# Research Papers — 2025-12 · bundle page 60 of 175 > Full-text markdown of papers #2951–3000 of 8703 published in 2025-12, newest first. AI-readable mirrors of the full 8703-paper 2025-12 archive (50 papers per file). Canonical page: https://aipod.dev/papers/archive/2025-12/page/60 [← Previous bundle](https://aipod.dev/papers/archive/2025-12/page/59.md) [Next bundle →](https://aipod.dev/papers/archive/2025-12/page/61.md) ### Physics-Informed Lightweight Machine Learning for Aviation Visibility Nowcasting Across Multiple Climatic Regimes Authors: Marcelo Cerda Castillo Published: 2025-12-18 Categories: cs.LG, physics.ao-ph arXiv: https://arxiv.org/abs/2512.16967v1 Short-term prediction (nowcasting) of low-visibility and precipitation events is critical for aviation safety and operational efficiency. Current operational approaches rely on computationally intensive numerical weather prediction guidance and human-issued TAF products, which often exhibit conservative biases and limited temporal resolution. This study presents a lightweight gradient boosting framework (XGBoost) trained exclusively on surface observation data (METAR) and enhanced through physics-guided feature engineering based on thermodynamic principles. The framework is evaluated across 11 international airports representing distinct climatic regimes (including SCEL, KJFK, KORD, KDEN, SBGR, and VIDP) using historical data from 2000 to 2024. Results suggest that the model successfully captures underlying local physical processes without manual configuration. In a blind comparative evaluation against operational TAF forecasts, the automated model achieved substantially higher detection rates at tactical horizons (3 hours), with a 2.5 to 4.0 times improvement in recall while reducing false alarms. Furthermore, SHAP analysis reveals that the model performs an implicit reconstruction of local physical drivers (advection, radiation, and subsidence), providing actionable explainability for operational situational awareness. Keywords: aviation meteorology; physics-guided machine learning; explainable artificial intelligence; lightweight machine learning; nowcasting; METAR; TAF verification; edge computing --- ### Colormap-Enhanced Vision Transformers for MRI-Based Multiclass (4-Class) Alzheimer's Disease Classification Authors: Faisal Ahmed Published: 2025-12-18 Categories: eess.IV, cs.CV, cs.LG arXiv: https://arxiv.org/abs/2512.16964v1 Magnetic Resonance Imaging (MRI) plays a pivotal role in the early diagnosis and monitoring of Alzheimer's disease (AD). However, the subtle structural variations in brain MRI scans often pose challenges for conventional deep learning models to extract discriminative features effectively. In this work, we propose PseudoColorViT-Alz, a colormap-enhanced Vision Transformer framework designed to leverage pseudo-color representations of MRI images for improved Alzheimer's disease classification. By combining colormap transformations with the global feature learning capabilities of Vision Transformers, our method amplifies anatomical texture and contrast cues that are otherwise subdued in standard grayscale MRI scans. We evaluate PseudoColorViT-Alz on the OASIS-1 dataset using a four-class classification setup (non-demented, moderate dementia, mild dementia, and very mild dementia). Our model achieves a state-of-the-art accuracy of 99.79% with an AUC of 100%, surpassing the performance of recent 2024--2025 methods, including CNN-based and Siamese-network approaches, which reported accuracies ranging from 96.1% to 99.68%. These results demonstrate that pseudo-color augmentation combined with Vision Transformers can significantly enhance MRI-based Alzheimer's disease classification. PseudoColorViT-Alz offers a robust and interpretable framework that outperforms current methods, providing a promising tool to support clinical decision-making and early detection of Alzheimer's disease. --- ### Compression is Routing: Reconstruction Error as an Intrinsic Signal for Modular Language Models Authors: Zhongpan Tang Published: 2025-12-18 Categories: cs.LG arXiv: https://arxiv.org/abs/2512.16963v2 Current Large Language Models (LLMs) face three major challenges: context length limitations, high inference costs, and catastrophic forgetting during continual learning. While Mixture-of-Experts (MoE) architectures mitigate some of these conflicts, their routing mechanisms typically rely on explicitly trained auxiliary classifiers. This not only increases system complexity but also often lacks interpretability when handling mixed-domain inputs. Building upon the premise that ``Compression is Intelligence,'' this paper proposes a novel architectural philosophy: Compression is Routing. We trained an 87M-parameter end-to-end Transformer Autoencoder, achieving a 64x sequence length compression (compressing 512 tokens into 8 latent vectors). Experimental results demonstrate that this compressor possesses extreme domain discriminative capability: it achieves a reconstruction accuracy of 99.47% on the in-domain (code) validation set; accuracy drops sharply to 47.76% on a semi-out-of-distribution domain (Wiki text); and further plummets to just 0.57% on a fully out-of-distribution domain (random sequences). This extreme and systematic performance discrepancy establishes the validity of reconstruction error as an Intrinsic Distribution Fingerprint. Based on this, we propose that expert modules can be automatically scheduled using reconstruction residuals directly, without the need for explicit gating networks. This mechanism offers excellent scalability. Furthermore, this architecture provides a new perspective on ``VRAM compression'' for handling ultra-long contexts. This report aims to verify the physical validity of this foundational architecture, offering a new research perspective for the next generation of scalable modular neural networks. --- ### MemoryGraft: Persistent Compromise of LLM Agents via Poisoned Experience Retrieval Authors: Saksham Sahai Srivastava, Haoyu He Published: 2025-12-18 Categories: cs.CR, cs.AI, cs.LG arXiv: https://arxiv.org/abs/2512.16962v1 Large Language Model (LLM) agents increasingly rely on long-term memory and Retrieval-Augmented Generation (RAG) to persist experiences and refine future performance. While this experience learning capability enhances agentic autonomy, it introduces a critical, unexplored attack surface, i.e., the trust boundary between an agent's reasoning core and its own past. In this paper, we introduce MemoryGraft. It is a novel indirect injection attack that compromises agent behavior not through immediate jailbreaks, but by implanting malicious successful experiences into the agent's long-term memory. Unlike traditional prompt injections that are transient, or standard RAG poisoning that targets factual knowledge, MemoryGraft exploits the agent's semantic imitation heuristic which is the tendency to replicate patterns from retrieved successful tasks. We demonstrate that an attacker who can supply benign ingestion-level artifacts that the agent reads during execution can induce it to construct a poisoned RAG store where a small set of malicious procedure templates is persisted alongside benign experiences. When the agent later encounters semantically similar tasks, union retrieval over lexical and embedding similarity reliably surfaces these grafted memories, and the agent adopts the embedded unsafe patterns, leading to persistent behavioral drift across sessions. We validate MemoryGraft on MetaGPT's DataInterpreter agent with GPT-4o and find that a small number of poisoned records can account for a large fraction of retrieved experiences on benign workloads, turning experience-based self-improvement into a vector for stealthy and durable compromise. To facilitate reproducibility and future research, our code and evaluation data are available at https://github.com/Jacobhhy/Agent-Memory-Poisoning. --- ### QSMOTE-PGM/kPGM: QSMOTE Based PGM and kPGM for Imbalanced Dataset Classification Authors: Bikash K. Behera, Giuseppe Sergioli, Roberto Giuntini Published: 2025-12-18 Categories: cs.LG, quant-ph arXiv: https://arxiv.org/abs/2512.16960v2 Quantum-inspired machine learning (QiML) employs mathematical principles from quantum theory, such as Hilbert-space representations and quantum state discrimination, to enhance classical learning algorithms. In this work, we investigate the integration of Quantum Synthetic Minority Oversampling Technique (QSMOTE) variants with two quantum-inspired classifiers: the Pretty Good Measurement (PGM) classifier and the kernelized Pretty Good Measurement (KPGM) classifier. We propose and analyze three QSMOTE variants, namely KNN-based, Fidelity-based, and Margin-based QSMOTE, designed to improve minority-class representation in imbalanced datasets through quantum-inspired similarity and sampling mechanisms. A unified theoretical and empirical comparison of PGM and KPGM is presented under amplitude and stereo encoding strategies with multiple quantum copies. Experimental evaluations on the Telco Customer Churn dataset demonstrate that the proposed quantum-inspired approaches consistently outperform a classical Random Forest baseline, particularly in terms of recall and balanced F1-score. Among all configurations, PGM with stereo encoding and n_{copies}=2 achieves the best performance with an accuracy of 0.8512 and an F1-score of 0.8234, while KPGM exhibits competitive and more stable behavior across different QSMOTE variants, reaching accuracies of 0.8511 under stereo encoding and 0.8483 under amplitude encoding. The results further show that increasing the number of quantum copies systematically improves classification performance, especially for minority-class detection. This work highlights the effectiveness of combining quantum-inspired oversampling and classification strategies for imbalanced learning, while providing practical insights into the complementary strengths of measurement-based and kernel-based quantum-inspired machine learning frameworks. --- ### SpIDER: Spatially Informed Dense Embedding Retrieval for Software Issue Localization Authors: Shravan Chaudhari, Rahul Thomas Jacob, Jiajun Cao, Shihab Rashid, Mononito Goswami, Christian Bock Published: 