# Research Papers — 2025-12 · bundle page 42 of 175 > Full-text markdown of papers #2051–2100 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/42 [← Previous bundle](https://aipod.dev/papers/archive/2025-12/page/41.md) [Next bundle →](https://aipod.dev/papers/archive/2025-12/page/43.md) ### Sign Language Recognition using Parallel Bidirectional Reservoir Computing Authors: Nitin Kumar Singh, Arie Rachmad Syulistyo, Yuichiro Tanaka, Hakaru Tamukoh Published: 2025-12-22 Categories: cs.CV, cs.RO arXiv: https://arxiv.org/abs/2512.19451v1 Sign language recognition (SLR) facilitates communication between deaf and hearing communities. Deep learning based SLR models are commonly used but require extensive computational resources, making them unsuitable for deployment on edge devices. To address these limitations, we propose a lightweight SLR system that combines parallel bidirectional reservoir computing (PBRC) with MediaPipe. MediaPipe enables real-time hand tracking and precise extraction of hand joint coordinates, which serve as input features for the PBRC architecture. The proposed PBRC architecture consists of two echo state network (ESN) based bidirectional reservoir computing (BRC) modules arranged in parallel to capture temporal dependencies, thereby creating a rich feature representation for classification. We trained our PBRC-based SLR system on the Word-Level American Sign Language (WLASL) video dataset, achieving top-1, top-5, and top-10 accuracies of 60.85%, 85.86%, and 91.74%, respectively. Training time was significantly reduced to 18.67 seconds due to the intrinsic properties of reservoir computing, compared to over 55 minutes for deep learning based methods such as Bi-GRU. This approach offers a lightweight, cost-effective solution for real-time SLR on edge devices. --- ### D2Pruner: Debiased Importance and Structural Diversity for MLLM Token Pruning Authors: Evelyn Zhang, Fufu Yu, Aoqi Wu, Zichen Wen, Ke Yan, Shouhong Ding, Biqing Qi, Linfeng Zhang Published: 2025-12-22 Categories: cs.CV arXiv: https://arxiv.org/abs/2512.19443v2 Processing long visual token sequences poses a significant computational burden on Multimodal Large Language Models (MLLMs). While token pruning offers a path to acceleration, we find that current methods, while adequate for general understanding, catastrophically fail on fine-grained localization tasks. We attribute this failure to the inherent flaws of the two prevailing strategies: importance-based methods suffer from a strong positional bias, an inherent model artifact that distracts from semantic content, while diversity-based methods exhibit structural blindness, disregarding the user's prompt and spatial redundancy. To address this, we introduce D2Pruner, a framework that rectifies these issues by uniquely combining debiased importance with a structural pruning mechanism. Our method first secures a core set of the most critical tokens as pivots based on a debiased attention score. It then performs a Maximal Independent Set (MIS) selection on the remaining tokens, which are modeled on a hybrid graph where edges signify spatial proximity and semantic similarity. This process iteratively preserves the most important and available token while removing its neighbors, ensuring that the supplementary tokens are chosen to maximize importance and diversity. Extensive experiments demonstrate that D2Pruner has exceptional efficiency and fidelity. Applied to LLaVA-1.5-7B for general understanding tasks, it reduces FLOPs by 74.2\% while retaining 99.2\% of its original performance. Furthermore, in challenging localization benchmarks with InternVL-2.5-8B, it maintains 85.7\% performance at a 90\% token reduction rate, marking a significant advancement with up to 63. 53\% improvement over existing methods. --- ### Real-Time Streamable Generative Speech Restoration with Flow Matching Authors: Simon Welker, Bunlong Lay, Maris Hillemann, Tal Peer, Timo Gerkmann Published: 2025-12-22 Categories: eess.SP, cs.LG, cs.SD arXiv: https://arxiv.org/abs/2512.19442v3 Diffusion-based generative models have greatly impacted the speech processing field in recent years, exhibiting high speech naturalness and spawning a new research direction. Their application in real-time communication is, however, still lagging behind due to their computation-heavy nature involving multiple calls of large DNNs. Here, we present Stream$.$FM, a frame-causal flow-based generative model with an algorithmic latency of 32 milliseconds (ms) and a total latency of 48 ms, paving the way for generative speech processing in real-time communication. We propose a buffered streaming inference scheme and an optimized DNN architecture, show how learned few-step numerical solvers can boost output quality at a fixed compute budget, explore model weight compression to find favorable points along a compute/quality tradeoff, and contribute a model variant with 24 ms total latency for the speech enhancement task. Our work looks beyond theoretical latencies, showing that high-quality streaming generative speech processing can be realized on consumer GPUs available today. Stream$.$FM can solve a variety of speech processing tasks in a streaming fashion: speech enhancement, dereverberation, codec post-filtering, bandwidth extension, STFT phase retrieval, and Mel vocoding. As we verify through comprehensive evaluations and a MUSHRA listening test, Stream$.$FM establishes a state-of-the-art for generative streaming speech restoration, exhibits only a reasonable reduction in quality compared to a non-streaming variant, and outperforms our recent work (Diffusion Buffer) on generative streaming speech enhancement while operating at a lower latency. --- ### Binary Kernel Logistic Regression: a sparsity-inducing formulation and a convergent decomposition training algorithm Authors: Antonio Consolo, Andrea Manno, Edoardo Amaldi Published: 2025-12-22 Categories: cs.LG, math.OC arXiv: https://arxiv.org/abs/2512.19440v1 Kernel logistic regression (KLR) is a widely used supervised learning method for binary and multi-class classification, which provides estimates of the conditional probabilities of class membership for the data points. Unlike other kernel methods such as Support Vector Machines (SVMs), KLRs are generally not sparse. Previous attempts to deal with sparsity in KLR include a heuristic method referred to as the Import Vector Machine (IVM) and ad hoc regularizations such as the $\ell_{1/2}$-based one. Achieving a good trade-off between prediction accuracy and sparsity is still a challenging issue with a potential significant impact from the application point of view. In this work, we revisit binary KLR and propose an extension of the training formulation proposed by Keerthi et al., which is able to induce sparsity in the trained model, while maintaining good testing accuracy. To efficiently solve the dual of this formulation, we devise a decomposition algorithm of Sequential Minimal Optimization type which exploits second-order information, and for which we establish global convergence. Numerical experiments conducted on 12 datasets from the