Explore · 每周精选
发现最受关注的研究论文,追踪研究趋势,订阅感兴趣的期刊与关键词。
50 篇论文 · 按点赞排序
Jiaqi Peng, Wenzhe Cai, Yuqiang Yang +3 authors
LoGoPlanner is an end-to-end navigation framework that improves trajectory planning in unstructured environments by integrating localization, scene geometry reconstruction, and policy conditioning.
Zhe Cao, Tao Wang, Jiaming Wang +10 authors
Text-to-Audio-Video (T2AV) generation aims to synthesize temporally coherent video and semantically synchronized audio from natural language, yet its evaluation remains fragmented, often relying on unimodal metrics or narrowly scoped benchmarks that fail to capture cross-modal alignment, instruction following, and perceptual realism under complex prompts. To address this limitation, we present T2AV-Compass, a unified benchmark for comprehensive evaluation of T2AV systems, consisting of 500 diverse and complex prompts constructed via a taxonomy-driven pipeline to ensure semantic richness and physical plausibility. Besides, T2AV-Compass introduces a dual-level evaluation framework that integrates objective signal-level metrics for video quality, audio quality, and cross-modal alignment with a subjective MLLM-as-a-Judge protocol for instruction following and realism assessment. Extensive evaluation of 11 representative T2AVsystems reveals that even the strongest models fall substantially short of human-level realism and cross-modal consistency, with persistent failures in audio realism, fine-grained synchronization, instruction following, etc. These results indicate significant improvement room for future models and highlight the value of T2AV-Compass as a challenging and diagnostic testbed for advancing text-to-audio-video generation.
Ming Li, Han Chen, Yunze Xiao +3 authors
Large Language Models struggle to accurately estimate human cognitive difficulty due to a misalignment with human perceptions and a lack of introspection regarding their own limitations.
Bowen Shi, Andros Tjandra, John Hoffman +11 authors
SAM Audio, a diffusion transformer-based foundation model, achieves superior performance in general audio separation using unified text, visual, and temporal span prompts across various audio types.
Haonan Qiu, Shikun Liu, Zijian Zhou +10 authors
High-resolution video generation, while crucial for digital media and film, is computationally bottlenecked by the quadratic complexity of diffusion models, making practical inference infeasible. To address this, we introduce HiStream, an efficient autoregressive framework that systematically reduces redundancy across three axes: i) Spatial Compression: denoising at low resolution before refining at high resolution with cached features; ii) Temporal Compression: a chunk-by-chunk strategy with a fixed-size anchor cache, ensuring stable inference speed; and iii) Timestep Compression: applying fewer denoising steps to subsequent, cache-conditioned chunks. On 1080p benchmarks, our primary HiStream model (i+ii) achieves state-of-the-art visual quality while demonstrating up to 76.2x faster denoising compared to the Wan2.1 baseline and negligible quality loss. Our faster variant, HiStream+, applies all three optimizations (i+ii+iii), achieving a 107.5x acceleration over the baseline, offering a compelling trade-off between speed and quality, thereby making high-resolution video generation both practical and scalable.
Chao Xu, Suyu Zhang, Yang Liu +11 authors
This survey provides a structured overview of Vision-Language-Action models, detailing key challenges and opportunities in areas such as representation, execution, generalization, safety, and datasets in the field of robotics.
Junho Lee, Kwanseok Kim, Joonseok Lee
A novel framework for flow matching combines norm-aligned training with directionally-pruned sampling to improve generation quality and sampling efficiency by addressing mode discrepancy and path entanglement.
Prime Intellect Team, Mika Senghaas, Fares Obeid +20 authors
INTELLECT-3, a large Mixture-of-Experts model trained with reinforcement learning, achieves top performance across various benchmarks and is supported by an open-source RL infrastructure framework.
Rang Li, Lei Li, Shuhuai Ren +10 authors
GroundingME benchmark evaluates multimodal large language models' visual grounding capabilities across complexity dimensions, revealing significant gaps and proposing strategies for improvement.