2025-12-18 Categories: cs.SE, cs.LG arXiv: https://arxiv.org/abs/2512.16956v3 Retrieving code functions, classes or files relevant to a user query, bug report or feature request from large codebases is a fundamental challenge for Large Language Model (LLM)-based coding agents. Agentic approaches typically employ sparse methods like BM25 or dense embedding strategies to identify semantically relevant units. While dense embedding approaches can outperform BM25 by large margins, both ignore the graph-structured characteristics of the codebase. To address this, we propose SpIDER (Spatially Informed Dense Embedding Retrieval), a dense retrieval approach that integrates LLM-based reasoning with graph-based exploration of the codebase. We further introduce SpIDER-Bench, a graph-structured benchmark curated from SWEPolyBench, SWEBench-Verified and Multi-SWE-bench, spanning Python, Java, JavaScript and TypeScript repositories. SpIDER's graph-based candidate expansion attaches a structural reason for inclusion to each surfaced function (its seed and the edge type linking them), making the candidate set auditable while keeping the retrieval budget fixed. The graph is built from per-repository syntax trees, so it can be constructed on-demand at the start of a developer session rather than precomputed offline. Empirical results show that SpIDER consistently improves dense retrieval Recall@20 across every language and benchmark in SpIDER-Bench: by at least 13% relative (+0.05 to +0.12 absolute) along containment edges, rising to at least 27% relative (+0.11 to +0.19 absolute) once call edges are explored. --- ### The World is Your Canvas: Painting Promptable Events with Reference Images, Trajectories, and Text Authors: Hanlin Wang, Hao Ouyang, Qiuyu Wang, Yue Yu, Yihao Meng, Wen Wang, Ka Leong Cheng, Shuailei Ma, Qingyan Bai, Yixuan Li, Cheng Chen, Yanhong Zeng et al. Published: 2025-12-18 Categories: cs.CV arXiv: https://arxiv.org/abs/2512.16924v1 We present WorldCanvas, a framework for promptable world events that enables rich, user-directed simulation by combining text, trajectories, and reference images. Unlike text-only approaches and existing trajectory-controlled image-to-video methods, our multimodal approach combines trajectories -- encoding motion, timing, and visibility -- with natural language for semantic intent and reference images for visual grounding of object identity, enabling the generation of coherent, controllable events that include multi-agent interactions, object entry/exit, reference-guided appearance and counterintuitive events. The resulting videos demonstrate not only temporal coherence but also emergent consistency, preserving object identity and scene despite temporary disappearance. By supporting expressive world events generation, WorldCanvas advances world models from passive predictors to interactive, user-shaped simulators. Our project page is available at: https://worldcanvas.github.io/. --- ### Generative Refocusing: Flexible Defocus Control from a Single Image Authors: Chun-Wei Tuan Mu, Cheng-De Fan, Jia-Bin Huang, Yu-Lun Liu Published: 2025-12-18 Categories: cs.CV arXiv: https://arxiv.org/abs/2512.16923v3 Depth-of-field control is essential in photography, but achieving perfect focus often requires multiple attempts or specialized equipment. Single-image refocusing is still difficult. It involves recovering sharp content and creating realistic bokeh. Current methods have significant drawbacks. They require all-in-focus inputs, rely on synthetic data from simulators, and have limited control over the aperture. We introduce Generative Refocusing, a two-step process that uses DeblurNet to recover all-in-focus images from diverse inputs and BokehNet to create controllable bokeh. This method combines synthetic and real bokeh images to achieve precise control while preserving authentic optical characteristics. Our experiments show we achieve top performance in defocus deblurring, bokeh synthesis, and refocusing benchmarks. Additionally, our Generative Refocusing allows custom aperture shapes. Project page: https://generative-refocusing.github.io/ --- ### Generative Refocusing: Flexible Defocus Control from a Single Image Authors: Chun-Wei Tuan Mu, Jia-Bin Huang, Yu-Lun Liu Published: 2025-12-18 Categories: cs.CV arXiv: https://arxiv.org/abs/2512.16923v1 Depth-of-field control is essential in photography, but getting the perfect focus often takes several tries or special equipment. Single-image refocusing is still difficult. It involves recovering sharp content and creating realistic bokeh. Current methods have significant drawbacks. They need all-in-focus inputs, depend on synthetic data from simulators, and have limited control over aperture. We introduce Generative Refocusing, a two-step process that uses DeblurNet to recover all-in-focus images from various inputs and BokehNet for creating controllable bokeh. Our main innovation is semi-supervised training. This method combines synthetic paired data with unpaired real bokeh images, using EXIF metadata to capture real optical characteristics beyond what simulators can provide. Our experiments show we achieve top performance in defocus deblurring, bokeh synthesis, and refocusing benchmarks. Additionally, our Generative Refocusing allows text-guided adjustments and custom aperture shapes. --- ### Next-Embedding Prediction Makes Strong Vision Learners Authors: Sihan Xu, Ziqiao Ma, Wenhao Chai, Xuweiyi Chen, Weiyang Jin, Joyce Chai, Saining Xie, Stella X. Yu Published: 2025-12-18 Categories: cs.CV arXiv: https://arxiv.org/abs/2512.16922v2 Inspired by the success of generative pretraining in natural language, we ask whether the same principles can yield strong self-supervised visual learners. Instead of training models to output features for downstream use, we train them to generate embeddings to perform predictive tasks directly. This work explores such a shift from learning representations to learning models. Specifically, models learn to predict future patch embeddings conditioned on past ones, using causal masking and stop gradient, which we refer to as Next-Embedding Predictive Autoregression (NEPA). We demonstrate that a simple Transformer pretrained on ImageNet-1k with next embedding prediction as its sole learning objective is effective - no pixel reconstruction, discrete tokens, contrastive loss, or task-specific heads. This formulation retains architectural simplicity and scalability, without requiring additional design complexity. NEPA achieves strong results across tasks, attaining 83.8% and 85.3% top-1 accuracy on ImageNet-1K with ViT-B and ViT-L backbones after fine-tuning, and transferring effectively to semantic segmentation on ADE20K. We believe generative pretraining from embeddings provides a simple, scalable, and potentially modality-agnostic alternative to visual self-supervised learning. --- ### Next-Embedding Prediction Makes Strong Vision Learners Authors: Sihan Xu, Ziqiao Ma, Wenhao Chai, Xuweiyi Chen, Weiyang Jin, Joyce Chai, Saining Xie, Stella X. Yu Published: 2025-12-18 Categories: cs.CV arXiv: https://arxiv.org/abs/2512.16922v1 Inspired by the success of generative pretraining in natural language, we ask whether the same principles can yield strong self-supervised visual learners. Instead of training models to output features for downstream use, we train them to generate embeddings to perform predictive tasks directly. This work explores such a shift from learning representations to learning models. Specifically, models learn to predict future patch embeddings conditioned on past ones, using causal masking and stop gradient, which we refer to as Next-Embedding Predictive Autoregression (NEPA). We demonstrate that a simple Transformer pretrained on ImageNet-1k with next embedding prediction as its sole learning objective is effective - no pixel reconstruction, discrete tokens, contrastive loss, or task-specific heads. This formulation retains architectural simplicity and scalability, without requiring additional design complexity. NEPA achieves strong results across tasks, attaining 83.8% and 85.3% top-1 accuracy on ImageNet-1K with ViT-B and ViT-L backbones after fine-tuning, and transferring effectively to semantic segmentation on ADE20K. We believe generative pretraining from embeddings provides a simple, scalable, and potentially modality-agnostic alternative to visual self-supervised learning. --- ### Differences That Matter: Auditing Models for Capability Gap Discovery and Rectification Authors: Qihao Liu, Chengzhi Mao, Yaojie Liu, Alan Yuille, Wen-Sheng Chu Published: 2025-12-18 Categories: cs.CV, cs.AI arXiv: https://arxiv.org/abs/2512.16921v1 Conventional evaluation methods for multimodal LLMs (MLLMs) lack interpretability and are often insufficient to fully disclose significant capability gaps across models. To address this, we introduce AuditDM, an automated framework that actively discovers and rectifies MLLM failure modes by auditing their divergence. AuditDM fine-tunes an MLLM as an auditor via reinforcement learning to generate challenging questions and counterfactual images that maximize disagreement among target models. Once trained, the auditor uncovers diverse, interpretable exemplars that reveal model weaknesses and serve as annotation-free data for rectification. When applied to SoTA models like Gemma-3 and PaliGemma-2, AuditDM discovers more than 20 distinct failure types. Fine-tuning on these discoveries consistently improves all models across 16 benchmarks, and enables a 3B model to surpass its 28B counterpart. Our results suggest that as data scaling hits diminishing returns, targeted model auditing offers an effective path to model diagnosis and improvement. --- ### EasyV2V: A High-quality Instruction-based Video Editing Framework Authors: Jinjie Mai, Chaoyang Wang, Guocheng Gordon Qian, Willi Menapace, Sergey Tulyakov, Bernard Ghanem, Peter Wonka, Ashkan Mirzaei Published: 2025-12-18 Categories: cs.CV, cs.AI arXiv: https://arxiv.org/abs/2512.16920v1 While image editing has advanced rapidly, video editing remains less explored, facing challenges in consistency, control, and generalization. We study the design space of data, architecture, and control, and introduce \emph{EasyV2V}, a