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。
Shih-Yang Liu, Xin Dong, Ximing Lu +10 authors
Multi-reward reinforcement learning suffers from reward normalization collapse in GRPO, which GDPO addresses by decoupling reward normalization for improved training stability and performance across reasoning tasks.
50 篇论文 · 按点赞排序
Yoav HaCohen, Benny Brazowski, Nisan Chiprut +26 authors
LTX-2 is an open-source audiovisual diffusion model that generates synchronized video and audio content using a dual-stream transformer architecture with cross-modal attention and classifier-free guidance.
Yuxue Yang, Lue Fan, Ziqi Shi +3 authors
NeoVerse is a scalable 4D world model that enables pose-free reconstruction and novel-trajectory video generation from monocular videos with state-of-the-art performance.
Yuchen Shi, Yuzheng Cai, Siqi Cai +15 authors
Existing Large Language Model (LLM) agent frameworks face two significant challenges: high configuration costs and static capabilities. Building a high-quality agent often requires extensive manual effort in tool integration and prompt engineering, while deployed agents struggle to adapt to dynamic environments without expensive fine-tuning. To address these issues, we propose Youtu-Agent, a modular framework designed for the automated generation and continuous evolution of LLM agents. Youtu-Agent features a structured configuration system that decouples execution environments, toolkits, and context management, enabling flexible reuse and automated synthesis. We introduce two generation paradigms: a Workflow mode for standard tasks and a Meta-Agent mode for complex, non-standard requirements, capable of automatically generating tool code, prompts, and configurations. Furthermore, Youtu-Agent establishes a hybrid policy optimization system: (1) an Agent Practice module that enables agents to accumulate experience and improve performance through in-context optimization without parameter updates; and (2) an Agent RL module that integrates with distributed training frameworks to enable scalable and stable reinforcement learning of any Youtu-Agents in an end-to-end, large-scale manner. Experiments demonstrate that Youtu-Agent achieves state-of-the-art performance on WebWalkerQA (71.47\%) and GAIA (72.8\%) using open-weight models. Our automated generation pipeline achieves over 81\% tool synthesis success rate, while the Practice module improves performance on AIME 2024/2025 by +2.7\% and +5.4\% respectively. Moreover, our Agent RL training achieves 40\% speedup with steady performance improvement on 7B LLMs, enhancing coding/reasoning and searching capabilities respectively up to 35\% and 21\% on Maths and general/multi-hop QA benchmarks.
Muxi Diao, Lele Yang, Wuxuan Gong +6 authors
Entropy-Adaptive Fine-Tuning addresses catastrophic forgetting in supervised fine-tuning by using token-level entropy to distinguish uncertainty from knowledge conflict, enabling better preservation of general capabilities.
Hao Yu, Haotong Lin, Jiawei Wang +7 authors
InfiniDepth represents depth as neural implicit fields using a local implicit decoder, enabling continuous 2D coordinate querying for arbitrary-resolution depth estimation and superior performance in fine-detail regions.
Alex L. Zhang, Tim Kraska, Omar Khattab
We study allowing large language models (LLMs) to process arbitrarily long prompts through the lens of inference-time scaling. We propose Recursive Language Models (RLMs), a general inference strategy that treats long prompts as part of an external environment and allows the LLM to programmatically examine, decompose, and recursively call itself over snippets of the prompt. We find that RLMs successfully handle inputs up to two orders of magnitude beyond model context windows and, even for shorter prompts, dramatically outperform the quality of base LLMs and common long-context scaffolds across four diverse long-context tasks, while having comparable (or cheaper) cost per query.
Eunbi Choi, Kibong Choi, Seokhee Hong +62 authors
K-EXAONE is a multilingual language model with a Mixture-of-Experts architecture that achieves competitive performance on various benchmarks while supporting multiple languages and long-context windows.
Haochen Shi, Xingdi Yuan, Bang Liu
Programmatic Skill Network enables continual skill acquisition through executable symbolic programs that evolve via reflection, progressive optimization, and structural refactoring mechanisms.
Amirhosein Ghasemabadi, Di Niu
Large language models (LLMs) generate fluent and complex outputs but often fail to recognize their own mistakes and hallucinations. Existing approaches typically rely on external judges, multi-sample consistency, or text-based self-critique, which incur additional compute or correlate weakly with true correctness. We ask: can LLMs predict their own failures by inspecting internal states during inference? We introduce Gnosis, a lightweight self-awareness mechanism that enables frozen LLMs to perform intrinsic self-verification by decoding signals from hidden states and attention patterns. Gnosis passively observes internal traces, compresses them into fixed-budget descriptors, and predicts correctness with negligible inference cost, adding only ~5M parameters and operating independently of sequence length. Across math reasoning, open-domain question answering, and academic knowledge benchmarks, and over frozen backbones ranging from 1.7B to 20B parameters, Gnosis consistently outperforms strong internal baselines and large external judges in both accuracy and calibration. Moreover, it generalizes zero-shot to partial generations, enabling early detection of failing trajectories and compute-aware control. These results show that reliable correctness cues are intrinsic to generation process and can be extracted efficiently without external supervision.
