3827 papers
This paper introduces Fine-grained Group Policy Optimization (FGO), a reinforcement learning algorithm that effectively compresses verbose Chain-of-Thought reasoning in Large Language Models while simultaneously addressing the data inefficiency and entropy collapse limitations of Group Relative Policy Optimization (GRPO).
UniWeTok is a unified binary tokenizer featuring a massive $2^{128}$ codebook, a convolution-attention hybrid architecture with SigLu activation, and a novel three-stage training framework that achieves state-of-the-art performance in image generation and multimodal understanding with significantly lower computational costs than existing models.
TikArt is a reinforcement learning-trained multimodal agent that stabilizes fine-grained visual reasoning by employing a Think-Aperture-Observe loop to sequentially acquire and linguistically encode evidence through zooming and segmentation, thereby overcoming the limitations of single-pass global image encoding.
The paper proposes GOT-JEPA, a model-predictive pretraining framework that learns to predict robust tracking models from corrupted observations to improve generalization, and introduces OccuSolver to enhance occlusion handling through iterative, object-aware visibility estimation.
This paper demonstrates that fully autonomous AI analysts can cheaply replicate the analytic diversity and conflicting conclusions observed in human many-analyst studies, revealing that empirical results are highly sensitive to analytic choices and prompting a new transparency norm requiring multiverse-style reporting and full prompt disclosure for AI-generated science.
This paper introduces LAVIDA, a zero-shot video anomaly detection framework that leverages a Multimodal Large Language Model and an Anomaly Exposure Sampler to train exclusively on pseudo-anomalies, achieving state-of-the-art performance across multiple benchmarks without requiring real anomaly data.
PatchDenoiser is a lightweight, parameter-efficient multi-scale patch-based framework that effectively denoises low-dose CT images by balancing noise suppression with anatomical detail preservation, outperforming state-of-the-art CNN and GAN methods while significantly reducing computational costs and energy consumption.
This paper introduces Hubscan, an open-source security scanner that utilizes a multi-detector architecture to identify and mitigate hubness poisoning attacks in Retrieval-Augmented Generation (RAG) systems, achieving high recall rates in detecting adversarial hubs across various vector databases and real-world benchmarks.
This paper proposes Alignment-Aware Masked Learning (AML), a training strategy that improves Referring Image Segmentation by quantifying pixel-level vision-language alignment to mask unreliable regions during optimization, thereby achieving state-of-the-art performance without architectural changes or inference overhead.
The paper introduces a minimal, open-source agentic baseline for automated theorem proving that achieves competitive performance against state-of-the-art systems through iterative refinement and library search, demonstrating superior sample efficiency and cost-effectiveness while serving as a reference for future research.
This paper identifies and quantifies "Defensive Refusal Bias," a safety alignment failure in large language models where legitimate cybersecurity defenders are disproportionately denied assistance for critical tasks due to the presence of security-sensitive keywords, a problem exacerbated by explicit authorization attempts and current reliance on semantic similarity rather than intent reasoning.
The paper introduces CARE, an evidence-grounded agentic framework that enhances clinical accountability and reasoning accuracy in multi-modal medical AI by decomposing tasks into specialized modules for entity proposal, pixel-level localization, and evidence-based reasoning, thereby outperforming state-of-the-art models on medical VQA benchmarks.
The paper proposes SEED-SET, a Bayesian experimental design framework that combines objective evaluations and subjective stakeholder preferences via hierarchical Gaussian Processes to efficiently and interpretably benchmark the ethical alignment of autonomous systems in high-stakes domains.
The paper introduces BrandFusion, a novel multi-agent framework that enables seamless brand integration in text-to-video generation by combining an offline brand knowledge base with an online iterative refinement process to balance semantic fidelity, brand recognizability, and contextual naturalness.
This paper presents the first comprehensive evaluation of parameter-efficient fine-tuning (PEFT) for multitask code analysis, demonstrating that a single shared PEFT module can match or surpass full fine-tuning performance while significantly reducing computational and storage costs, provided that tasks are strategically grouped based on factors like complementarity and stability.
This paper proposes Targeted Reasoning Unlearning (TRU), a novel framework that utilizes a reasoning-based unlearning target to guide models in precisely removing specific undesirable knowledge while preserving general capabilities and enhancing robustness against attacks.
This paper introduces AraModernBERT, an Arabic adaptation of the ModernBERT encoder that leverages transtokenized initialization and native long-context modeling up to 8,192 tokens to achieve significant improvements in both language modeling and downstream discriminative tasks.
MoE-SpAc is an efficient inference framework for Mixture-of-Experts models on heterogeneous edge devices that repurposes speculative decoding as a predictive sensor for memory management, achieving significant throughput improvements through dynamic workload balancing and asynchronous execution.
This empirical study demonstrates that large language models exhibit a Dunning-Kruger-like cognitive bias, where poorly performing models display significantly higher overconfidence and worse calibration than their more accurate counterparts.
This paper quantifies hallucinations in medical textbook-based QA by demonstrating that LLaMA-70B-Instruct hallucinated in nearly 20% of answers despite high plausibility, and found that lower hallucination rates generally correlate with higher clinician-rated usefulness across models.