simple and effective framework for instruction-based video editing. On the data side, we compose existing experts with fast inverses to build diverse video pairs, lift image edit pairs into videos via single-frame supervision and pseudo pairs with shared affine motion, mine dense-captioned clips for video pairs, and add transition supervision to teach how edits unfold. On the model side, we observe that pretrained text-to-video models possess editing capability, motivating a simplified design. Simple sequence concatenation for conditioning with light LoRA fine-tuning suffices to train a strong model. For control, we unify spatiotemporal control via a single mask mechanism and support optional reference images. Overall, EasyV2V works with flexible inputs, e.g., video+text, video+mask+text, video+mask+reference+text, and achieves state-of-the-art video editing results, surpassing concurrent and commercial systems. Project page: https://snap-research.github.io/easyv2v/ --- ### DVGT: Driving Visual Geometry Transformer Authors: Sicheng Zuo, Zixun Xie, Wenzhao Zheng, Shaoqing Xu, Fang Li, Shengyin Jiang, Long Chen, Zhi-Xin Yang, Jiwen Lu Published: 2025-12-18 Categories: cs.CV, cs.AI, cs.RO arXiv: https://arxiv.org/abs/2512.16919v2 Perceiving and reconstructing 3D scene geometry from visual inputs is crucial for autonomous driving. However, there still lacks a driving-targeted dense geometry perception model that can adapt to different scenarios and camera configurations. To bridge this gap, we propose a Driving Visual Geometry Transformer (DVGT), which reconstructs a global dense 3D point map from a sequence of unposed multi-view visual inputs. We first extract visual features for each image using a DINO backbone, and employ alternating intra-view local attention, cross-view spatial attention, and cross-frame temporal attention to infer geometric relations across images. We then use multiple heads to decode a global point map in the ego coordinate of the first frame and the ego poses for each frame. Unlike conventional methods that rely on precise camera parameters, DVGT is free of explicit 3D geometric priors, enabling flexible processing of arbitrary camera configurations. DVGT directly predicts metric-scaled geometry from image sequences, eliminating the need for post-alignment with external sensors. Trained on a large mixture of driving datasets including nuScenes, OpenScene, Waymo, KITTI, and DDAD, DVGT significantly outperforms existing models on various scenarios. Code is available at https://github.com/wzzheng/DVGT. --- ### DVGT: Driving Visual Geometry Transformer Authors: Sicheng Zuo, Zixun Xie, Wenzhao Zheng, Shaoqing Xu, Fang Li, Shengyin Jiang, Long Chen, Zhi-Xin Yang, Jiwen Lu Published: 2025-12-18 Categories: cs.CV, cs.AI, cs.RO arXiv: https://arxiv.org/abs/2512.16919v1 Perceiving and reconstructing 3D scene geometry from visual inputs is crucial for autonomous driving. However, there still lacks a driving-targeted dense geometry perception model that can adapt to different scenarios and camera configurations. To bridge this gap, we propose a Driving Visual Geometry Transformer (DVGT), which reconstructs a global dense 3D point map from a sequence of unposed multi-view visual inputs. We first extract visual features for each image using a DINO backbone, and employ alternating intra-view local attention, cross-view spatial attention, and cross-frame temporal attention to infer geometric relations across images. We then use multiple heads to decode a global point map in the ego coordinate of the first frame and the ego poses for each frame. Unlike conventional methods that rely on precise camera parameters, DVGT is free of explicit 3D geometric priors, enabling flexible processing of arbitrary camera configurations. DVGT directly predicts metric-scaled geometry from image sequences, eliminating the need for post-alignment with external sensors. Trained on a large mixture of driving datasets including nuScenes, OpenScene, Waymo, KITTI, and DDAD, DVGT significantly outperforms existing models on various scenarios. Code is available at https://github.com/wzzheng/DVGT. --- ### AdaTooler-V: Adaptive Tool-Use for Images and Videos Authors: Chaoyang Wang, Kaituo Feng, Dongyang Chen, Zhongyu Wang, Zhixun Li, Sicheng Gao, Meng Meng, Xu Zhou, Manyuan Zhang, Yuzhang Shang, Xiangyu Yue Published: 2025-12-18 Categories: cs.CV arXiv: https://arxiv.org/abs/2512.16918v3 Recent advances have shown that multimodal large language models (MLLMs) benefit from multimodal interleaved chain-of-thought (CoT) with vision tool interactions. However, existing open-source models often exhibit blind tool-use reasoning patterns, invoking vision tools even when they are unnecessary, which significantly increases inference overhead and degrades model performance. To this end, we propose AdaTooler-V, an MLLM that performs adaptive tool-use by determining whether a visual problem truly requires tools. First, we introduce AT-GRPO, a reinforcement learning algorithm that adaptively adjusts reward scales based on the Tool Benefit Score of each sample, encouraging the model to invoke tools only when they provide genuine improvements. Moreover, we construct two datasets to support training: AdaTooler-V-CoT-100k for SFT cold start and AdaTooler-V-300k for RL with verifiable rewards across single-image, multi-image, and video data. Experiments across twelve benchmarks demonstrate the strong reasoning capability of AdaTooler-V, outperforming existing methods in diverse visual reasoning tasks. Notably, AdaTooler-V-7B achieves an accuracy of 89.8\% on the high-resolution benchmark V*, surpassing the commercial proprietary model GPT-4o and Gemini 1.5 Pro. All code, models, and data are released. --- ### AdaTooler-V: Adaptive Tool-Use for Images and Videos Authors: Chaoyang Wang, Kaituo Feng, Dongyang Chen, Zhongyu Wang, Zhixun Li, Sicheng Gao, Meng Meng, Xu Zhou, Manyuan Zhang, Yuzhang Shang, Xiangyu Yue Published: 2025-12-18 Categories: cs.CV arXiv: https://arxiv.org/abs/2512.16918v1 Recent advances have shown that multimodal large language models (MLLMs) benefit from multimodal interleaved chain-of-thought (CoT) with vision tool interactions. However, existing open-source models often exhibit blind tool-use reasoning patterns, invoking vision tools even when they are unnecessary, which significantly increases inference overhead and degrades model performance. To this end, we propose AdaTooler-V, an MLLM that performs adaptive tool-use by determining whether a visual problem truly requires tools. First, we introduce AT-GRPO, a reinforcement learning algorithm that adaptively adjusts reward scales based on the Tool Benefit Score of each sample, encouraging the model to invoke tools only when they provide genuine improvements. Moreover, we construct two datasets to support training: AdaTooler-V-CoT-100k for SFT cold start and AdaTooler-V-300k for RL with verifiable rewards across single-image, multi-image, and video data. Experiments across twelve benchmarks demonstrate the strong reasoning capability of AdaTooler-V, outperforming existing methods in diverse visual reasoning tasks. Notably, AdaTooler-V-7B achieves an accuracy of 89.8\% on the high-resolution benchmark V*, surpassing the commercial proprietary model GPT-4o and Gemini 1.5 Pro. All code, models, and data are released. --- ### Generative Adversarial Reasoner: Enhancing LLM Reasoning with Adversarial Reinforcement Learning Authors: Qihao Liu, Luoxin Ye, Wufei Ma, Yu-Cheng Chou, Alan Yuille Published: 2025-12-18 Categories: cs.AI, cs.CL, cs.LG arXiv: https://arxiv.org/abs/2512.16917v3 Large language models (LLMs) with explicit reasoning capabilities excel at mathematical reasoning yet still commit process errors, such as incorrect calculations, brittle logic, and superficially plausible but invalid steps. In this paper, we introduce Generative Adversarial Reasoner, an on-policy joint training framework designed to enhance reasoning by co-evolving an LLM reasoner and an LLM-based discriminator through adversarial reinforcement learning. A compute-efficient review schedule partitions each reasoning chain into logically complete slices of comparable length, and the discriminator evaluates each slice's soundness with concise, structured justifications. Learning couples complementary signals: the LLM reasoner is rewarded for logically consistent steps that yield correct answers, while the discriminator earns rewards for correctly detecting errors or distinguishing traces in the reasoning process. This produces dense, well-calibrated, on-policy step-level rewards that supplement sparse exact-match signals, improving credit assignment, increasing sample efficiency, and enhancing overall reasoning quality of LLMs. Across various mathematical benchmarks, the method delivers consistent gains over strong baselines with standard RL post-training. Specifically, on AIME24, we improve DeepSeek-R1-Distill-Qwen-7B from 54.0 to 61.3 (+7.3) and DeepSeek-R1-Distill-Llama-8B from 43.7 to 53.7 (+10.0). The modular discriminator also enables flexible reward shaping for objectives such as teacher distillation, preference alignment, and mathematical proof-based reasoning. --- ### Generative Adversarial Reasoner: Enhancing LLM Reasoning with Adversarial Reinforcement Learning Authors: Qihao Liu, Luoxin Ye, Wufei Ma, Yu-Cheng Chou, Alan Yuille Published: 2025-12-18 Categories: cs.AI, cs.CL, cs.LG arXiv: https://arxiv.org/abs/2512.16917v1 Large language models (LLMs) with explicit reasoning capabilities excel at mathematical reasoning yet still commit process errors, such as incorrect calculations, brittle logic, and superficially plausible but invalid steps. In this paper, we introduce Generative Adversarial Reasoner, an on-policy joint training framework designed to enhance reasoning by co-evolving an LLM reasoner and an LLM-based discriminator through adversarial reinforcement learning. A compute-efficient review schedule partitions each reasoning chain into logically complete slices of comparable length, and the discriminator evaluates each slice's soundness with concise, structured justifications. Learning couples complementary signals: the LLM reasoner is rewarded for logically consistent