MOSI. AI, Donghua Yu, Zhengyuan Lin +19 authors
A unified multimodal large language model for end-to-end speaker-attributed, time-stamped transcription with extended context window and strong generalization across benchmarks.
Huichao Zhang, Liao Qu, Yiheng Liu +33 authors
NextFlow is a unified decoder-only autoregressive transformer that processes interleaved text-image tokens, enabling fast multimodal generation through novel next-token and next-scale prediction strategies.
Zhe Huang, Hao Wen, Aiming Hao +6 authors
Multimodal Large Language Models (MLLMs) have made remarkable progress in video understanding. However, they suffer from a critical vulnerability: an over-reliance on language priors, which can lead to visual ungrounded hallucinations, especially when processing counterfactual videos that defy common sense. This limitation, stemming from the intrinsic data imbalance between text and video, is challenging to address due to the substantial cost of collecting and annotating counterfactual data. To address this, we introduce DualityForge, a novel counterfactual data synthesis framework that employs controllable, diffusion-based video editing to transform real-world videos into counterfactual scenarios. By embedding structured contextual information into the video editing and QA generation processes, the framework automatically produces high-quality QA pairs together with original-edited video pairs for contrastive training. Based on this, we build DualityVidQA, a large-scale video dataset designed to reduce MLLM hallucinations. In addition, to fully exploit the contrastive nature of our paired data, we propose Duality-Normalized Advantage Training (DNA-Train), a two-stage SFT-RL training regime where the RL phase applies pair-wise ell_1 advantage normalization, thereby enabling a more stable and efficient policy optimization. Experiments on DualityVidQA-Test demonstrate that our method substantially reduces model hallucinations on counterfactual videos, yielding a relative improvement of 24.0% over the Qwen2.5-VL-7B baseline. Moreover, our approach achieves significant gains across both hallucination and general-purpose benchmarks, indicating strong generalization capability. We will open-source our dataset and code.
Taekyung Ki, Sangwon Jang, Jaehyeong Jo +2 authors
Avatar Forcing framework enables real-time interactive head avatar generation with low latency and expressive motion through diffusion forcing and label-free preference optimization.
Xu Guo, Fulong Ye, Xinghui Li +6 authors
A novel video face swapping framework combines image face swapping techniques with diffusion transformers and curriculum learning to achieve superior identity preservation and visual realism.
Ruiyan Han, Zhen Fang, XinYu Sun +9 authors
UniCorn, a self-improvement framework for unified multimodal models, addresses generation gaps through self-play and cognitive pattern reconstruction, achieving state-of-the-art results in text-to-image generation.
Yuan-Kang Lee, Kuan-Lin Chen, Chia-Che Chang +1 authors
A novel nighttime color constancy framework combines statistical methods with deep reinforcement learning to improve white balance adjustment under low-light conditions.
Loïc Magne, Anas Awadalla, Guanzhi Wang +11 authors
NitroGen is a vision-action foundation model trained on extensive gameplay data that demonstrates strong cross-game generalization and effective transfer learning capabilities.
Ali Behrouz, Meisam Razaviyayn, Peilin Zhong +1 authors
Despite the recent progresses, particularly in developing Language Models, there are fundamental challenges and unanswered questions about how such models can continually learn/memorize, self-improve, and find effective solutions. In this paper, we present a new learning paradigm, called Nested Learning (NL), that coherently represents a machine learning model with a set of nested, multi-level, and/or parallel optimization problems, each of which with its own context flow. Through the lenses of NL, existing deep learning methods learns from data through compressing their own context flow, and in-context learning naturally emerges in large models. NL suggests a philosophy to design more expressive learning algorithms with more levels, resulting in higher-order in-context learning and potentially unlocking effective continual learning capabilities. We advocate for NL by presenting three core contributions: (1) Expressive Optimizers: We show that known gradient-based optimizers, such as Adam, SGD with Momentum, etc., are in fact associative memory modules that aim to compress the gradients' information (by gradient descent). Building on this insight, we present other more expressive optimizers with deep memory and/or more powerful learning rules; (2) Self-Modifying Learning Module: Taking advantage of NL's insights on learning algorithms, we present a sequence model that learns how to modify itself by learning its own update algorithm; and (3) Continuum Memory System: We present a new formulation for memory system that generalizes the traditional viewpoint of long/short-term memory. Combining our self-modifying sequence model with the continuum memory system, we present a continual learning module, called Hope, showing promising results in language modeling, knowledge incorporation, and few-shot generalization tasks, continual learning, and long-context reasoning tasks.
Jinyang Wu, Guocheng Zhai, Ruihan Jin +5 authors
ATLAS is a dual-path framework that dynamically selects optimal model-tool combinations for cross-domain reasoning through cluster-based routing and reinforcement learning-based multi-step routing, achieving superior performance on complex reasoning tasks.