literature show that the proposed binary KLR approach achieves a competitive trade-off between accuracy and sparsity with respect to IVM, $\ell_{1/2}$-based regularization for KLR, and SVM while retaining the advantages of providing informative estimates of the class membership probabilities. --- ### An Inverse Scattering Inspired Fourier Neural Operator for Time-Dependent PDE Learning Authors: Rixin Yu Published: 2025-12-22 Categories: cs.LG arXiv: https://arxiv.org/abs/2512.19439v1 Learning accurate and stable time-advancement operators for nonlinear partial differential equations (PDEs) remains challenging, particularly for chaotic, stiff, and long-horizon dynamical systems. While neural operator methods such as the Fourier Neural Operator (FNO) and Koopman-inspired extensions achieve good short-term accuracy, their long-term stability is often limited by unconstrained latent representations and cumulative rollout errors. In this work, we introduce an inverse scattering inspired Fourier Neural Operator(IS-FNO), motivated by the reversibility and spectral evolution structure underlying the classical inverse scattering transform. The proposed architecture enforces a near-reversible pairing between lifting and projection maps through an explicitly invertible neural transformation, and models latent temporal evolution using exponential Fourier layers that naturally encode linear and nonlinear spectral dynamics. We systematically evaluate IS-FNO against baseline FNO and Koopman-based models on a range of benchmark PDEs, including the Michelson-Sivashinsky and Kuramoto-Sivashinsky equations (in one and two dimensions), as well as the integrable Korteweg-de Vries and Kadomtsev-Petviashvili equations. The results demonstrate that IS-FNO achieves lower short-term errors and substantially improved long-horizon stability in non-stiff regimes. For integrable systems, reduced IS-FNO variants that embed analytical scattering structure retain competitive long-term accuracy despite limited model capacity. Overall, this work shows that incorporating physical structure -- particularly reversibility and spectral evolution -- into neural operator design significantly enhances robustness and long-term predictive fidelity for nonlinear PDE dynamics. --- ### MT-Mark: Rethinking Image Watermarking via Mutual-Teacher Collaboration with Adaptive Feature Modulation Authors: Fei Ge, Ying Huang, Jie Liu, Guixuan Zhang, Zhi Zeng, Shuwu Zhang, Hu Guan Published: 2025-12-22 Categories: cs.CV, cs.AI arXiv: https://arxiv.org/abs/2512.19438v1 Existing deep image watermarking methods follow a fixed embedding-distortion-extraction pipeline, where the embedder and extractor are weakly coupled through a final loss and optimized in isolation. This design lacks explicit collaboration, leaving no structured mechanism for the embedder to incorporate decoding-aware cues or for the extractor to guide embedding during training. To address this architectural limitation, we rethink deep image watermarking by reformulating embedding and extraction as explicitly collaborative components. To realize this reformulation, we introduce a Collaborative Interaction Mechanism (CIM) that establishes direct, bidirectional communication between the embedder and extractor, enabling a mutual-teacher training paradigm and coordinated optimization. Built upon this explicitly collaborative architecture, we further propose an Adaptive Feature Modulation Module (AFMM) to support effective interaction. AFMM enables content-aware feature regulation by decoupling modulation structure and strength, guiding watermark embedding toward stable image features while suppressing host interference during extraction. Under CIM, the AFMMs on both sides form a closed-loop collaboration that aligns embedding behavior with extraction objectives. This architecture-level redesign changes how robustness is learned in watermarking systems. Rather than relying on exhaustive distortion simulation, robustness emerges from coordinated representation learning between embedding and extraction. Experiments on real-world and AI-generated datasets demonstrate that the proposed method consistently outperforms state-of-the-art approaches in watermark extraction accuracy while maintaining high perceptual quality, showing strong robustness and generalization. --- ### dMLLM-TTS: Self-Verified and Efficient Test-Time Scaling for Diffusion Multi-Modal Large Language Models Authors: Yi Xin, Siqi Luo, Tianxiang Xu, Qi Qin, Haoxing Chen, Kaiwen Zhu, Zhiwei Zhang, Yangfan He, Rongchao Zhang, Jinbin Bai, Shuo Cao, Bin Fu et al. Published: 2025-12-22 Categories: cs.CV arXiv: https://arxiv.org/abs/2512.19433v2 Diffusion Multi-modal Large Language Models (dMLLMs) have recently emerged as a novel architecture unifying image generation and understanding. However, developing effective and efficient Test-Time Scaling (TTS) methods to unlock their full generative potential remains an underexplored challenge. To address this, we propose dMLLM-TTS, a novel framework operating on two complementary scaling axes: (1) trajectory exploration scaling to enhance the diversity of generated hypotheses, and (2) iterative refinement scaling for stable generation. Conventional TTS approaches typically perform linear search across these two dimensions, incurring substantial computational costs of O(NT) and requiring an external verifier for best-of-N selection. To overcome these limitations, we propose two innovations. First, we design an efficient hierarchical search algorithm with O(N+T) complexity that adaptively expands and prunes sampling trajectories. Second, we introduce a self-verified feedback mechanism that leverages the dMLLMs' intrinsic image understanding capabilities to assess text-image alignment, eliminating the need for external verifier. Extensive experiments on the GenEval benchmark across three representative dMLLMs (e.g., Lumina-DiMOO, MMaDA, Muddit) show that our framework substantially improves generation quality while achieving up to 6x greater efficiency than linear search. Project page: https://github.com/Alpha-VLLM/Lumina-DiMOO. --- ### MobileWorld: Benchmarking Autonomous Mobile Agents in Agent-User Interactive and MCP-Augmented Environments Authors: Quyu Kong, Xu Zhang, Zhenyu Yang, Nolan Gao, Chen Liu, Panrong Tong, Chenglin Cai, Hanzhang Zhou, Jianan Zhang, Liangyu Chen, Zhidan Liu, Steven Hoi et al. Published: 2025-12-22 Categories: cs.CL arXiv: https://arxiv.org/abs/2512.19432v3 Among existing online mobile-use benchmarks, AndroidWorld has emerged as the dominant benchmark due to its reproducible environment and deterministic evaluation; however, recent agents achieving over 90% success rates indicate its saturation and motivate the need for a more challenging benchmark. In addition, its environment lacks key application categories, such as e-commerce and enterprise communication, and does not reflect realistic mobile-use scenarios characterized by vague user instructions and hybrid tool usage. We introduce MobileWorld, a substantially more challenging benchmark designed to reflect real-world usage through 201 tasks