Proposes inference scaling laws showing smaller models with advanced strategies outperform larger ones under limited compute, achieving Pareto optimality.

Yangzhen Wu, Zhiqing Sun, Shanda Li et al.

SAM 2 employs streaming memory within a transformer framework for promptable image and video segmentation, achieving 3x fewer interactions and 6x faster speeds than prior models.

Nikhila Ravi, Valentin Gabeur, Yuan-Ting Hu et al.

Proposes MotionFix dataset and TMED diffusion model for text-driven 3D human motion editing, achieving significant improvements in editing accuracy.

Nikos Athanasiou, Alpár Cseke, Markos Diomataris et al.

DiscipLink leverages LLMs to generate exploratory questions, expand queries, and connect papers across disciplines, enhancing interdisciplinary literature search.

Chengbo Zheng, Yuanhao Zhang, Zeyu Huang et al.

RFNet+RFFT lift ad-image availability with RF1M, a 1M+ human-labeled dataset

Zhenbang Du, Wei Feng, Haohan Wang et al.

OmniParser significantly enhances GPT-4V performance on ScreenSpot benchmark using pure vision-based UI parsing.

Yadong Lu, Jianwei Yang, Yelong Shen et al.

Negative Attention Score Alignment (NASA) reduces negative bias in large language models, enhancing precision and recall.

Sangwon Yu, Jongyoon Song, Bongkyu Hwang et al.

Repeated sampling enhances inference coverage; on SWE-bench Lite, coverage increased from 15.9% to 56% with 250 samples, outperforming single-sample SOTA (43%).

Bradley Brown, Jordan Juravsky, Ryan Ehrlich et al.

Llama 3 employs a 405B-parameter dense Transformer supporting 128K tokens, multi-modal, and multi-language capabilities, rivaling GPT-4.

Aaron Grattafiori, Abhimanyu Dubey, Abhinav Jauhri et al.

MoMa employs modality-aware mixture-of-experts, achieving 3.7× FLOPs savings on 1.4B parameters with minimal performance loss.

Xi Victoria Lin, Akshat Shrivastava, Liang Luo et al.

Informed corrector using model guidance and hollow Transformer architecture improves discrete diffusion sampling efficiency and quality.

Yixiu Zhao, Jiaxin Shi, Feng Chen et al.

GradCraft uses dynamic gradient magnitude adjustment and global direction projection to improve multi-task recommendation models' gradient balance and performance.

Yimeng Bai, Yang Zhang, Fuli Feng et al.

Meta-Reward introduces self-judgment and meta-judgment to improve LLMs, achieving 39.4% win rate on AlpacaEval 2, surpassing previous methods.

Tianhao Wu, Weizhe Yuan, Olga Golovneva et al.

Proposes a Stackelberg game-based robust meta-learning framework using normalizing flows for explicit task distribution modeling, enhancing adaptation under distribution shifts.

Cheems Wang, Yiqin Lv, Yixiu Mao et al.

Proposes ctPuLSE, combining close-talk enhancement and pseudo-labels, to improve far-field speech enhancement generalization.

Zhong-Qiu Wang

SETO enriches samples via temporary operations on subsequences, effectively addressing data sparsity.

Shu Chen, Jinwei Luo, Weike Pan et al.

Nerva leverages sparse matrix operations via Intel MKL to accelerate neural network training, reducing time by 4× at 99% sparsity with comparable accuracy.

Wieger Wesselink, Bram Grooten, Qiao Xiao et al.

Hierarchical prompting with ROME boosts complex HDL generation success from 0 to 78%+ in LLMs, reducing time by 30%.

Andre Nakkab, Sai Qian Zhang, Ramesh Karri et al.

This study compares RAG and long-context LLMs, proposing Self-Route for cost-effective hybrid routing, achieving performance close to long-context models.

Zhuowan Li, Cheng Li, Mingyang Zhang et al.

C3T method improves accuracy by at least 8% in unsupervised modality adaptation.

Abhi Kamboj, Anh Duy Nguyen, Minh N. Do