Maksim Velikanov, Ilyas Chahed, Jingwei Zuo +3 authors
Learnable multipliers are introduced to address weight decay-induced normalization artifacts in large language model training, outperforming traditional methods while reducing computational overhead.
Jiawei Chen, Xintian Shen, Lihao Zheng +28 authors
MindWatcher is a tool-integrated reasoning agent that combines interleaved thinking and multimodal chain-of-thought reasoning to autonomously perform complex decision-making tasks through coordinated tool invocation without human prompts.
Nuoya Xiong, Yuhang Zhou, Hanqing Zeng +5 authors
FusionRoute is a token-level multi-LLM collaboration framework that uses a lightweight router to select optimal experts and add complementary logits, outperforming existing methods in diverse tasks while maintaining efficiency.
Bangjun Xiao, Bingquan Xia, Bo Yang +122 authors
MiMo-V2-Flash is a sparse Mixture-of-Experts model with hybrid attention architecture and efficient distillation technique that achieves strong performance with reduced parameters and improved inference speed.
Shuming Liu, Mingchen Zhuge, Changsheng Zhao +20 authors
VideoAuto-R1 framework employs a reason-when-necessary strategy for video understanding, using a Thinking Once, Answering Twice training paradigm with verifiable rewards and confidence-based reasoning activation during inference.
Jiaqi Liu, Yaofeng Su, Peng Xia +5 authors
To support reliable long-term interaction in complex environments, LLM agents require memory systems that efficiently manage historical experiences. Existing approaches either retain full interaction histories via passive context extension, leading to substantial redundancy, or rely on iterative reasoning to filter noise, incurring high token costs. To address this challenge, we introduce SimpleMem, an efficient memory framework based on semantic lossless compression. We propose a three-stage pipeline designed to maximize information density and token utilization: (1) Semantic Structured Compression, which applies entropy-aware filtering to distill unstructured interactions into compact, multi-view indexed memory units; (2) Recursive Memory Consolidation, an asynchronous process that integrates related units into higher-level abstract representations to reduce redundancy; and (3) Adaptive Query-Aware Retrieval, which dynamically adjusts retrieval scope based on query complexity to construct precise context efficiently. Experiments on benchmark datasets show that our method consistently outperforms baseline approaches in accuracy, retrieval efficiency, and inference cost, achieving an average F1 improvement of 26.4% while reducing inference-time token consumption by up to 30-fold, demonstrating a superior balance between performance and efficiency. Code is available at https://github.com/aiming-lab/SimpleMem.
Yiheng Wang, Yixin Chen, Shuo Li +33 authors
SciEvalKit is a unified benchmarking toolkit for evaluating AI models across diverse scientific disciplines, focusing on core scientific intelligence capabilities and supporting customizable, reproducible assessments.
Yong Xien Chng, Tao Hu, Wenwen Tong +10 authors
While Vision-Language Models (VLMs) can solve complex tasks through agentic reasoning, their capabilities remain largely constrained to text-oriented chain-of-thought or isolated tool invocation. They fail to exhibit the human-like proficiency required to seamlessly interleave dynamic tool manipulation with continuous reasoning, particularly in knowledge-intensive and visually complex scenarios that demand coordinated external tools such as search and image cropping. In this work, we introduce SenseNova-MARS, a novel Multimodal Agentic Reasoning and Search framework that empowers VLMs with interleaved visual reasoning and tool-use capabilities via reinforcement learning (RL). Specifically, SenseNova-MARS dynamically integrates the image search, text search, and image crop tools to tackle fine-grained and knowledge-intensive visual understanding challenges. In the RL stage, we propose the Batch-Normalized Group Sequence Policy Optimization (BN-GSPO) algorithm to improve the training stability and advance the model's ability to invoke tools and reason effectively. To comprehensively evaluate the agentic VLMs on complex visual tasks, we introduce the HR-MMSearch benchmark, the first search-oriented benchmark composed of high-resolution images with knowledge-intensive and search-driven questions. Experiments demonstrate that SenseNova-MARS achieves state-of-the-art performance on open-source search and fine-grained image understanding benchmarks. Specifically, on search-oriented benchmarks, SenseNova-MARS-8B scores 67.84 on MMSearch and 41.64 on HR-MMSearch, surpassing proprietary models such as Gemini-3-Flash and GPT-5. SenseNova-MARS represents a promising step toward agentic VLMs by providing effective and robust tool-use capabilities. To facilitate further research in this field, we will release all code, models, and datasets.
Qi Qian, Chengsong Huang, Jingwen Xu +13 authors
Researchers developed Benchmark^2, a framework with three metrics to evaluate benchmark quality for large language models, revealing significant variations in existing benchmarks and enabling more efficient evaluation through selective benchmark construction.
Yifan Zhang, Yifeng Liu, Mengdi Wang +1 authors
Deep Delta Learning introduces a novel residual connection mechanism that uses a learnable geometric transformation to generalize identity shortcuts, enabling more flexible feature modeling while maintaining stable training properties.
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