across 20 applications. MobileWorld derives its difficulty from an emphasis on long-horizon, cross-application workflows, requiring nearly twice as many completion steps on average (27.8 vs. 14.3) and featuring a significantly higher proportion of multi-app tasks (62.2% vs. 9.5%) than AndroidWorld. To overcome the limitations of existing environments, MobileWorld achieves a balance between production-grade utility and reproducible evaluation by utilizing open-source alternatives to industry standards (e.g., Mattermost for Slack). This approach enables a fully observable and controlled environment through source code modification and direct backend database access for precise verification. MobileWorld also introduces novel task categories, including agent-user interaction and Model Context Protocol (MCP)-augmented tasks, for evaluating agents in user-aware, hybrid-tool scenarios. To facilitate evaluation, we develop a planner-executor agentic framework with extended action spaces to support user interactions and MCP calls. Our results reveal a sharp performance drop compared to AndroidWorld, with the best agentic framework and end-to-end model achieving 51.7% and 20.9% success rates, respectively, highlighting ample headroom for future research. --- ### Attention Is Not What You Need Authors: Zhang Chong Published: 2025-12-22 Categories: cs.LG, cs.AI, math.AG arXiv: https://arxiv.org/abs/2512.19428v1 We revisit a basic question in sequence modeling: is explicit self-attention actually necessary for strong performance and reasoning? We argue that standard multi-head attention is best seen as a form of tensor lifting: hidden vectors are mapped into a high-dimensional space of pairwise interactions, and learning proceeds by constraining this lifted tensor through gradient descent. This mechanism is extremely expressive but mathematically opaque, because after many layers it becomes very hard to describe the model with a small family of explicit invariants. To explore an alternative, we propose an attention-free architecture based on Grassmann flows. Instead of forming an L by L attention matrix, our Causal Grassmann layer (i) linearly reduces token states, (ii) encodes local token pairs as two-dimensional subspaces on a Grassmann manifold via Plucker coordinates, and (iii) fuses these geometric features back into the hidden states through gated mixing. Information therefore propagates by controlled deformations of low-rank subspaces over multi-scale local windows, so the core computation lives on a finite-dimensional manifold rather than in an unstructured tensor space. On the Wikitext-2 language modeling benchmark, purely Grassmann-based models with 13 to 18 million parameters achieve validation perplexities within about 10 to 15 percent of size-matched Transformers. On the SNLI natural language inference task, a Grassmann-Plucker head on top of DistilBERT slightly outperforms a Transformer head, with best validation and test accuracies of 0.8550 and 0.8538 compared to 0.8545 and 0.8511. We analyze the complexity of Grassmann mixing, show linear scaling in sequence length for fixed rank, and argue that such manifold-based designs offer a more structured route toward geometric and invariant-based interpretations of neural reasoning. --- ### CodeSimpleQA: Scaling Factuality in Code Large Language Models Authors: Jian Yang, Wei Zhang, Yizhi Li, Shawn Guo, Haowen Wang, Aishan Liu, Ge Zhang, Zili Wang, Zhoujun Li, Xianglong Liu, Weifeng Lv Published: 2025-12-22 Categories: cs.CL arXiv: https://arxiv.org/abs/2512.19424v1 Large language models (LLMs) have made significant strides in code generation, achieving impressive capabilities in synthesizing code snippets from natural language instructions. However, a critical challenge remains in ensuring LLMs generate factually accurate responses about programming concepts, technical implementations, etc. Most previous code-related benchmarks focus on code execution correctness, overlooking the factual accuracy of programming knowledge. To address this gap, we present CodeSimpleQA, a comprehensive bilingual benchmark designed to evaluate the factual accuracy of code LLMs in answering code-related questions, which contains carefully curated question-answer pairs in both English and Chinese, covering diverse programming languages and major computer science domains. Further, we create CodeSimpleQA-Instruct, a large-scale instruction corpus with 66M samples, and develop a post-training framework combining supervised fine-tuning and reinforcement learning. Our comprehensive evaluation of diverse LLMs reveals that even frontier LLMs struggle with code factuality. Our proposed framework demonstrates substantial improvements over the base model, underscoring the critical importance of factuality-aware alignment in developing reliable code LLMs. --- ### Non-Contrast CT Esophageal Varices Grading through Clinical Prior-Enhanced Multi-Organ Analysis Authors: Xiaoming Zhang, Chunli Li, Jiacheng Hao, Yuan Gao, Danyang Tu, Jianyi Qiao, Xiaoli Yin, Le Lu, Ling Zhang, Ke Yan, Yang Hou, Yu Shi Published: 2025-12-22 Categories: cs.CV arXiv: https://arxiv.org/abs/2512.19415v2 Esophageal varices (EV) represent a critical complication of portal hypertension, affecting approximately 60% of cirrhosis patients with a significant bleeding risk of ~30%. While traditionally diagnosed through invasive endoscopy, non-contrast computed tomography (NCCT) presents a potential non-invasive alternative that has yet to be fully utilized in clinical practice. We present Multi-Organ-COhesion Network++ (MOON++), a novel multimodal framework that enhances EV assessment through comprehensive analysis of NCCT scans. Inspired by clinical evidence correlating organ volumetric relationships with liver disease severity, MOON++ synthesizes imaging characteristics of the esophagus, liver, and spleen through multimodal learning. We evaluated our approach using 1,631 patients, those with endoscopically confirmed EV were classified into four severity grades. Validation in 239 patient cases and independent testing in 289 cases demonstrate superior performance compared to conventional single organ methods, achieving an AUC of 0.894 versus 0.803 for the severe grade EV classification (G3 versus <G3) and 0.921 versus 0.793 for the differentiation of moderate to severe grades (>=G2 versus <G2). We conducted a reader study involving experienced radiologists to further validate the performance of MOON++. To our knowledge, MOON++ represents the first comprehensive multi-organ NCCT analysis framework incorporating clinical knowledge priors for EV assessment, potentially offering a promising non-invasive diagnostic alternative. --- ### From Retrieval to Reasoning: A Framework for Cyber Threat Intelligence NER with Explicit and Adaptive Instructions Authors: Jiaren Peng, Hongda Sun, Xuan Tian, Cheng Huang, Zeqing Li, Rui Yan Published: 2025-12-22 Categories: cs.CR, cs.CL arXiv: https://arxiv.org/abs/2512.19414v1 The automation of Cyber Threat Intelligence (CTI) relies heavily on Named Entity Recognition (NER) to extract critical entities from