steps that yield correct answers, while the discriminator earns rewards for correctly detecting errors or distinguishing traces in the reasoning process. This produces dense, well-calibrated, on-policy step-level rewards that supplement sparse exact-match signals, improving credit assignment, increasing sample efficiency, and enhancing overall reasoning quality of LLMs. Across various mathematical benchmarks, the method delivers consistent gains over strong baselines with standard RL post-training. Specifically, on AIME24, we improve DeepSeek-R1-Distill-Qwen-7B from 54.0 to 61.3 (+7.3) and DeepSeek-R1-Distill-Llama-8B from 43.7 to 53.7 (+10.0). The modular discriminator also enables flexible reward shaping for objectives such as teacher distillation, preference alignment, and mathematical proof-based reasoning. --- ### StereoPilot: Learning Unified and Efficient Stereo Conversion via Generative Priors Authors: Guibao Shen, Yihua Du, Wenhang Ge, Jing He, Chirui Chang, Donghao Zhou, Zhen Yang, Luozhou Wang, Xin Tao, Ying-Cong Chen Published: 2025-12-18 Categories: cs.CV arXiv: https://arxiv.org/abs/2512.16915v1 The rapid growth of stereoscopic displays, including VR headsets and 3D cinemas, has led to increasing demand for high-quality stereo video content. However, producing 3D videos remains costly and complex, while automatic Monocular-to-Stereo conversion is hindered by the limitations of the multi-stage ``Depth-Warp-Inpaint'' (DWI) pipeline. This paradigm suffers from error propagation, depth ambiguity, and format inconsistency between parallel and converged stereo configurations. To address these challenges, we introduce UniStereo, the first large-scale unified dataset for stereo video conversion, covering both stereo formats to enable fair benchmarking and robust model training. Building upon this dataset, we propose StereoPilot, an efficient feed-forward model that directly synthesizes the target view without relying on explicit depth maps or iterative diffusion sampling. Equipped with a learnable domain switcher and a cycle consistency loss, StereoPilot adapts seamlessly to different stereo formats and achieves improved consistency. Extensive experiments demonstrate that StereoPilot significantly outperforms state-of-the-art methods in both visual fidelity and computational efficiency. Project page: https://hit-perfect.github.io/StereoPilot/. --- ### Constructive Circuit Amplification: Improving Math Reasoning in LLMs via Targeted Sub-Network Updates Authors: Nikhil Prakash, Donghao Ren, Dominik Moritz, Yannick Assogba Published: 2025-12-18 Categories: cs.CL arXiv: https://arxiv.org/abs/2512.16914v1 Prior studies investigating the internal workings of LLMs have uncovered sparse subnetworks, often referred to as circuits, that are responsible for performing specific tasks. Additionally, it has been shown that model performance improvement through fine-tuning often results from the strengthening of existing circuits in the model. Taken together, these findings suggest the possibility of intervening directly on such circuits to make precise, task-targeted updates. Motivated by these findings, we propose a novel method called Constructive Circuit Amplification which identifies pivotal tokens from model reasoning traces as well as model components responsible for the desired task, and updates only those components. Applied to mathematical reasoning, it improves accuracy by up to +11.4% across multiple models while modifying as little as 1.59% of model components, with minimal impact on other abilities as measured by MMLU, TriviaQA, and TruthfulQA. These results demonstrate that targeted capabilities can be reliably enhanced by selectively updating a sparse set of model components. --- ### Depth Any Panoramas: A Foundation Model for Panoramic Depth Estimation Authors: Xin Lin, Meixi Song, Dizhe Zhang, Wenxuan Lu, Haodong Li, Bo Du, Ming-Hsuan Yang, Truong Nguyen, Lu Qi Published: 2025-12-18 Categories: cs.CV arXiv: https://arxiv.org/abs/2512.16913v1 In this work, we present a panoramic metric depth foundation model that generalizes across diverse scene distances. We explore a data-in-the-loop paradigm from the view of both data construction and framework design. We collect a large-scale dataset by combining public datasets, high-quality synthetic data from our UE5 simulator and text-to-image models, and real panoramic images from the web. To reduce domain gaps between indoor/outdoor and synthetic/real data, we introduce a three-stage pseudo-label curation pipeline to generate reliable ground truth for unlabeled images. For the model, we adopt DINOv3-Large as the backbone for its strong pre-trained generalization, and introduce a plug-and-play range mask head, sharpness-centric optimization, and geometry-centric optimization to improve robustness to varying distances and enforce geometric consistency across views. Experiments on multiple benchmarks (e.g., Stanford2D3D, Matterport3D, and Deep360) demonstrate strong performance and zero-shot generalization, with particularly robust and stable metric predictions in diverse real-world scenes. The project page can be found at: \href{https://insta360-research-team.github.io/DAP_website/} {https://insta360-research-team.github.io/DAP\_website/} --- ### Exploration vs Exploitation: Rethinking RLVR through Clipping, Entropy, and Spurious Reward Authors: Peter Chen, Xiaopeng Li, Ziniu Li, Wotao Yin, Xi Chen, Tianyi Lin Published: 2025-12-18 Categories: cs.LG, cs.AI, cs.CL arXiv: https://arxiv.org/abs/2512.16912v3 This paper examines the exploration-exploitation trade-off in reinforcement learning with verifiable rewards (RLVR), a framework for improving the reasoning of Large Language Models (LLMs). Recent studies suggest that RLVR can elicit strong mathematical reasoning in LLMs through two seemingly paradoxical mechanisms: spurious rewards, which suppress exploitation by rewarding outcomes unrelated to the ground truth, and entropy minimization, which suppresses exploration by pushing the model toward more confident and deterministic outputs, highlighting a puzzling dynamic: both discouraging exploitation and discouraging exploration improve reasoning performance, yet the underlying principles that reconcile these effects remain poorly understood. We focus on two fundamental questions: (i) how policy entropy relates to performance, and (ii) whether spurious rewards yield gains, potentially through the interplay of clipping bias and model contamination. Our results show that clipping bias under spurious rewards reduces policy entropy, leading to more confident and deterministic outputs, while entropy minimization alone is insufficient for improvement. We further propose a reward-misalignment model explaining why spurious rewards can enhance performance beyond contaminated settings. Our findings clarify the mechanisms behind spurious-reward benefits and provide principles for more effective RLVR training. --- ### Exploration v.s. Exploitation: Rethinking RLVR through Clipping, Entropy, and Spurious Reward Authors: Peter Chen, Xiaopeng Li, Ziniu Li, Wotao Yin, Xi Chen, Tianyi Lin Published: 2025-12-18 Categories: cs.LG, cs.AI, cs.CL arXiv: https://arxiv.org/abs/2512.16912v1 This paper examines the exploration-exploitation trade-off in reinforcement learning with verifiable rewards (RLVR), a framework for improving the reasoning of Large Language Models (LLMs). Recent studies suggest that RLVR can elicit strong mathematical reasoning in LLMs through two seemingly paradoxical mechanisms: spurious rewards, which suppress exploitation by rewarding outcomes unrelated to the ground truth, and entropy minimization, which suppresses exploration by pushing the model toward more confident and deterministic outputs, highlighting a puzzling dynamic: both discouraging exploitation and discouraging exploration improve reasoning performance, yet the underlying principles that reconcile these effects remain poorly understood. We focus on two fundamental questions: (i) how policy entropy relates to performance, and (ii) whether spurious rewards yield gains, potentially through the interplay of clipping bias and model contamination. Our results show that clipping bias under spurious rewards reduces policy entropy, leading to more confident and deterministic outputs, while entropy minimization alone is insufficient for improvement. We further propose a reward-misalignment model explaining why spurious rewards can enhance performance beyond contaminated settings. Our findings clarify the mechanisms behind spurious-reward benefits and provide principles for more effective RLVR training. --- ### Posterior Behavioral Cloning: Pretraining BC Policies for Efficient RL Finetuning Authors: Andrew Wagenmaker, Perry Dong, Raymond Tsao, Chelsea Finn, Sergey Levine Published: 2025-12-18 Categories: cs.LG, cs.AI, cs.RO arXiv: https://arxiv.org/abs/2512.16911v1 Standard practice across domains from robotics to language is to first pretrain a policy on a large-scale demonstration dataset, and then finetune this policy, typically with reinforcement learning (RL), in order to improve performance on deployment domains. This finetuning step has proved critical in achieving human or super-human performance, yet while much attention has been given to developing more effective finetuning algorithms, little attention has been given to ensuring the pretrained policy is an effective initialization for RL finetuning. In this work we seek to understand how the pretrained policy affects finetuning performance, and how to pretrain policies in order to ensure they are effective initializations for finetuning. We first show theoretically that standard behavioral cloning (BC) -- which trains a policy to directly match the actions played by the demonstrator -- can fail to ensure coverage over the demonstrator's actions, a minimal condition necessary for effective RL finetuning. We then show that if, instead of exactly fitting the observed demonstrations, we train a policy to model the posterior distribution of the demonstrator's behavior given the demonstration dataset, we do obtain a policy that ensures coverage over the demonstrator's actions, enabling more effective finetuning. Furthermore, this policy -- which