable to CatBoost.

Han-Jia Ye, Huai-Hong Yin, De-Chuan Zhan et al.

Programmable photonic extreme learning machine (PELM) using hexagonal waveguide mesh, achieving >98% accuracy on classification tasks, enhanced by evolutionary optimization and WDM ensemble.

Jose Roberto Rausell-Campo, Antonio Hurtado, Daniel Pérez-López et al.

Proposes Dy-DCA, a content-aware single-model super-resolution framework with dynamic routing and compiler optimization, achieving 33FPS on mobile with 1.7× speedup.

Gen Li, Zhihao Shu, Jie Ji et al.

GlyphDraw2 combines diffusion models and LLMs for automatic complex poster generation, supporting multilingual text and font control.

Jian Ma, Yonglin Deng, Chen Chen et al.

Proposes Entropic Optimal Transport eigenmaps for nonlinear alignment and joint embedding of high-dimensional datasets, with theoretical guarantees.

Boris Landa, Yuval Kluger, Rong Ma

Introduced MMLongBench-Doc, a benchmark with 135 long PDFs, evaluating 14 LVLMs; top F1 score is 42.7%.

Yubo Ma, Yuhang Zang, Liangyu Chen et al.

Open-TeleVision integrates active stereo vision and immersive VR to enable high-precision remote teleoperation, improving data collection for imitation learning.

Xuxin Cheng, Jialong Li, Shiqi Yang et al.

FoleyCrafter integrates pre-trained text-to-audio models with semantic adapters and temporal controllers for high-quality, synchronized video sound synthesis.

Yiming Zhang, Yicheng Gu, Yanhong Zeng et al.

Proposes active inheritance using non-differentiable metrics to steer synthetic data, improving attributes like lexical diversity and reducing toxicity in LLMs

Luísa Shimabucoro, Sebastian Ruder, Julia Kreutzer et al.

uDistil-Whisper achieves label-free knowledge distillation in low-data regimes, improving performance by 5-7 WER points.

Abdul Waheed, Karima Kadaoui, Bhiksha Raj et al.

FineSurE uses LLMs for fine-grained summarization evaluation, improving completeness and conciseness.

Hwanjun Song, Hang Su, Igor Shalyminov et al.

VIS framework calibrates OMSV for efficient yield estimation, achieving 29.03x speedup.

Yanfang Liu, Lei He, Wei W. Xing

This study assesses the necessity of Bengali-specific LLMs, revealing challenges in tokenization and bias, with LLaMA-3 outperforming fine-tuned models in understanding tasks.

Tamzeed Mahfuz, Satak Kumar Dey, Ruwad Naswan et al.

Proposes a two-dimensional taxonomy for long-context task difficulty based on information diffusion and scope, highlighting under-explored high-diffusion, high-scope scenarios.

Omer Goldman, Alon Jacovi, Aviv Slobodkin et al.

Perception Stitching enables zero-shot transfer of visuomotor policies, significantly improving success rates.

Pingcheng Jian, Easop Lee, Zachary Bell et al.

ShortcutsBench provides a large-scale benchmark to evaluate API agents in complex tasks.

Haiyang Shen, Yue Li, Desong Meng et al.

Layer swapping and deletion experiments reveal four universal inference stages in LLMs, showing high robustness especially in middle layers, with detailed neural mechanisms identified.

Vedang Lad, Jin Hwa Lee, Wes Gurnee et al.

CORE4D combines real MoCap and synthetic retargeting to create 11K 4D human-object-human interaction sequences for collaborative object rearrangement.

Yun Liu, Chengwen Zhang, Ruofan Xing et al.

ColPali employs vision-language models with multi-vector embeddings and late interaction to enhance document retrieval, outperforming traditional methods by 20% on ViDoRe.

Manuel Faysse, Hugues Sibille, Tony Wu et al.