unstructured text. Currently, Large Language Models (LLMs) primarily address this task through retrieval-based In-Context Learning (ICL). This paper analyzes this mainstream paradigm, revealing a fundamental flaw: its success stems not from global semantic similarity but largely from the incidental overlap of entity types within retrieved examples. This exposes the limitations of relying on unreliable implicit induction. To address this, we propose TTPrompt, a framework shifting from implicit induction to explicit instruction. TTPrompt maps the core concepts of CTI's Tactics, Techniques, and Procedures (TTPs) into an instruction hierarchy: formulating task definitions as Tactics, guiding strategies as Techniques, and annotation guidelines as Procedures. Furthermore, to handle the adaptability challenge of static guidelines, we introduce Feedback-driven Instruction Refinement (FIR). FIR enables LLMs to self-refine guidelines by learning from errors on minimal labeled data, adapting to distinct annotation dialects. Experiments on five CTI NER benchmarks demonstrate that TTPrompt consistently surpasses retrieval-based baselines. Notably, with refinement on just 1% of training data, it rivals models fine-tuned on the full dataset. For instance, on LADDER, its Micro F1 of 71.96% approaches the fine-tuned baseline, and on the complex CTINexus, its Macro F1 exceeds the fine-tuned ACLM model by 10.91%. --- ### Research Program: Theory of Learning in Dynamical Systems Authors: Elad Hazan, Shai Shalev Shwartz, Nathan Srebro Published: 2025-12-22 Categories: cs.LG, cs.AI arXiv: https://arxiv.org/abs/2512.19410v1 Modern learning systems increasingly interact with data that evolve over time and depend on hidden internal state. We ask a basic question: when is such a dynamical system learnable from observations alone? This paper proposes a research program for understanding learnability in dynamical systems through the lens of next-token prediction. We argue that learnability in dynamical systems should be studied as a finite-sample question, and be based on the properties of the underlying dynamics rather than the statistical properties of the resulting sequence. To this end, we give a formulation of learnability for stochastic processes induced by dynamical systems, focusing on guarantees that hold uniformly at every time step after a finite burn-in period. This leads to a notion of dynamic learnability which captures how the structure of a system, such as stability, mixing, observability, and spectral properties, governs the number of observations required before reliable prediction becomes possible. We illustrate the framework in the case of linear dynamical systems, showing that accurate prediction can be achieved after finite observation without system identification, by leveraging improper methods based on spectral filtering. We survey the relationship between learning in dynamical systems and classical PAC, online, and universal prediction theories, and suggest directions for studying nonlinear and controlled systems. --- ### Symplectic Representation of Legendre Dynamics Authors: Robert Simon Fong, Gouhei Tanaka, Kazuyuki Aihara Published: 2025-12-22 Categories: cs.LG arXiv: https://arxiv.org/abs/2512.19409v2 Modern learning systems act on internal representations of data, yet how these representations encode underlying physical or statistical structure is often left implicit. In physics, symplecticity keeps Hamiltonian systems faithful to their phase-space geometry. Recent learning methods impose such geometric structure either in the dynamics or through training losses. Here we ask a different question: what would it mean for the representation itself to obey a symplectic conservation law? We pose this representation-level constraint through Legendre duality: the relation $p = dψ(q)$ between primal and dual coordinates, which in exponential family models is the information-geometric pairing of natural and expectation parameters. We formalize Legendre dynamics as stochastic processes whose trajectories remain on Legendre graphs, where the evolving primal-dual parameters stay Legendre dual. We show that this class includes linear time-invariant Gaussian process regression and Ornstein-Uhlenbeck dynamics. Geometrically, we characterize the symplectomorphisms of cotangent bundles that preserve all Legendre graphs. We show that these maps are exactly cotangent lifts of base diffeomorphisms followed by exact fibre translations. This gives an explicit normal form for Legendre-preserving representation updates. Dynamically, we prove that the normal form is realized by Hamiltonians that are at most linear in the momentum. This realization principle is used to construct linear and nonlinear Hamiltonian Symplectic Reservoirs (SR) whose recurrent updates preserve Legendre graphs by construction. This is the only normal form that preserves Legendre duality, so the architecture follows from the invariant. Numerical experiments confirm the normal-form identities and distinguish Legendre preserving Hamiltonian SRs from generic symplectic and standard reservoir baselines. --- ### Real2Edit2Real: Generating Robotic Demonstrations via a 3D Control Interface Authors: Yujie Zhao, Hongwei Fan, Di Chen, Shengcong Chen, Liliang Chen, Xiaoqi Li, Guanghui Ren, Hao Dong Published: 2025-12-22 Categories: cs.RO, cs.CV, cs.GR arXiv: https://arxiv.org/abs/2512.19402v2 Recent progress in robot learning has been driven by large-scale datasets and powerful visuomotor policy architectures, yet policy robustness remains limited by the substantial cost of collecting diverse demonstrations, particularly for spatial generalization in manipulation tasks. To reduce repetitive data collection, we present Real2Edit2Real, a framework that generates new demonstrations by bridging 3D editability with 2D visual data through a 3D control interface. Our approach first reconstructs scene geometry from multi-view RGB observations with a metric-scale 3D reconstruction model. Based on the reconstructed geometry, we perform depth-reliable 3D editing on point clouds to generate new manipulation trajectories while geometrically correcting the robot poses to recover physically consistent depth, which serves as a reliable condition for synthesizing new demonstrations. Finally, we propose a multi-conditional video generation model guided by depth as the primary control signal, together with action, edge, and ray maps, to synthesize spatially augmented multi-view manipulation videos. Experiments on four real-world manipulation tasks demonstrate that policies trained on data generated from only 1-5 source demonstrations can match or outperform those trained on 50 real-world demonstrations, improving data efficiency by up to 10-50x. Moreover, experimental results on height and texture editing demonstrate the framework's flexibility and extensibility, indicating its potential to serve as a unified data generation framework. Project website is https://real2edit2real.github.io/. --- ### Kunnafonidilaw ka Cadeau: an ASR dataset of present-day Bambara Authors: Yacouba Diarra, Panga Azazia