we refer to as the posterior behavioral cloning (PostBC) policy -- achieves this while ensuring pretrained performance is no worse than that of the BC policy. We then show that PostBC is practically implementable with modern generative models in robotic control domains -- relying only on standard supervised learning -- and leads to significantly improved RL finetuning performance on both realistic robotic control benchmarks and real-world robotic manipulation tasks, as compared to standard behavioral cloning. --- ### SFTok: Bridging the Performance Gap in Discrete Tokenizers Authors: Qihang Rao, Borui Zhang, Wenzhao Zheng, Jie Zhou, Jiwen Lu Published: 2025-12-18 Categories: cs.CV, cs.LG arXiv: https://arxiv.org/abs/2512.16910v1 Recent advances in multimodal models highlight the pivotal role of image tokenization in high-resolution image generation. By compressing images into compact latent representations, tokenizers enable generative models to operate in lower-dimensional spaces, thereby improving computational efficiency and reducing complexity. Discrete tokenizers naturally align with the autoregressive paradigm but still lag behind continuous ones, limiting their adoption in multimodal systems. To address this, we propose \textbf{SFTok}, a discrete tokenizer that incorporates a multi-step iterative mechanism for precise reconstruction. By integrating \textbf{self-forcing guided visual reconstruction} and \textbf{debias-and-fitting training strategy}, SFTok resolves the training-inference inconsistency in multi-step process, significantly enhancing image reconstruction quality. At a high compression rate of only 64 tokens per image, SFTok achieves state-of-the-art reconstruction quality on ImageNet (rFID = 1.21) and demonstrates exceptional performance in class-to-image generation tasks (gFID = 2.29). --- ### MomaGraph: State-Aware Unified Scene Graphs with Vision-Language Model for Embodied Task Planning Authors: Yuanchen Ju, Yongyuan Liang, Yen-Jen Wang, Nandiraju Gireesh, Yuanliang Ju, Seungjae Lee, Qiao Gu, Elvis Hsieh, Furong Huang, Koushil Sreenath Published: 2025-12-18 Categories: cs.CV, cs.RO arXiv: https://arxiv.org/abs/2512.16909v2 Mobile manipulators in households must both navigate and manipulate. This requires a compact, semantically rich scene representation that captures where objects are, how they function, and which parts are actionable. Scene graphs are a natural choice, yet prior work often separates spatial and functional relations, treats scenes as static snapshots without object states or temporal updates, and overlooks information most relevant for accomplishing the current task. To address these limitations, we introduce MomaGraph, a unified scene representation for embodied agents that integrates spatial-functional relationships and part-level interactive elements. However, advancing such a representation requires both suitable data and rigorous evaluation, which have been largely missing. We thus contribute MomaGraph-Scenes, the first large-scale dataset of richly annotated, task-driven scene graphs in household environments, along with MomaGraph-Bench, a systematic evaluation suite spanning six reasoning capabilities from high-level planning to fine-grained scene understanding. Built upon this foundation, we further develop MomaGraph-R1, a 7B vision-language model trained with reinforcement learning on MomaGraph-Scenes. MomaGraph-R1 predicts task-oriented scene graphs and serves as a zero-shot task planner under a Graph-then-Plan framework. Extensive experiments demonstrate that our model achieves state-of-the-art results among open-source models, reaching 71.6% accuracy on the benchmark (+11.4% over the best baseline), while generalizing across public benchmarks and transferring effectively to real-robot experiments. --- ### MomaGraph: State-Aware Unified Scene Graphs with Vision-Language Model for Embodied Task Planning Authors: Yuanchen Ju, Yongyuan Liang, Yen-Jen Wang, Nandiraju Gireesh, Yuanliang Ju, Seungjae Lee, Qiao Gu, Elvis Hsieh, Furong Huang, Koushil Sreenath Published: 2025-12-18 Categories: cs.CV, cs.RO arXiv: https://arxiv.org/abs/2512.16909v1 Mobile manipulators in households must both navigate and manipulate. This requires a compact, semantically rich scene representation that captures where objects are, how they function, and which parts are actionable. Scene graphs are a natural choice, yet prior work often separates spatial and functional relations, treats scenes as static snapshots without object states or temporal updates, and overlooks information most relevant for accomplishing the current task. To address these limitations, we introduce MomaGraph, a unified scene representation for embodied agents that integrates spatial-functional relationships and part-level interactive elements. However, advancing such a representation requires both suitable data and rigorous evaluation, which have been largely missing. We thus contribute MomaGraph-Scenes, the first large-scale dataset of richly annotated, task-driven scene graphs in household environments, along with MomaGraph-Bench, a systematic evaluation suite spanning six reasoning capabilities from high-level planning to fine-grained scene understanding. Built upon this foundation, we further develop MomaGraph-R1, a 7B vision-language model trained with reinforcement learning on MomaGraph-Scenes. MomaGraph-R1 predicts task-oriented scene graphs and serves as a zero-shot task planner under a Graph-then-Plan framework. Extensive experiments demonstrate that our model achieves state-of-the-art results among open-source models, reaching 71.6% accuracy on the benchmark (+11.4% over the best baseline), while generalizing across public benchmarks and transferring effectively to real-robot experiments. --- ### SceneDiff: A Benchmark and Method for Multiview Object Change Detection Authors: Yuqun Wu, Chih-hao Lin, Henry Che, Aditi Tiwari, Chuhang Zou, Shenlong Wang, Derek Hoiem Published: 2025-12-18 Categories: cs.CV arXiv: https://arxiv.org/abs/2512.16908v2 We investigate the problem of identifying objects that have been added, removed, or moved between a pair of captures (images or videos) of the same scene at different times. Accurately identifying verifiable changes is extremely challenging -- some objects may appear to be missing because they are occluded or out of frame, while others may appear different due to large viewpoint changes. To study this problem, we introduce the SceneDiff Benchmark, the first multiview change detection dataset for scenes captured along different camera trajectories, comprising 350 diverse video pairs with dense object instance-level annotations. We also introduce the SceneDiff algorithm, a training-free approach that solves for image poses, segments images into objects, and compares them using semantic and geometric features. By building on pretrained models, SceneDiff generalizes across domains without retraining and naturally improves as the underlying models advance. Experiments on multiview and two-view benchmarks demonstrate that our method outperforms existing approaches by large margins (53.0\% and 30.6\% relative AP improvements). Project page: https://yuqunw.github.io/SceneDiff --- ### SceneDiff: A Benchmark and Method for Multiview Object Change Detection Authors: Yuqun Wu, Chih-hao Lin, Henry Che, Aditi Tiwari, Chuhang Zou, Shenlong Wang, Derek Hoiem Published: 2025-12-18 Categories: cs.CV arXiv: https://arxiv.org/abs/2512.16908v1 We investigate the problem of identifying objects that have been added, removed, or moved between a pair of captures (images or videos) of the same scene at different times. Detecting such changes is important for many applications, such as robotic tidying or construction progress and safety monitoring. A major challenge is that varying viewpoints can cause objects to falsely appear changed. We introduce SceneDiff Benchmark, the first multiview change detection benchmark with object instance annotations, comprising 350 diverse video pairs with thousands of changed objects. We also introduce the SceneDiff method, a new training-free approach for multiview object change detection that leverages pretrained 3D, segmentation, and image encoding models to robustly predict across multiple benchmarks. Our method aligns the captures in 3D, extracts object regions, and compares spatial and semantic region features to detect changes. Experiments on multi-view and two-view benchmarks demonstrate that our method outperforms existing approaches by large margins (94% and 37.4% relative AP improvements). The benchmark and code will be publicly released. --- ### Flowing from Reasoning to Motion: Learning 3D Hand Trajectory Prediction from Egocentric Human Interaction Videos Authors: Mingfei Chen, Yifan Wang, Zhengqin Li, Homanga Bharadhwaj, Yujin Chen, Chuan Qin, Ziyi Kou, Yuan Tian, Eric Whitmire, Rajinder Sodhi, Hrvoje Benko, Eli Shlizerman et al. Published: 2025-12-18 Categories: cs.CV, cs.AI, cs.RO arXiv: https://arxiv.org/abs/2512.16907v2 Prior works on 3D hand trajectory prediction are constrained by datasets that decouple motion from semantic supervision and by models that weakly link reasoning and action. To address these, we first present the EgoMAN dataset, a large-scale egocentric dataset for interaction stage-aware 3D hand trajectory prediction with 219K 6DoF trajectories and 3M structured QA pairs for semantic, spatial, and motion reasoning. We then introduce the EgoMAN model, a reasoning-to-motion framework that links vision-language reasoning and motion generation via a trajectory-token interface. Trained progressively to align reasoning with motion dynamics, our approach yields accurate and stage-aware trajectories with generalization across real-world scenes. --- ### Flowing from Reasoning to Motion: Learning 3D Hand Trajectory Prediction from Egocentric Human Interaction Videos Authors: Mingfei Chen, Yifan Wang, Zhengqin Li, Homanga Bharadhwaj, Yujin Chen, Chuan Qin, Ziyi Kou, Yuan Tian, Eric Whitmire, Rajinder Sodhi, Hrvoje Benko, Eli Shlizerman et al. Published: 2025-12-18 Categories: cs.CV, cs.AI, cs.RO arXiv: https://arxiv.org/abs/2512.16907v1 Prior works on 3D hand trajectory prediction are constrained by datasets that decouple