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。
686 篇论文 · 按点赞排序
Yi Pan, Miao Pan, Qi Lu +8 authors
VLA-Corrector addresses limitations of action chunking in vision-language-action models by introducing a lightweight latent-space vision monitor that enables adaptive corrective replanning, improving robustness in contact-rich manipulation tasks.
Bytedance Seed
Seed2.0 addresses complex real-world tasks by tackling long-tail knowledge and complex instruction following challenges while enhancing reasoning, visual understanding, and search capabilities through a robust evaluation framework grounded in user needs.
Shangqing Tu, Daniel Zhang-Li, Yucheng Wang +21 authors
CogEvol is a family of models that generate structured learning artifacts in a single pass using supervised fine-tuning and reinforcement learning with vision-language rewards, achieving high quality with far fewer parameters and lower cost.
Zhoushun Yu, Xiaoyu Hu, Xiangyu Xu
Latent world models enable efficient planning by predicting future states in a compact representation space, but their performance depends critically on the quality of the learned latent distribution. LeWorldModel (LeWM) regularizes its latents toward an isotropic Gaussian using the Epps-Pulley (EP) objective. We show that the corrective gradients of EP rapidly vanish for isolated tail samples, leaving heavy-tailed deviations insufficiently controlled. To address this limitation, we propose QQWorld, which replaces EP with a quantile-quantile matching objective that directly aligns projected latent samples with rank-matched Gaussian quantiles, thereby maintaining effective corrective gradients in the tails. We further develop cross-batch QQ, which enlarges the effective ranking pool using detached samples from previous batches, and characterize its bias-variance trade-off. Across four control environments, QQWorld effectively improves the average planning success rate of LeWM, while consistently yielding better Gaussian alignment and thinner latent tails.
Sagi Polaczek, Noa Kraicer, Gal Metzer +4 authors
Audio-video diffusion models exhibit bidirectional semantic leakage through cross-modal attention pathways, which can be diagnosed via attention-derived signals and mitigated through inference-time alignment interventions.
Hangrui Xu, Zhengxian Wu, Yunyao Yu +6 authors
KBMR uses a multimodal language model to embed images by semantic identity rather than surface appearance, improving retrieval and visual question answering via continuous distillation and hard negative sampling.
Ayoub Kirouane, Christos Petrocheilos
Modern Greek is absent from NVIDIA's Nemotron retrieval models and from major multilingual retrieval benchmarks, despite being important for retrieval-augmented generation (RAG) in legal, energy, financial, and medical applications. We present an end-to-end adaptation of the Nemotron retrieval stack for Modern Greek, including corpus mining, synthetic supervision, retrieval model training, reranker adaptation, reader fine-tuning, and a new benchmark called HERA. Our study shows that a parameter-free BM25 baseline outperforms several off-the-shelf multilingual dense retrieval models on specialist Greek corpora. After fine-tuning on 65,773 Greek retrieval pairs, a Nemotron 1B embedder improves nDCG@10 from 0.362 to 0.835 and substantially outperforms its unadapted counterpart. The learned language competence transfers to general-domain Greek, although the advantage over BM25 remains domain-dependent. We further adapt a cross-encoder reranker and demonstrate consistent improvements across specialist domains. Finally, we LoRA-tune a Nemotron 30B-A3B mixture-of-experts reader for grounded generation, increasing judged answer correctness from 29.4% to 66.9% while significantly improving faithfulness and citation quality. We also introduce HERA, the first large-scale Greek benchmark for retrieval-augmented generation, and release our adapted models and benchmark to support future research on Greek-language RAG systems.
Zhishang Xiang, Zerui Chen, Yunbo Tang +5 authors
Memory plays a crucial role in LLM-based agents, but retrieved memories can cause sycophancy issues where agents over-align with users at the expense of factual accuracy, necessitating new evaluation benchmarks that assess memory's impact on reasoning and decision-making rather than just storage and retrieval.
Zhiqin Yang, Jingwen Fu, Yuhan Liu +16 authors
This work proposes a structured ladder for scaling large reasoning models beyond human supervision by tracing autonomous rewards and self-generated experience, while identifying risks and evaluation dimensions.
Cheng-De Fan, Chun-Wei Tuan Mu, Chen-Wei Chang +4 authors
LongE2V enables high-quality video recovery from sparse event streams by leveraging pre-trained video diffusion priors and addressing temporal stability and frame interpolation challenges.