Kamate, Nouhoum Souleymane Coulibaly, Michael Leventhal Published: 2025-12-22 Categories: cs.CL arXiv: https://arxiv.org/abs/2512.19400v1 We present Kunkado, a 160-hour Bambara ASR dataset compiled from Malian radio archives to capture present-day spontaneous speech across a wide range of topics. It includes code-switching, disfluencies, background noise, and overlapping speakers that practical ASR systems encounter in real-world use. We finetuned Parakeet-based models on a 33.47-hour human-reviewed subset and apply pragmatic transcript normalization to reduce variability in number formatting, tags, and code-switching annotations. Evaluated on two real-world test sets, finetuning with Kunkado reduces WER from 44.47\% to 37.12\% on one and from 36.07\% to 32.33\% on the other. In human evaluation, the resulting model also outperforms a comparable system with the same architecture trained on 98 hours of cleaner, less realistic speech. We release the data and models to support robust ASR for predominantly oral languages. --- ### Brain-Grounded Axes for Reading and Steering LLM States Authors: Sandro Andric Published: 2025-12-22 Categories: cs.LG, cs.AI, cs.CL arXiv: https://arxiv.org/abs/2512.19399v1 Interpretability methods for large language models (LLMs) typically derive directions from textual supervision, which can lack external grounding. We propose using human brain activity not as a training signal but as a coordinate system for reading and steering LLM states. Using the SMN4Lang MEG dataset, we construct a word-level brain atlas of phase-locking value (PLV) patterns and extract latent axes via ICA. We validate axes with independent lexica and NER-based labels (POS/log-frequency used as sanity checks), then train lightweight adapters that map LLM hidden states to these brain axes without fine-tuning the LLM. Steering along the resulting brain-derived directions yields a robust lexical (frequency-linked) axis in a mid TinyLlama layer, surviving perplexity-matched controls, and a brain-vs-text probe comparison shows larger log-frequency shifts (relative to the text probe) with lower perplexity for the brain axis. A function/content axis (axis 13) shows consistent steering in TinyLlama, Qwen2-0.5B, and GPT-2, with PPL-matched text-level corroboration. Layer-4 effects in TinyLlama are large but inconsistent, so we treat them as secondary (Appendix). Axis structure is stable when the atlas is rebuilt without GPT embedding-change features or with word2vec embeddings (|r|=0.64-0.95 across matched axes), reducing circularity concerns. Exploratory fMRI anchoring suggests potential alignment for embedding change and log frequency, but effects are sensitive to hemodynamic modeling assumptions and are treated as population-level evidence only. These results support a new interface: neurophysiology-grounded axes provide interpretable and controllable handles for LLM behavior. --- ### EchoTrail-GUI: Building Actionable Memory for GUI Agents via Critic-Guided Self-Exploration Authors: Runze Li, Yuwen Zhai, Bo Xu, LiWu Xu, Nian Shi, Wei Zhang, Ran Lin, Liang Wang Published: 2025-12-22 Categories: cs.AI arXiv: https://arxiv.org/abs/2512.19396v3 Contemporary GUI agents, while increasingly capable due to advances in Large Vision-Language Models (VLMs), often operate with a critical limitation: they treat each task in isolation, lacking a mechanism to systematically learn from past successes. This digital ''amnesia'' results in sub-optimal performance, repeated errors, and poor generalization to novel challenges. To bridge this gap, we introduce EchoTrail-GUI, a novel framework designed to mimic human-like experiential learning by equipping agents with a dynamic, accessible memory. Our framework operates in three distinct stages. First, during Experience Exploration, an agent autonomously interacts with GUI environments to build a curated database of successful task trajectories, validated by a reward model. Crucially, the entire knowledge base construction is thus fully automated, requiring no human supervision. Second, in the Memory Injection stage, upon receiving a new task, our system efficiently retrieves the most relevant past trajectories to serve as actionable ''memories''. Finally, during GUI Task Inference, these memories are injected as in-context guidance to inform the agent's reasoning and decision-making process. We demonstrate the efficacy of our approach on benchmarks including Android World and AndroidLab. The results show that EchoTrail-GUI significantly improves the task success rate and operational efficiency of baseline agents, validating the power of structured memory in creating more robust and intelligent GUI automation. --- ### TwinAligner: Visual-Dynamic Alignment Empowers Physics-aware Real2Sim2Real for Robotic Manipulation Authors: Hongwei Fan, Hang Dai, Jiyao Zhang, Jinzhou Li, Qiyang Yan, Yujie Zhao, Mingju Gao, Jinghang Wu, Hao Tang, Hao Dong Published: 2025-12-22 Categories: cs.RO, cs.CV, cs.GR arXiv: https://arxiv.org/abs/2512.19390v1 The robotics field is evolving towards data-driven, end-to-end learning, inspired by multimodal large models. However, reliance on expensive real-world data limits progress. Simulators offer cost-effective alternatives, but the gap between simulation and reality challenges effective policy transfer. This paper introduces TwinAligner, a novel Real2Sim2Real system that addresses both visual and dynamic gaps. The visual alignment module achieves pixel-level alignment through SDF reconstruction and editable 3DGS rendering, while the dynamic alignment module ensures dynamic consistency by identifying rigid physics from robot-object interaction. TwinAligner improves robot learning by providing scalable data collection and establishing a trustworthy iterative cycle, accelerating algorithm development. Quantitative evaluations highlight TwinAligner's strong capabilities in visual and dynamic real-to-sim alignment. This system enables policies trained in simulation to achieve strong zero-shot generalization to the real world. The high consistency between real-world and simulated policy performance underscores TwinAligner's potential to advance scalable robot learning. Code and data will be released on https://twin-aligner.github.io --- ### DSTED: Decoupling Temporal Stabilization and Discriminative Enhancement for Surgical Workflow Recognition Authors: Yueyao Chen, Kai-Ni Wang, Dario Tayupo, Arnaud Huaulm'e, Krystel Nyangoh Timoh, Pierre Jannin, Qi Dou Published: 2025-12-22 Categories: cs.CV, cs.AI arXiv: https://arxiv.org/abs/2512.19387v1 Purpose: Surgical workflow recognition enables context-aware assistance and skill assessment in computer-assisted interventions. Despite recent advances, current methods suffer from two critical challenges: prediction jitter across consecutive frames and poor discrimination of ambiguous phases. This paper aims to develop a stable framework by selectively propagating reliable historical information and explicitly modeling uncertainty for hard sample enhancement. Methods: We propose a dual-pathway framework DSTED with Reliable Memory Propagation (RMP) and Uncertainty-Aware Prototype Retrieval (UPR). RMP maintains temporal coherence by filtering and fusing high-confidence historical features through multi-criteria reliability assessment. UPR constructs learnable class-specific prototypes from high-uncertainty samples and performs adaptive prototype matching to refine ambiguous frame representations. Finally, a confidence-driven gate dynamically balances both pathways based on prediction certainty. Results: Our method achieves state-of-the-art performance on AutoLaparo-hysterectomy with 84.36% accuracy and 65.51% F1-score, surpassing the second-best method by 3.51% and 4.88% respectively. Ablations reveal complementary gains from RMP (2.19%) and UPR (1.93%), with synergistic effects when combined. Extensive analysis confirms substantial reduction in temporal jitter and marked improvement on challenging phase transitions. Conclusion: Our dual-pathway design introduces a novel paradigm for stable workflow recognition, demonstrating that decoupling the modeling of temporal consistency and phase ambiguity yields superior performance and clinical applicability. --- ### Real-Time Machine Learning for Embedded Anomaly Detection Authors: Abdelmadjid Benmachiche, Khadija Rais, Hamda Slimi Published: 2025-12-22 Categories: cs.LG arXiv: https://arxiv.org/abs/2512.19383v1 The spread of a resource-constrained Internet of Things (IoT) environment and embedded devices has put pressure on the real-time detection of anomalies occurring at the edge. This survey presents an overview of machine-learning methods aimed specifically at on-device anomaly detection with extremely strict constraints for latency, memory, and power consumption. Lightweight algorithms such as Isolation Forest, One-Class SVM, recurrent architectures, and statistical techniques are compared here according to the realities of embedded implementation. Our survey brings out significant trade-offs of accuracy and computational efficiency of detection, as well as how hardware constraints end up fundamentally redefining algorithm choice. The survey is completed with a set of practical recommendations on the choice of the algorithm depending on the equipment profiles and new trends in TinyML, which can help close the gap between detection capabilities and embedded reality. The paper serves as a strategic roadmap for engineers deploying anomaly detection in edge environments that are constrained by bandwidth and may be safety-critical. --- ### OmniMER: Auxiliary-Enhanced LLM Adaptation for Indonesian Multimodal Emotion Recognition Authors: Xueming Yan, Boyan Xu, Yaochu Jin, Lixian Xiao, Wenlong Ye, Runyang Cai, Zeqi Zheng, Jingfa Liu, Aimin Yang, Yongduan Song Published: 2025-12-22 Categories: cs.LG, cs.AI, cs.MM arXiv: https://arxiv.org/abs/2512.19379v3 Indonesian, spoken by over 200 million people, remains underserved in multimodal emotion recognition research despite its dominant presence on Southeast Asian social media platforms. We introduce IndoMER, the first multimodal emotion recognition benchmark for Indonesian, comprising 1,944 video segments from 203 speakers with temporally aligned text, audio, and visual annotations across seven emotion categories. The dataset exhibits realistic challenges including cross-modal inconsistency and long-tailed class distributions shaped by Indonesian cultural communication norms. To address these challenges, we propose OmniMER, a multimodal adaptation framework built upon Qwen2.5-Omni that enhances emotion recognition through three auxiliary modality-specific perception tasks: emotion keyword extraction for text, facial expression analysis for video, and prosody analysis for audio. These auxiliary tasks help the model identify emotion-relevant cues in each modality before fusion, reducing reliance on spurious correlations in low-resource settings. Experiments on IndoMER show that OmniMER achieves 0.582 Macro-F1 on sentiment classification and 0.454 on emotion recognition, outperforming the base model by 7.6 and 22.1 absolute points respectively. Cross-lingual evaluation on the Chinese CH-SIMS dataset further demonstrates the generalizability of the proposed framework. The dataset and code are publicly available. https://github.com/yanxm01/INDOMER --- ### HATS: High-Accuracy Triple-Set Watermarking for Large Language Models Authors: Zhiqing Hu, Chenxu Zhao, Jiazhong Lu, Xiaolei Liu Published: 2025-12-22 Categories: cs.CL arXiv: https://arxiv.org/abs/2512.19378v1 Misuse of LLM-generated text can be curbed by watermarking techniques that embed implicit signals into the output. We propose a watermark that partitions the vocabulary at each decoding step into three sets (Green/Yellow/Red) with fixed ratios and restricts sampling to the Green and Yellow sets. At detection time, we replay the same partitions, compute Green-enrichment and Red-depletion statistics, convert them to one-sided z-scores, and aggregate their p-values via Fisher's method to decide whether a passage is watermarked. We implement generation, detection, and testing on Llama 2 7B, and evaluate true-positive rate, false-positive rate, and text quality. Results show that the triple-partition scheme achieves high detection accuracy at fixed FPR while preserving readability. --- ### A Critical Assessment of Pattern Comparisons Between POD and Autoencoders in Intraventricular Flows Authors: Eneko Lazpita, Andrés Bell-Navas, Jesús Garicano-Mena, Petros Koumoutsakos, Soledad Le Clainche Published: 2025-12-22 Categories: physics.flu-dyn, cs.LG arXiv: https://arxiv.org/abs/2512.19376v1 Understanding intraventricular hemodynamics requires compact and physically interpretable representations of the underlying flow structures, as characteristic flow patterns are closely associated with cardiovascular conditions and can support early detection of cardiac deterioration. Conventional visualization of velocity or pressure fields, however, provides limited insight into the coherent mechanisms driving these dynamics. Reduced-order modeling techniques, like Proper Orthogonal Decomposition (POD) and Autoencoder (AE) architectures, offer powerful alternatives to extract dominant flow features from complex datasets. This study systematically compares POD with several AE variants (Linear, Nonlinear, Convolutional, and Variational) using left ventricular flow fields obtained from computational fluid dynamics simulations. We show that, for a suitably chosen latent dimension, AEs produce modes that become nearly orthogonal and qualitatively resemble POD modes that capture a given percentage of kinetic energy. As the number of latent modes increases, AE modes progressively lose orthogonality, leading to linear dependence, spatial redundancy, and the appearance of repeated modes with substantial high-frequency content. This degradation reduces interpretability and introduces noise-like components into AE-based reduced-order models, potentially complicating their integration with physics-based formulations or neural-network surrogates. The extent of interpretability loss varies across the AEs, with nonlinear, convolutional, and variational models