motion from semantic supervision and by models that weakly link reasoning and action. To address these, we first present the EgoMAN dataset, a large-scale egocentric dataset for interaction stage-aware 3D hand trajectory prediction with 219K 6DoF trajectories and 3M structured QA pairs for semantic, spatial, and motion reasoning. We then introduce the EgoMAN model, a reasoning-to-motion framework that links vision-language reasoning and motion generation via a trajectory-token interface. Trained progressively to align reasoning with motion dynamics, our approach yields accurate and stage-aware trajectories with generalization across real-world scenes. --- ### VIVA: VLM-Guided Instruction-Based Video Editing with Reward Optimization Authors: Xiaoyan Cong, Haotian Yang, Angtian Wang, Yizhi Wang, Yiding Yang, Canyu Zhang, Chongyang Ma Published: 2025-12-18 Categories: cs.CV arXiv: https://arxiv.org/abs/2512.16906v1 Instruction-based video editing aims to modify an input video according to a natural-language instruction while preserving content fidelity and temporal coherence. However, existing diffusion-based approaches are often trained on paired data of simple editing operations, which fundamentally limits their ability to generalize to diverse and complex, real-world instructions. To address this generalization gap, we propose VIVA, a scalable framework for instruction-based video editing that leverages VLM-guided encoding and reward optimization. First, we introduce a VLM-based instructor that encodes the textual instruction, the first frame of the source video, and an optional reference image into visually-grounded instruction representations, providing fine-grained spatial and semantic context for the diffusion transformer backbone. Second, we propose a post-training stage, Edit-GRPO, which adapts Group Relative Policy Optimization to the domain of video editing, directly optimizing the model for instruction-faithful, content-preserving, and aesthetically pleasing edits using relative rewards. Furthermore, we propose a data construction pipeline designed to synthetically generate diverse, high-fidelity paired video-instruction data of basic editing operations. Extensive experiments show that VIVA achieves superior instruction following, generalization, and editing quality over state-of-the-art methods. Website: https://viva-paper.github.io --- ### Alchemist: Unlocking Efficiency in Text-to-Image Model Training via Meta-Gradient Data Selection Authors: Kaixin Ding, Yang Zhou, Xi Chen, Miao Yang, Jiarong Ou, Rui Chen, Xin Tao, Hengshuang Zhao Published: 2025-12-18 Categories: cs.CV arXiv: https://arxiv.org/abs/2512.16905v1 Recent advances in Text-to-Image (T2I) generative models, such as Imagen, Stable Diffusion, and FLUX, have led to remarkable improvements in visual quality. However, their performance is fundamentally limited by the quality of training data. Web-crawled and synthetic image datasets often contain low-quality or redundant samples, which lead to degraded visual fidelity, unstable training, and inefficient computation. Hence, effective data selection is crucial for improving data efficiency. Existing approaches rely on costly manual curation or heuristic scoring based on single-dimensional features in Text-to-Image data filtering. Although meta-learning based method has been explored in LLM, there is no adaptation for image modalities. To this end, we propose **Alchemist**, a meta-gradient-based framework to select a suitable subset from large-scale text-image data pairs. Our approach automatically learns to assess the influence of each sample by iteratively optimizing the model from a data-centric perspective. Alchemist consists of two key stages: data rating and data pruning. We train a lightweight rater to estimate each sample's influence based on gradient information, enhanced with multi-granularity perception. We then use the Shift-Gsampling strategy to select informative subsets for efficient model training. Alchemist is the first automatic, scalable, meta-gradient-based data selection framework for Text-to-Image model training. Experiments on both synthetic and web-crawled datasets demonstrate that Alchemist consistently improves visual quality and downstream performance. Training on an Alchemist-selected 50% of the data can outperform training on the full dataset. --- ### How Good is Post-Hoc Watermarking With Language Model Rephrasing? Authors: Pierre Fernandez, Tom Sander, Hady Elsahar, Hongyan Chang, Tomáš Souček, Valeriu Lacatusu, Tuan Tran, Sylvestre-Alvise Rebuffi, Alexandre Mourachko Published: 2025-12-18 Categories: cs.CR, cs.CL arXiv: https://arxiv.org/abs/2512.16904v2 Generation-time text watermarking embeds statistical signals into text for traceability of AI-generated content. We explore *post-hoc watermarking* where an LLM rewrites existing text while applying generation-time watermarking, to protect copyrighted documents, or detect their use in training or RAG via watermark radioactivity. Unlike generation-time approaches, which is constrained by how LLMs are served, this setting offers additional degrees of freedom for both generation and detection. We investigate how allocating compute (through larger rephrasing models, beam search, multi-candidate generation, or entropy filtering at detection) affects the quality-detectability trade-off. Our strategies achieve strong detectability and semantic fidelity on open-ended text such as books. Among our findings, the simple Gumbel-max scheme surprisingly outperforms more recent alternatives under nucleus sampling, and most methods benefit significantly from beam search. However, most approaches struggle when watermarking verifiable text such as code, where we counterintuitively find that smaller models outperform larger ones. This study reveals both the potential and limitations of post-hoc watermarking, laying groundwork for practical applications and future research. --- ### How Good is Post-Hoc Watermarking With Language Model Rephrasing? Authors: Pierre Fernandez, Tom Sander, Hady Elsahar, Hongyan Chang, Tomáš Souček, Valeriu Lacatusu, Tuan Tran, Sylvestre-Alvise Rebuffi, Alexandre Mourachko Published: 2025-12-18 Categories: cs.CR, cs.CL arXiv: https://arxiv.org/abs/2512.16904v1 Generation-time text watermarking embeds statistical signals into text for traceability of AI-generated content. We explore *post-hoc watermarking* where an LLM rewrites existing text while applying generation-time watermarking, to protect copyrighted documents, or detect their use in training or RAG via watermark radioactivity. Unlike generation-time approaches, which is constrained by how LLMs are served, this setting offers additional degrees of freedom for both generation and detection. We investigate how allocating compute (through larger rephrasing models, beam search, multi-candidate generation, or entropy filtering at detection) affects the quality-detectability trade-off. Our strategies achieve strong detectability and semantic fidelity on open-ended text such as books. Among our findings, the simple Gumbel-max scheme surprisingly outperforms more recent alternatives under nucleus sampling, and most methods benefit significantly from beam search. However, most approaches struggle when watermarking verifiable text such as code, where we counterintuitively find that smaller models outperform larger ones. This study reveals both the potential and limitations of post-hoc watermarking, laying groundwork for practical applications and future research. --- ### In-Context Algebra Authors: Eric Todd, Jannik Brinkmann, Rohit Gandikota, David Bau Published: 2025-12-18 Categories: cs.CL, cs.LG arXiv: https://arxiv.org/abs/2512.16902v2 We investigate the mechanisms that arise when transformers are trained to solve arithmetic on sequences where tokens are variables whose meaning is determined only through their interactions in-context. While prior work has studied transformers in settings where the answer relies on fixed parametric or geometric information encoded in token embeddings, we devise a new in-context reasoning task where the assignment of tokens to specific algebraic elements varies from one sequence to another. Despite this challenging setup, transformers achieve near-perfect accuracy on the task and even generalize to unseen groups. We develop targeted data distributions to create causal tests of a set of hypothesized mechanisms, and we isolate three mechanisms models consistently learn: commutative copying where a dedicated head copies answers, identity element recognition that distinguishes identity-containing facts, and closure-based cancellation that tracks group membership to constrain valid answers. Our findings show that the kinds of reasoning strategies learned by transformers are dependent on the task structure and that models can develop symbolic reasoning mechanisms when trained to reason in-context about variables whose meanings are not fixed. --- ### In-Context Algebra Authors: Eric Todd, Jannik Brinkmann, Rohit Gandikota, David Bau Published: 2025-12-18 Categories: cs.CL, cs.LG arXiv: https://arxiv.org/abs/2512.16902v1 We investigate the mechanisms that arise when transformers are trained to solve arithmetic on sequences where tokens are variables whose meaning is determined only through their interactions. While prior work has found that transformers develop geometric embeddings that mirror algebraic structure, those previous findings emerge from settings where arithmetic-valued tokens have fixed meanings. We devise a new task in which the assignment of symbols to specific algebraic group elements varies from one sequence to another. Despite this challenging setup, transformers achieve near-perfect accuracy on the task and even generalize to unseen algebraic groups. We develop targeted data distributions to create causal tests of a set of hypothesized mechanisms, and we isolate three mechanisms models consistently learn: commutative copying where a dedicated head copies answers, identity element recognition that distinguishes identity-containing facts, and closure-based cancellation that tracks group membership to constrain valid answers. Complementary to the geometric representations found in fixed-symbol settings, our