Jiajun Wu, Jian Yang, Wei Zhang +6 authors
IPC is an unsupervised framework that uses internal probing of large language models to generate code without labeled datasets, achieving competitive performance with reduced resource dependency.
Rujiao Long, Yang Li, Xingyao Zhang +7 authors
Reasoning Palette enhances large language models by using a latent-modulation framework to guide internal planning and improve both inference and reinforcement learning performance.
Jiacheng Guo, Ling Yang, Peter Chen +6 authors
GenEnv, a framework using a co-evolutionary game with a generative environment simulator, enhances LLM agent performance by 40.3% over 7B baselines and uses less data than offline augmentation.
Kaiwen Zhang, Liming Jiang, Angtian Wang +6 authors
StoryMem enhances multi-shot video generation with cinematic quality and long-range consistency using a memory bank and pre-trained single-shot video diffusion models.
Benjamin Minixhofer, Tyler Murray, Tomasz Limisiewicz +6 authors
Bolmo, a family of competitive byte-level language models, is trained by converting existing subword-level models, overcoming character understanding and efficiency limitations while achieving performance comparable to subword models.
Gül Sena Altıntaş, Malikeh Ehghaghi, Brian Lester +4 authors
TokSuite enables systematic研究 of tokenization effects on language model performance by providing identical models with different tokenizers and a benchmark for real-world perturbations.
Jian Yang, Shawn Guo, Lin Jing +8 authors
Systematic exploration of scaling laws for multilingual code pre-training reveals language-specific benefits and proposes a strategy for optimal token allocation across programming languages.
Jin Qin, Zihan Liao, Ziyin Zhang +3 authors
We present C2LLM - Contrastive Code Large Language Models, a family of code embedding models in both 0.5B and 7B sizes. Building upon Qwen-2.5-Coder backbones, C2LLM adopts a Pooling by Multihead Attention (PMA) module for generating sequence embedding from token embeddings, effectively 1) utilizing the LLM's causal representations acquired during pretraining, while also 2) being able to aggregate information from all tokens in the sequence, breaking the information bottleneck in EOS-based sequence embeddings, and 3) supporting flexible adaptation of embedding dimension, serving as an alternative to MRL. Trained on three million publicly available data, C2LLM models set new records on MTEB-Code among models of similar sizes, with C2LLM-7B ranking 1st on the overall leaderboard.
Chenkai Xu, Yijie Jin, Jiajun Li +8 authors
LoPA, a training-free algorithm, enhances the parallelism of diffusion large language models, doubling the tokens per forward pass and boosting throughput with multi-GPU deployment.
Ming Li, Chenrui Fan, Yize Cheng +2 authors
Large language models increasingly expose reasoning traces, yet their underlying cognitive structure and steps remain difficult to identify and analyze beyond surface-level statistics. We adopt Schoenfeld's Episode Theory as an inductive, intermediate-scale lens and introduce ThinkARM (Anatomy of Reasoning in Models), a scalable framework that explicitly abstracts reasoning traces into functional reasoning steps such as Analysis, Explore, Implement, Verify, etc. When applied to mathematical problem solving by diverse models, this abstraction reveals reproducible thinking dynamics and structural differences between reasoning and non-reasoning models, which are not apparent from token-level views. We further present two diagnostic case studies showing that exploration functions as a critical branching step associated with correctness, and that efficiency-oriented methods selectively suppress evaluative feedback steps rather than uniformly shortening responses. Together, our results demonstrate that episode-level representations make reasoning steps explicit, enabling systematic analysis of how reasoning is structured, stabilized, and altered in modern language models.
Dan Ben-Ami, Gabriele Serussi, Kobi Cohen +1 authors
HERBench is a VideoQA benchmark that tests models' ability to integrate multiple, temporally separated visual cues, revealing significant challenges in current Video-LLMs.
北京市昌平区探索星信息技术及软件开发工作室
京ICP备2026059466号