Byeongho Heo, Jaehui Hwang, Sangdoo Yun +1 authors
On-Policy Distillation (OPD) is emerging as a promising alternative to reinforcement learning for LLM post-training, yet its effectiveness in multilingual settings remains underexplored. We study OPD and its advanced variant, On-Policy Delta Distillation (OPD^2), for mathematical reasoning in English, Korean, and Japanese. OPD^2 improves OPD by using the probability gap between a post-trained teacher and its base model as the learning signal. Experiments with Qwen3 show that OPD^2 consistently outperforms the original OPD, with particularly strong improvements in Korean and Japanese, and generally narrows the English-Korean performance gap. We further find that English-only OPD can also increase performance for Korean and Japanese, but often shifts the responses toward English, highlighting the importance of multilingual data to preserving target-language responses.
Yumeng Wang, Yuchen Wu, Cheng Qian +6 authors
The study proposes PRISK, a framework that reveals how personalized context in LLMs increases irrelevant personalization, preference narrowing, and sycophantic bias.
Fengqi Zhu, Shaoxuan Xu, Jingyang Ou +11 authors
Diffusion language models (dLLMs) offer an alternative to autoregressive (AR) language modeling, yet the scaling behavior of Mixture-of-Experts (MoE) dLLMs remains poorly understood. We systematically characterize how optimization hyperparameters, compute allocation, and architecture scale for MoE dLLMs, identifying quantitative differences from scaling trends previously reported for AR models. Specifically, for optimization, the optimal nominal batch size grows faster, while the optimal learning rate decays more rapidly with compute. For model--data allocation, IsoFLOP analysis reveals a slight data-side tilt: the optimal token budget grows faster than activated model-side computation. For MoE architecture, larger scales increasingly favor larger expert pools at fixed activated capacity, while moderate expert granularity remains consistently effective and the preferred fraction of activated capacity assigned to shared experts remains stable across scales. Guided by these findings, we train LLaDA MoE v2, a 30B-A3B dLLM, from scratch on 23.5T tokens. With approximately 65\% as many pretraining tokens as Qwen3, LLaDA MoE v2 approaches Qwen3 on several knowledge, reasoning, and coding benchmarks. After supervised fine-tuning alone, it outperforms SDAR Chat on seven of eight reasoning and coding benchmarks and remains close to Qwen3 on several tasks. These results establish practical scaling laws and design principles for MoE dLLMs.
Senyu Fei, Xiaopeng Yu, Siyin Wang +3 authors
Reinforcement learning (RL) post-training of Vision-Language-Action (VLA) models has shown strong promise for robotic manipulation. Among RL methods, critic-based approaches rely on a value estimator that predominantly operates on single-frame observations or single-frame VLM backbone latents, which is a fundamental mismatch with the partially observable nature of robot control. A naive approach to incorporate observation history into the critic incurs exponential complexity with high-dimensional visual space, and still fails because pure scalar-return regression provides insufficient supervision for learning cross-temporal dynamics. We identify the root cause as a state approximation problem: without an explicit world modeling objective, the critic's representation cannot capture the temporal structure needed for accurate value estimation. To address this, we propose the World Critic Model (WCM), built on a lightweight LeJEPA architecture; WCM jointly predicts future latent state and estimates values, such that the critic's representation is explicitly trained to capture temporal dynamics rather than merely regress scalar returns. WCM integrates seamlessly into both on-policy and off-policy training pipelines and is compatible with state-of-the-art VLA backbones including Pi0, Pi0.5, and OpenVLA-OFT. Extensive experiments on 149 tasks across four benchmarks demonstrate that WCM consistently achieves state-of-the-art performance in both in-distribution and out-of-distribution settings, with particularly strong generalization gains. We further validate WCM on seven real-world manipulation tasks using OpenVLA-OFT and Pi0.5 with off-policy RL, confirming stable deployment across diverse settings.
Jaewoo Park, Minyoung Lee, Sukmin Seo +11 authors
DroneCATS evaluates multimodal language models as drone controllers and finds that small open models navigate well but fail at protocol adherence and termination, highlighting a gap between spatial perception and disciplined action.
Shijie Li, Yilin Gao, Siyuan Yang +7 authors
Asymmetric Mutual Variational Learning addresses train-inference mismatch in multimodal reasoning by using bidirectional calibration to prevent answer leakage and improve latent-space stability.
Siyu Yan, Yizhen Gao, Yilin Wang +2 authors
DataEvolver is a self-evolving multi-agent framework that improves text-rich image generation by leveraging feedback from rejected samples to iteratively enhance data quality.