exhibiting distinct behaviors in orthogonality preservation and feature localization. Overall, the results indicate that AEs can reproduce POD-like coherent structures under specific latent-space configurations, while highlighting the need for careful mode selection to ensure physically meaningful representations of cardiac flow dynamics. --- ### Cluster-Based Generalized Additive Models Informed by Random Fourier Features Authors: Xin Huang, Jia Li, Jun Yu Published: 2025-12-22 Categories: stat.ML, cs.LG arXiv: https://arxiv.org/abs/2512.19373v3 In developing data-driven modeling methodologies, there is an ongoing need to reconcile the strong predictive performance of opaque black-box models with the transparency required for critical applications. This work introduces an interpretable and computationally tractable regression framework for heterogeneous data by combining response-informed spectral representation learning with localized additive modeling. The method first fits a random Fourier feature regression model and constructs a spectral feature map from the learned amplitudes and adaptively resampled frequencies, so that the representation reflects predictive variation in the data. This representation is then compressed by principal component analysis to obtain a low-dimensional latent embedding, in which a Gaussian mixture model performs soft regime discovery. Within each regime, a cluster-specific generalized additive model captures nonlinear covariate effects through interpretable spline-based univariate smooth functions. The final predictor is formed as a soft mixture of these local additive models, enabling flexible modeling of a nonlinear, heterogeneous structure while preserving interpretability. Numerical experiments across several benchmark regression datasets show that the proposed method consistently improves upon classical globally interpretable baselines while remaining competitive with more flexible black-box models. Overall, the framework provides a unified approach to heterogeneous regression that combines predictive adaptivity with interpretable local covariate effects. --- ### Sprecher Networks: A Parameter-Efficient Kolmogorov-Arnold Architecture Authors: Christian Hägg, Kathlén Kohn, Giovanni Luca Marchetti, Boris Shapiro Published: 2025-12-22 Categories: cs.LG, cs.AI, math.NA arXiv: https://arxiv.org/abs/2512.19367v2 We introduce Sprecher Networks (SNs), a family of trainable architectures derived from David Sprecher's 1965 constructive form of the Kolmogorov-Arnold representation. Each SN block implements a "sum of shifted univariate functions" using only two shared learnable splines per block, a monotone inner spline $φ$ and a general outer spline $Φ$, together with a learnable shift parameter $η$ and a mixing vector $λ$ shared across all output dimensions. Stacking these blocks yields deep, compositional models; for vector-valued outputs we append an additional non-summed output block. We also propose an optional lateral mixing operator enabling intra-block communication between output channels with only $O(d_{\mathrm{out}})$ additional parameters. Owing to the vector (not matrix) mixing weights and spline sharing, SNs scale linearly in width, approximately $O(\sum_{\ell}(d_{\ell-1}+d_{\ell}+G))$ parameters for $G$ spline knots, versus $O(\sum_{\ell} d_{\ell-1}d_{\ell})$ for dense MLPs and $O(G\sum_{\ell} d_{\ell-1}d_{\ell})$ for edge-spline KANs. This linear width-scaling is particularly attractive for extremely wide, shallow models, where low depth can translate into low inference latency. Finally, we describe a sequential forward implementation that avoids materializing the $d_{\mathrm{in}}\times d_{\mathrm{out}}$ shifted-input tensor, reducing peak forward-intermediate memory from quadratic to linear in layer width, relevant for memory-constrained settings such as on-device/edge inference; we demonstrate deployability via fixed-point real-time digit classification on resource-constrained embedded device with only 4 MB RAM. We provide empirical demonstrations on supervised regression, Fashion-MNIST classification (including stable training at 25 hidden layers with residual connections and normalization), and a Poisson PINN, with controlled comparisons to MLP and KAN baselines. --- ### Learning General Policies with Policy Gradient Methods Authors: Simon Ståhlberg, Blai Bonet, Hector Geffner Published: 2025-12-22 Categories: cs.AI, cs.LG arXiv: https://arxiv.org/abs/2512.19366v1 While reinforcement learning methods have delivered remarkable results in a number of settings, generalization, i.e., the ability to produce policies that generalize in a reliable and systematic way, has remained a challenge. The problem of generalization has been addressed formally in classical planning where provable correct policies that generalize over all instances of a given domain have been learned using combinatorial methods. The aim of this work is to bring these two research threads together to illuminate the conditions under which (deep) reinforcement learning approaches, and in particular, policy optimization methods, can be used to learn policies that generalize like combinatorial methods do. We draw on lessons learned from previous combinatorial and deep learning approaches, and extend them in a convenient way. From the former, we model policies as state transition classifiers, as (ground) actions are not general and change from instance to instance. From the latter, we use graph neural networks (GNNs) adapted to deal with relational structures for representing value functions over planning states, and in our case, policies. With these ingredients in place, we find that actor-critic methods can be used to learn policies that generalize almost as well as those obtained using combinatorial approaches while avoiding the scalability bottleneck and the use of feature pools. Moreover, the limitations of the DRL methods on the benchmarks considered have little to do with deep learning or reinforcement learning algorithms, and result from the well-understood expressive limitations of GNNs, and the tradeoff between optimality and generalization (general policies cannot be optimal in some domains). Both of these limitations are addressed without changing the basic DRL methods by adding derived predicates and an alternative cost structure to optimize. --- ### Efficient Spike-driven Transformer for High-performance Drone-View Geo-Localization Authors: Zhongwei Chen, Hai-Jun Rong, Zhao-Xu Yang, Guoqi Li Published: 2025-12-22 Categories: cs.CV arXiv: https://arxiv.org/abs/2512.19365v2 Traditional drone-view geo-localization (DVGL) methods based on artificial neural networks (ANNs) have achieved remarkable performance. However, ANNs rely on dense computation, which results in high power consumption. In contrast, spiking neural networks (SNNs), which benefit from spike-driven computation, inherently provide low power consumption. Regrettably, the potential of SNNs for DVGL has yet to be thoroughly investigated. Meanwhile, the inherent sparsity of spike-driven computation for representation learning scenarios also