This survey reviews LLM-based multi-agent systems, highlighting architectures like DyLAN, ProAgent, and their performance in software, social, and strategic tasks.

Taicheng Guo, Xiuying Chen, Yaqi Wang et al.

Introduced Missing Modality Token (MMT) to reduce performance drop to 10% on Ego4D dataset.

Merey Ramazanova, Alejandro Pardo, Humam Alwassel et al.

Medusa accelerates LLM inference by adding multiple decoding heads and tree attention, achieving 2.2-3.6x speedup without quality loss.

Tianle Cai, Yuhong Li, Zhengyang Geng et al.

Proposed OFMS-FT algorithm enables memory-constrained devices to perform federated online model selection and fine-tuning, achieving superior performance on regression and classification tasks.

Pouya M. Ghari, Yanning Shen

ChatQA surpasses GPT-4 in conversational QA and RAG using a two-stage instruction tuning and dense retriever.

Zihan Liu, Wei Ping, Rajarshi Roy et al.

Self-Rewarding language model using iterative DPO training, surpassing Claude 2, Gemini Pro, and GPT-4 0613 on AlpacaEval 2.0 with Llama 2 70B.

Weizhe Yuan, Richard Yuanzhe Pang, Kyunghyun Cho et al.

Introduces R-Judge benchmark, evaluates LLMs' safety risk detection with 74.45% top score, revealing significant room for improvement.

Tongxin Yuan, Zhiwei He, Lingzhong Dong et al.

SymbolNet uses adaptive dynamic pruning for neural symbolic regression, enhancing compression efficiency.

Ho Fung Tsoi, Vladimir Loncar, Sridhara Dasu et al.

Introduces geometrically smoothed momentum in randomized Kaczmarz, accelerating convergence along small singular vectors.

Seth J. Alderman, Roan W. Luikart, Nicholas F. Marshall

Vlogger combines LLM and diffusion models to generate 5-minute coherent long videos, maintaining script-actor consistency.

Shaobin Zhuang, Kunchang Li, Xinyuan Chen et al.

POP-3D introduces a zero-shot open-vocabulary 3D occupancy prediction model leveraging multimodal self-supervised learning, achieving 78% mIoU on nuScenes without manual annotations.

Antonin Vobecky, Oriane Siméoni, David Hurych et al.

SceneVerse constructs a million-scale 3D scene-language dataset and introduces GPS, a contrastive pre-training framework, achieving SOTA in grounded scene understanding.

Baoxiong Jia, Yixin Chen, Huangyue Yu et al.

Proposes fine-tuning spatial modules with high-quality images, enabling high-quality video diffusion training using only low-quality videos and synthetic images.

Haoxin Chen, Yong Zhang, Xiaodong Cun et al.

SeeClick leverages GUI grounding pre-training, significantly improving visual GUI agent performance across platforms, enabling screenshot-based task automation.

Kanzhi Cheng, Qiushi Sun, Yougang Chu et al.

Introduced IQC method to create MMIQC dataset; Qwen-72B-MMIQC achieves 45.0% accuracy on MATH benchmark.

Haoxiong Liu, Yifan Zhang, Yifan Luo et al.

Proposes MMToM-QA benchmark integrating Bayesian inverse planning and language models to enhance machine Theory of Mind in multimodal scenarios.

Chuanyang Jin, Yutong Wu, Jing Cao et al.

Introduces multi-track timeline control for text-driven 3D human motion, using a test-time denoising method with pre-trained diffusion models, achieving fine-grained multi-action synthesis.

Mathis Petrovich, Or Litany, Umar Iqbal et al.

Contrastive Preference Optimization (CPO) boosts 7B/13B LLM translation, surpassing SFT, matching GPT-4 performance.

Haoran Xu, Amr Sharaf, Yunmo Chen et al.

TACO constructs a large-scale bimanual tool-action-object dataset supporting generalization tasks with 2.5K sequences and 5.2M frames.

Yun Liu, Haolin Yang, Xu Si et al.

Proposes a low-latency, resource-efficient multichannel speech enhancement framework using joint training of two lightweight DNNs and a neural Wiener filter, achieving 2ms latency.

Tsun-An Hsieh, Jacob Donley, Daniel Wong et al.