Jiwen Yu, Jianxiong Gao, Jianhong Bai +7 authors
MemLearner improves video world models by using learning-based adaptive context querying with query tokens to enhance scene consistency and memory in long video sequences with occlusions and dynamic objects.
Daegyu Sung, Yukyeong Lee, Geon Park +2 authors
A shared-library agent framework reduces redundancy and structural erosion when sequentially generating related applications by guiding cross-application code extraction and migration.
Yinghui He, Ling Yang, Jiarui Liu +6 authors
Long-horizon reasoning in recent LLMs demands that the model switch between distinct skills inside a reasoning chain, such as first doing a math derivation, then using the result to plan a schedule. We call such problems cross-skill long-horizon tasks: multi-step tasks whose steps require different reasoning skills and depend on earlier outputs. Existing benchmarks often evaluate individual skills, lacking a principled way to measure how well a model switches between skills. We address this gap from both the evaluation and training sides. We introduce Skill Entropy, a measure of the difficulty of switching from one skill to another. We then propose Skill^2-Bench, a benchmark of cross-skill long-horizon tasks built over 558 skills across 9 verifiable and open-ended domains. Each task is assigned a task-level skill-entropy score and grouped into three difficulty levels. Evaluating 8 frontier and 4 open-source models on Skill^2-Bench reveals a skill-switching gap: accuracy decreases on higher-entropy tasks. We then turn skill entropy from a benchmark scale into a training signal. We propose Skill-Entropy RL, an RL framework where the model predicts not only the answer at each step but also the skill used to produce it. The reward combines step-level correctness with a skill-entropy reward that measures the alignment between the model-predicted skill sequence and the gold skill sequence. On Qwen3-4B-Instruct and Qwen3-1.7B, Skill-Entropy RL improves the Skill^2-Bench score from 34.4% to 68.4% and from 14.6% to 40.1%, respectively, outperforming competitive baselines. The same pipeline can be applied to off-the-shelf training data such as OpenR1-Math, indicating that skill entropy is a reusable training signal. Code available at: https://github.com/Gen-Verse/Skill-Entropy-RL
Koorosh Roohi, Javad Rajabi, Andrew Fleet +1 authors
PhotoQuilt is a training-free framework that generates high-resolution photomosaics by combining global layout composition with separate tile generation in latent space, overcoming limitations of diffusion models in balancing local detail and global structure.
Wanshun Su, Yang Shi, Feihu Liu +10 authors
Omni-modal large language models (Omni-LLMs) have achieved remarkable performance on audio-visual understanding tasks, but processing long and highly redundant visual and audio token sequences incurs substantial computational overhead, demanding aggressive token compression for efficient deployment. Existing methods often degrade at low token budgets: pre-LLM compression may discard structurally important and globally distributed evidence, whereas inner-LLM compression often underexploits query-conditioned audio-visual collaboration. To address these limitations, we propose OmniPack, a training-free framework that coordinates structural compression before the LLM with task-relevant semantic refinement within the LLM. Before the LLM, OmniPack removes structural redundancy through modality-specific importance, global coverage, and similarity-aware merging. After sufficient multimodal interaction, it further consolidates diverse, task-relevant representations through textual guidance and audio-visual collaboration. Extensive experiments on five benchmarks with three Omni-LLM backbones demonstrate that OmniPack consistently achieves the best performance-efficiency trade-off across diverse retention ratios, outperforming all existing methods. Notably, on Qwen2.5-Omni-7B, OmniPack preserves 98.0% of the original performance while reducing FLOPs to 16.7%, and still retains 92.9% of the original performance with only 6.8% of the original FLOPs.
Junha Jung, Minbyul Jeong, Suhyeon Lim +5 authors
A reinforcement learning approach called MRPO is introduced to improve clinical image reasoning by addressing cascading errors through step-wise process rewards, demonstrating superior performance over existing methods.
Siming Fu, Zheming Fu, Ruizhe He +7 authors
On-policy distillation, in which a teacher corrects samples that the student itself generates, presupposes that the two models speak the same language: identical VAE latents, matching architectures, and a common timestep grid. We ask what happens when none of this holds, as when the strongest teacher available and the student one wishes to deploy come from different model families, and find that the standard recipes have no answer: teacher latents cannot serve as targets in a foreign coordinate system, per-pixel losses against a teacher that stochastically re-draws local detail degenerate into blur or divergence, and timestep indices lose their meaning across mismatched schedules. We present Any-OPD, to our knowledge the first framework for on-policy distillation between arbitrary pairs of latent flow-matching generators. Any-OPD treats the teacher purely as a black-box sampler and connects the two models at exactly one point: a frozen, model-agnostic vision representation in which their independently decoded outputs are compared, sidestepping every assumption about latents, features, or architecture. Trajectory correspondence is recovered by matching continuous noise levels instead of step indices, and a brief anchoring phase, in which teacher samples are re-encoded through the student's own VAE, ensures the on-policy gradient measures sample quality rather than domain mismatch. Distilling the 12B FLUX.1-dev into the 2.5B SD3.5-Medium, Any-OPD lifts the student's PickScore from 0.846 to 0.884 and HPSv3 from 9.12 to 10.97, rivaling the teacher at a fifth of its size, where direct latent regression fails to train at all.