findings show that models develop symbolic reasoning mechanisms when trained to reason in-context with variables whose meanings are not fixed. --- ### Impacts of Racial Bias in Historical Training Data for News AI Authors: Rahul Bhargava, Malene Hornstrup Jespersen, Emily Boardman Ndulue, Vivica Dsouza Published: 2025-12-18 Categories: cs.LG, cs.AI, cs.CL, cs.CY arXiv: https://arxiv.org/abs/2512.16901v1 AI technologies have rapidly moved into business and research applications that involve large text corpora, including computational journalism research and newsroom settings. These models, trained on extant data from various sources, can be conceptualized as historical artifacts that encode decades-old attitudes and stereotypes. This paper investigates one such example trained on the broadly-used New York Times Annotated Corpus to create a multi-label classifier. Our use in research settings surfaced the concerning "blacks" thematic topic label. Through quantitative and qualitative means we investigate this label's use in the training corpus, what concepts it might be encoding in the trained classifier, and how those concepts impact our model use. Via the application of explainable AI methods, we find that the "blacks" label operates partially as a general "racism detector" across some minoritized groups. However, it performs poorly against expectations on modern examples such as COVID-19 era anti-Asian hate stories, and reporting on the Black Lives Matter movement. This case study of interrogating embedded biases in a model reveals how similar applications in newsroom settings can lead to unexpected outputs that could impact a wide variety of potential uses of any large language model-story discovery, audience targeting, summarization, etc. The fundamental tension this exposes for newsrooms is how to adopt AI-enabled workflow tools while reducing the risk of reproducing historical biases in news coverage. --- ### FlashPortrait: 6x Faster Infinite Portrait Animation with Adaptive Latent Prediction Authors: Shuyuan Tu, Yueming Pan, Yinming Huang, Xintong Han, Zhen Xing, Qi Dai, Kai Qiu, Chong Luo, Zuxuan Wu Published: 2025-12-18 Categories: cs.CV arXiv: https://arxiv.org/abs/2512.16900v1 Current diffusion-based acceleration methods for long-portrait animation struggle to ensure identity (ID) consistency. This paper presents FlashPortrait, an end-to-end video diffusion transformer capable of synthesizing ID-preserving, infinite-length videos while achieving up to 6x acceleration in inference speed. In particular, FlashPortrait begins by computing the identity-agnostic facial expression features with an off-the-shelf extractor. It then introduces a Normalized Facial Expression Block to align facial features with diffusion latents by normalizing them with their respective means and variances, thereby improving identity stability in facial modeling. During inference, FlashPortrait adopts a dynamic sliding-window scheme with weighted blending in overlapping areas, ensuring smooth transitions and ID consistency in long animations. In each context window, based on the latent variation rate at particular timesteps and the derivative magnitude ratio among diffusion layers, FlashPortrait utilizes higher-order latent derivatives at the current timestep to directly predict latents at future timesteps, thereby skipping several denoising steps and achieving 6x speed acceleration. Experiments on benchmarks show the effectiveness of FlashPortrait both qualitatively and quantitatively. --- ### Multimodal RewardBench 2: Evaluating Omni Reward Models for Interleaved Text and Image Authors: Yushi Hu, Reyhane Askari-Hemmat, Melissa Hall, Emily Dinan, Luke Zettlemoyer, Marjan Ghazvininejad Published: 2025-12-18 Categories: cs.CL, cs.CV arXiv: https://arxiv.org/abs/2512.16899v3 Reward models (RMs) are essential for training large language models (LLMs), but remain underexplored for omni models that handle interleaved image and text sequences. We introduce Multimodal RewardBench 2 (MMRB2), the first comprehensive benchmark for reward models on multimodal understanding and (interleaved) generation. MMRB2 spans four tasks: text-to-image, image editing, interleaved generation, and multimodal reasoning ("thinking-with-images"), providing 1,000 expert-annotated preference pairs per task from 23 models and agents across 21 source tasks. MMRB2 is designed with: (1) practical but challenging prompts; (2) responses from state-of-the-art models and agents; and (3) preference pairs with strong human-expert consensus, curated via an ensemble filtering strategy. Using MMRB2, we study existing judges for each subtask, including multimodal LLM-as-a-judge and models trained with human preferences. The latest Gemini 3 Pro attains 75-80% accuracy. GPT-5 and Gemini 2.5 Pro reach 66-75% accuracy, compared to >90% for humans, yet surpass the widely used GPT-4o (59%). The best performing open-source model Qwen3-VL-32B achieves similar accuracies as Gemini 2.5 Flash (64%). We also show that MMRB2 performance strongly correlates with downstream task success using Best-of-N sampling and conduct an in-depth analysis that shows key areas to improve the reward models going forward. --- ### Multimodal RewardBench 2: Evaluating Omni Reward Models for Interleaved Text and Image Authors: Yushi Hu, Reyhane Askari-Hemmat, Melissa Hall, Emily Dinan, Luke Zettlemoyer, Marjan Ghazvininejad Published: 2025-12-18 Categories: cs.CL, cs.CV arXiv: https://arxiv.org/abs/2512.16899v1 Reward models (RMs) are essential for training large language models (LLMs), but remain underexplored for omni models that handle interleaved image and text sequences. We introduce Multimodal RewardBench 2 (MMRB2), the first comprehensive benchmark for reward models on multimodal understanding and (interleaved) generation. MMRB2 spans four tasks: text-to-image, image editing, interleaved generation, and multimodal reasoning ("thinking-with-images"), providing 1,000 expert-annotated preference pairs per task from 23 models and agents across 21 source tasks. MMRB2 is designed with: (1) practical but challenging prompts; (2) responses from state-of-the-art models and agents; and (3) preference pairs with strong human-expert consensus, curated via an ensemble filtering strategy. Using MMRB2, we study existing judges for each subtask, including multimodal LLM-as-a-judge and models trained with human preferences. The latest Gemini 3 Pro attains 75-80% accuracy. GPT-5 and Gemini 2.5 Pro reach 66-75% accuracy, compared to >90% for humans, yet surpass the widely used GPT-4o (59%). The best performing open-source model Qwen3-VL-32B achieves similar accuracies as Gemini 2.5 Flash (64%). We also show that MMRB2 performance strongly correlates with downstream task success using Best-of-N sampling and conduct an in-depth analysis that shows key areas to improve the reward models going forward. --- ### Sceniris: A Fast Procedural Scene Generation Framework Authors: Jinghuan Shang, Harsh Patel, Ran Gong, Karl Schmeckpeper Published: 2025-12-18 Categories: cs.RO, cs.CV, cs.GR arXiv: https://arxiv.org/abs/2512.16896v1 Synthetic 3D scenes are essential for developing Physical AI and generative models. Existing procedural generation methods often have low output throughput, creating a significant bottleneck in scaling up dataset creation. In this work, we introduce Sceniris, a highly efficient procedural scene generation framework for rapidly generating large-scale, collision-free scene variations. Sceniris also provides an optional robot reachability check, providing manipulation-feasible scenes for robot tasks. Sceniris is designed for maximum efficiency by addressing the primary performance limitations of the prior method, Scene Synthesizer. Leveraging batch sampling and faster collision checking in cuRobo, Sceniris achieves at least 234x speed-up over Scene Synthesizer. Sceniris also expands the object-wise spatial relationships available in prior work to support diverse scene requirements. Our code is available at https://github.com/rai-inst/sceniris --- ### Instant Expressive Gaussian Head Avatars at Over 100 FPS Authors: Kaiwen Jiang, Xueting Li, Seonwook Park, Ravi Ramamoorthi, Shalini De Mello, Koki Nagano Published: 2025-12-18 Categories: cs.CV arXiv: https://arxiv.org/abs/2512.16893v2 Portrait animation has witnessed tremendous quality improvements thanks to recent advances in video diffusion models. However, these 2D methods often compromise 3D consistency and speed, limiting their applicability in real-world scenarios, such as digital twins or telepresence. In contrast, 3D-aware feedforward facial animation methods -- built upon 3D representations, such as neural radiance fields or Gaussian splatting -- ensure 3D consistency and achieve faster inference speed, but come with inferior expression details. In this paper, we address this portrait animation trilemma (speed, 3D consistency, and expressiveness) and propose a pipeline that instantly converts an in-the-wild single image into a 3D-consistent, fast yet expressive animatable representation via a feed-forward encoder. Unlike previous computationally intensive global fusion mechanisms (e.g., multiple attention layers) for fusing 3D structural and animation information, our design employs an efficient lightweight local fusion strategy to achieve high animation expressivity. Furthermore, our animation representation is decoupled from the face's 3D representation and learns motion implicitly from data, eliminating the dependency on pre-defined parametric models that often constrain animation capabilities. Our method runs at 107.31 FPS for animation and pose control, representing a 3-4 order of magnitude speedup versus the state of the art while achieving comparable animation quality, thus surpassing alternative designs that trade speed for quality or vice versa. --- ### Instant Expressive Gaussian Head Avatar via 3D-Aware Expression Distillation Authors: Kaiwen Jiang, Xueting Li, Seonwook Park, Ravi Ramamoorthi, Shalini De Mello, Koki Nagano Published: 2025-12-18 Categories: cs.CV arXiv: https://arxiv.org/abs/2512.16893v1 Portrait animation has witnessed tremendous quality improvements thanks to recent advances in video diffusion