Tingyu Song, Mingxin Li, Yanzhao Zhang +5 authors
CORE distills compositional ranking judgments from a cross-attentive reranker into an embedding model via synthesized multi-level candidates and a Rank-KL objective, improving compositional retrieval without degrading standard performance.
Aurélien Lac, Tony Wu
NeoMME introduces small bidirectional multimodal encoders pretrained with masked discrete diffusion that achieve strong visual document retrieval and high compression of late-interaction embeddings.
Mingxuan Xia, Yuhang Yang, Chao Ye +7 authors
Rubric-based RL has recently shown promise in improving LLMs on open-ended tasks. A widely recognized limitation of rubric-based RL is limited exploration: criteria that no rollout manages to satisfy (Unexplored Criteria, UC) receive no optimization signal. Recent methods address this by incorporating rubric information as external guidance during rollout, yet they introduce a train-inference mismatch: the policy is optimized on rollouts produced under external guidance while this guidance is absent at inference time, causing error accumulation through autoregressive decoding. Moreover, these exploration-focused approaches overlook a fundamentally different failure mode that we term Suppressed Criteria (SC) -- criteria that are satisfied by some rollouts yet whose learning signals are lost during optimization because scalar reward aggregation assigns them non-positive aggregate advantages. Our analysis reveals that SC are remarkably prevalent: over 57% of samples exhibit this failure mode throughout training, with an average of 1.8 SC per sample. To simultaneously address both UC and SC without introducing training-inference mismatch, we propose Criterion-Distilled Policy Optimization (CriPO), which enhances rubric-based RL via on-policy self-distillation. For UC, CriPO constructs a criterion-injection self-teacher and computes a localized forward-KL loss to inject missing behaviors into the policy. For SC, CriPO employs a counterfactual self-teacher to locate criterion-relevant tokens in negative-advantage rollouts and flips their token-level advantages to positive values, preserving useful patterns that would otherwise be suppressed. Experiments on medicine and science benchmarks demonstrate that CriPO consistently outperforms rubric-based RL, achieving stronger final performance with approximately 2times fewer optimization steps.
Mingda Lin, Weijie Wang, Zeyu Zhang +7 authors
ZipTok3D organizes 3D geometry into compact global-token prefixes with iterative decoding to achieve high-fidelity reconstruction from extremely short token sequences.
Yuqiao Tan, Jinxiang Meng, Fangyu Lei +4 authors
Real-world software development requires coding agents to operate in shared workspaces where users may inspect and modify code during an ongoing task, yet existing repository-level benchmarks typically evaluate agents working alone or restrict user participation to messages. This leads us to ask: how do coding agents understand and respond to code changes in a shared workspace? We introduce SWE-Touch, a framework that stress-tests this setting through validated Counter-Edits: plausible edits to task-relevant code that conflict with task completion. SWE-Touch mines task-critical regions from multiple repair trajectories, uses a separate User Patch Generator to construct the edits, and injects them with contextual user messages when agents reach the relevant code. We evaluate nine coding models on SWE-bench Verified, with additional experiments on longer-horizon tasks from SWE-Bench Pro and DeepSWE. Counter-Edit lowers average resolve rate by 7.7 percentage points on SWE-bench Verified, with degradation also persisting on both longer-horizon benchmarks. Trajectory analysis links these failures to limited awareness of the evolving workspace: agents may retain conflicting code or replace it without sufficiently re-inspecting the repository and validating the revised code with targeted tests. These findings show that strong autonomous performance does not yet ensure the state awareness and adaptive behavior needed for shared-workspace collaboration, and point to detecting workspace changes, reconciling conflicting edits with the task, and verifying the affected behavior as key capabilities for future optimization.
Julia Belikova, Rauf Parchiev, Evgeny Egorov +4 authors
Procedural memory enhances LLM agents on workplace tasks through skill transfer across roles and models, with varying generalization capabilities affecting deployment strategies.
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