models. However, these 2D methods often compromise 3D consistency and speed, limiting their applicability in real-world scenarios, such as digital twins or telepresence. In contrast, 3D-aware facial animation feedforward methods -- built upon explicit 3D representations, such as neural radiance fields or Gaussian splatting -- ensure 3D consistency and achieve faster inference speed, but come with inferior expression details. In this paper, we aim to combine their strengths by distilling knowledge from a 2D diffusion-based method into a feed-forward encoder, which instantly converts an in-the-wild single image into a 3D-consistent, fast yet expressive animatable representation. Our animation representation is decoupled from the face's 3D representation and learns motion implicitly from data, eliminating the dependency on pre-defined parametric models that often constrain animation capabilities. Unlike previous computationally intensive global fusion mechanisms (e.g., multiple attention layers) for fusing 3D structural and animation information, our design employs an efficient lightweight local fusion strategy to achieve high animation expressivity. As a result, our method runs at 107.31 FPS for animation and pose control while achieving comparable animation quality to the state-of-the-art, surpassing alternative designs that trade speed for quality or vice versa. Project website is https://research.nvidia.com/labs/amri/projects/instant4d --- ### LinkedOut: Linking World Knowledge Representation Out of Video LLM for Next-Generation Video Recommendation Authors: Haichao Zhang, Yao Lu, Lichen Wang, Yunzhe Li, Daiwei Chen, Yunpeng Xu, Yun Fu Published: 2025-12-18 Categories: cs.CV, cs.AI, cs.IR, cs.LG, cs.MM arXiv: https://arxiv.org/abs/2512.16891v1 Video Large Language Models (VLLMs) unlock world-knowledge-aware video understanding through pretraining on internet-scale data and have already shown promise on tasks such as movie analysis and video question answering. However, deploying VLLMs for downstream tasks such as video recommendation remains challenging, since real systems require multi-video inputs, lightweight backbones, low-latency sequential inference, and rapid response. In practice, (1) decode-only generation yields high latency for sequential inference, (2) typical interfaces do not support multi-video inputs, and (3) constraining outputs to language discards fine-grained visual details that matter for downstream vision tasks. We argue that these limitations stem from the absence of a representation that preserves pixel-level detail while leveraging world knowledge. We present LinkedOut, a representation that extracts VLLM world knowledge directly from video to enable fast inference, supports multi-video histories, and removes the language bottleneck. LinkedOut extracts semantically grounded, knowledge-aware tokens from raw frames using VLLMs, guided by promptable queries and optional auxiliary modalities. We introduce a cross-layer knowledge fusion MoE that selects the appropriate level of abstraction from the rich VLLM features, enabling personalized, interpretable, and low-latency recommendation. To our knowledge, LinkedOut is the first VLLM-based video recommendation method that operates on raw frames without handcrafted labels, achieving state-of-the-art results on standard benchmarks. Interpretability studies and ablations confirm the benefits of layer diversity and layer-wise fusion, pointing to a practical path that fully leverages VLLM world-knowledge priors and visual reasoning for downstream vision tasks such as recommendation. --- ### M-PhyGs: Multi-Material Object Dynamics from Video Authors: Norika Wada, Kohei Yamashita, Ryo Kawahara, Ko Nishino Published: 2025-12-18 Categories: cs.CV arXiv: https://arxiv.org/abs/2512.16885v1 Knowledge of the physical material properties governing the dynamics of a real-world object becomes necessary to accurately anticipate its response to unseen interactions. Existing methods for estimating such physical material parameters from visual data assume homogeneous single-material objects, pre-learned dynamics, or simplistic topologies. Real-world objects, however, are often complex in material composition and geometry lying outside the realm of these assumptions. In this paper, we particularly focus on flowers as a representative common object. We introduce Multi-material Physical Gaussians (M-PhyGs) to estimate the material composition and parameters of such multi-material complex natural objects from video. From a short video captured in a natural setting, M-PhyGs jointly segments the object into similar materials and recovers their continuum mechanical parameters while accounting for gravity. M-PhyGs achieves this efficiently with newly introduced cascaded 3D and 2D losses, and by leveraging temporal mini-batching. We introduce a dataset, Phlowers, of people interacting with flowers as a novel platform to evaluate the accuracy of this challenging task of multi-material physical parameter estimation. Experimental results on Phlowers dataset demonstrate the accuracy and effectiveness of M-PhyGs and its components. --- ### AdaSearch: Balancing Parametric Knowledge and Search in Large Language Models via Reinforcement Learning Authors: Tzu-Han Lin, Wei-Lin Chen, Chen-An Li, Hung-yi Lee, Yun-Nung Chen, Yu Meng Published: 2025-12-18 Categories: cs.CL arXiv: https://arxiv.org/abs/2512.16883v2 Equipping large language models (LLMs) with search engines via reinforcement learning (RL) promises effective search agents. However, adaptively balancing internal parametric knowledge with external search remains a challenge, as overreliance on search introduces unnecessary cost and risks exposure to noisy or malicious content, while relying solely on parametric knowledge risks hallucination. Prior efforts mitigate search overuse through tool-call reward shaping, which requires heavy reward engineering and conflates necessary and unnecessary search. To address these limitations, we revisit the evaluation of search agents through an F1-based decision metric, revealing that prior methods often overlook readily available parametric knowledge. Motivated by this, we propose AdaSearch, a simple two-stage, outcome-driven RL framework that disentangles problem-solving from the decision to search, making the decision process explicit and interpretable. Extensive experiments demonstrate that AdaSearch significantly improves search-decision quality and reduces unnecessary search calls, with only a small trade-off in QA accuracy relative to always-search. --- ### AdaSearch: Balancing Parametric Knowledge and Search in Large Language Models via Reinforcement Learning Authors: Tzu-Han Lin, Wei-Lin Chen, Chen-An Li, Hung-yi Lee, Yun-Nung Chen, Yu Meng Published: 2025-12-18 Categories: cs.CL arXiv: https://arxiv.org/abs/2512.16883v1 Equipping large language models (LLMs) with search engines via reinforcement learning (RL) has emerged as an effective approach for building search agents. However, overreliance on search introduces unnecessary cost and risks exposure to noisy or malicious content, while relying solely on parametric knowledge risks hallucination. The central challenge is to develop agents that adaptively balance parametric knowledge with external search, invoking search only when necessary. Prior work mitigates search overuse by shaping rewards around the number of tool calls. However, these penalties require substantial reward engineering, provide ambiguous credit assignment, and can be exploited by agents that superficially reduce calls. Moreover, evaluating performance solely through call counts conflates necessary and unnecessary search, obscuring the measurement of true adaptive behavior. To address these limitations, we first quantify the self-knowledge awareness of existing search agents via an F1-based decision metric, revealing that methods such as Search-R1 often overlook readily available parametric knowledge. Motivated by these findings, we propose AdaSearch, a simple two-stage, outcome-driven RL framework that disentangles problem solving from the decision of whether to invoke search, and makes this decision process explicit and interpretable. This transparency is crucial for high-stakes domains such as finance and medical question answering, yet is largely neglected by prior approaches. Experiments across multiple model families and sizes demonstrate that AdaSearch substantially improves knowledge-boundary awareness, reduces unnecessary search calls, preserves strong task performance, and offers more transparent, interpretable decision behaviors. --- ### A Cartesian-3j Framework for Machine Learning Interatomic Potentials Authors: Zemin Xu, Chenyu Wu, Wenbo Xie, P. Hu Published: 2025-12-18 Categories: physics.chem-ph, cond-mat.mtrl-sci, cs.LG arXiv: https://arxiv.org/abs/2512.16882v2 Machine learning interatomic potentials (MLIPs) have brought substantial gains in the extrapolation capability in computational chemistry. However, most equivariant models are typically built with spherical tensors (STs), while Cartesian tensor formulations remain less developed despite their natural alignment with atomic coordinates and tensorial targets. In this work, we develop a Cartesian framework for irreducible Cartesian tensors (ICTs) by introduce the \texttt{Cartesian-3j} symbol and Cartesian Generalized Clebsch-Gordan Coefficients, which serve as direct analogues of the \texttt{Wigner-3j} symbol and Generalized Clebsch-Gordan coefficients defined for ST coupling. We extend the \texttt{e3nn} library to support ICT product, and use this framework to build Cartesian counterparts of \texttt{MACE}, \texttt{NequIP}, and \texttt{Allegro}, allowing the first controlled comparison where architectures are held fixed and only the tensor basis is changed. Our experiments show that irreducible Cartesian models can achieve accuracy comparable to spherical counterparts, but direct Cartesianization incurs unfavorable compute and memory scaling, motivating dedicated Cartesian architectural choices. Leveraging ICTs and our framework, we introduce \texttt{TACE-v1-OAM-M} and demonstrate that it achieves competitive performance on Matbench Discovery compared to state-of-the-art ST models. ---