results in loss of critical information and difficulties in learning long-range dependencies when aligning heterogeneous visual data sources. To address these, we propose SpikeViMFormer, the first SNN framework designed for DVGL. In this framework, a lightweight spike-driven transformer backbone is adopted to extract coarse-grained features. To mitigate the loss of critical information, the spike-driven selective attention (SSA) block is designed, which uses a spike-driven gating mechanism to achieve selective feature enhancement and highlight discriminative regions. Furthermore, a spike-driven hybrid state space (SHS) block is introduced to learn long-range dependencies using a hybrid state space. Moreover, only the backbone is utilized during the inference stage to reduce computational cost. To ensure backbone effectiveness, a novel hierarchical re-ranking alignment learning (HRAL) strategy is proposed. It refines features via neighborhood re-ranking and maintains cross-batch consistency to directly optimize the backbone. Experimental results demonstrate that SpikeViMFormer outperforms state-of-the-art SNNs. Compared with advanced ANNs, it also achieves competitive performance.Our code is available at https://github.com/ISChenawei/SpikeViMFormer --- ### From Points to Coalitions: Hierarchical Contrastive Shapley Values for Prioritizing Data Samples Authors: Canran Xiao, Jiabao Dou, Zhiming Lin, Zong Ke, Liwei Hou Published: 2025-12-22 Categories: cs.LG arXiv: https://arxiv.org/abs/2512.19363v1 How should we quantify the value of each training example when datasets are large, heterogeneous, and geometrically structured? Classical Data-Shapley answers in principle, but its O(n!) complexity and point-wise perspective are ill-suited to modern scales. We propose Hierarchical Contrastive Data Valuation (HCDV), a three-stage framework that (i) learns a contrastive, geometry-preserving representation, (ii) organizes the data into a balanced coarse-to-fine hierarchy of clusters, and (iii) assigns Shapley-style payoffs to coalitions via local Monte-Carlo games whose budgets are propagated downward. HCDV collapses the factorial burden to O(T sum_{l} K_{l}) = O(T K_max log n), rewards examples that sharpen decision boundaries, and regularizes outliers through curvature-based smoothness. We prove that HCDV approximately satisfies the four Shapley axioms with surplus loss O(eta log n), enjoys sub-Gaussian coalition deviation tilde O(1/sqrt{T}), and incurs at most k epsilon_infty regret for top-k selection. Experiments on four benchmarks--tabular, vision, streaming, and a 45M-sample CTR task--plus the OpenDataVal suite show that HCDV lifts accuracy by up to +5 pp, slashes valuation time by up to 100x, and directly supports tasks such as augmentation filtering, low-latency streaming updates, and fair marketplace payouts. --- ### Interpretable Hybrid Deep Q-Learning Framework for IoT-Based Food Spoilage Prediction with Synthetic Data Generation and Hardware Validation Authors: Isshaan Singh, Divyansh Chawla, Anshu Garg, Shivin Mangal, Pallavi Gupta, Khushi Agarwal, Nimrat Singh Khalsa, Nandan Patel Published: 2025-12-22 Categories: cs.LG arXiv: https://arxiv.org/abs/2512.19361v1 The need for an intelligent, real-time spoilage prediction system has become critical in modern IoT-driven food supply chains, where perishable goods are highly susceptible to environmental conditions. Existing methods often lack adaptability to dynamic conditions and fail to optimize decision making in real time. To address these challenges, we propose a hybrid reinforcement learning framework integrating Long Short-Term Memory (LSTM) and Recurrent Neural Networks (RNN) for enhanced spoilage prediction. This hybrid architecture captures temporal dependencies within sensor data, enabling robust and adaptive decision making. In alignment with interpretable artificial intelligence principles, a rule-based classifier environment is employed to provide transparent ground truth labeling of spoilage levels based on domain-specific thresholds. This structured design allows the agent to operate within clearly defined semantic boundaries, supporting traceable and interpretable decisions. Model behavior is monitored using interpretability-driven metrics, including spoilage accuracy, reward-to-step ratio, loss reduction rate, and exploration decay. These metrics provide both quantitative performance evaluation and insights into learning dynamics. A class-wise spoilage distribution visualization is used to analyze the agents decision profile and policy behavior. Extensive evaluations on simulated and real-time hardware data demonstrate that the LSTM and RNN based agent outperforms alternative reinforcement learning approaches in prediction accuracy and decision efficiency while maintaining interpretability. The results highlight the potential of hybrid deep reinforcement learning with integrated interpretability for scalable IoT-based food monitoring systems. --- ### First-Order Representation Languages for Goal-Conditioned RL Authors: Simon Ståhlberg, Hector Geffner Published: 2025-12-22 Categories: cs.AI arXiv: https://arxiv.org/abs/2512.19355v1 First-order relational languages have been used in MDP planning and reinforcement learning (RL) for two main purposes: specifying MDPs in compact form, and representing and learning policies that are general and not tied to specific instances or state spaces. In this work, we instead consider the use of first-order languages in goal-conditioned RL and generalized planning. The question is how to learn goal-conditioned and general policies when the training instances are large and the goal cannot be reached by random exploration alone. The technique of Hindsight Experience Replay (HER) provides an answer to this question: it relabels unsuccessful trajectories as successful ones by replacing the original goal with one that was actually achieved. If the target policy must generalize across states and goals, trajectories that do not reach the original goal states can enable more data- and time-efficient learning. In this work, we show that further performance gains can be achieved when states and goals are represented by sets of atoms. We consider three versions: goals as full states, goals as subsets of the original goals, and goals as lifted versions of these subgoals. The result is that the latter two successfully learn general policies on large planning instances with sparse rewards by automatically creating a curriculum of easier goals of increasing complexity. The experiments illustrate the computational gains of these versions, their limitations, and opportunities for addressing them. --- ### ReasonCD: A Multimodal Reasoning Large Model for Implicit Change-of-Interest Semantic Mining Authors: Zhenyang Huang, Xiao Yu, Yi Zhang, Decheng Wang, Hang Ruan Published: 2025-12-22 Categories: cs.CV arXiv: https://arxiv.org/abs/2512.19354v1 Remote sensing image change detection is one of the fundamental tasks in remote sensing intelligent interpretation. Its core objective is to identify changes within change regions of interest (CRoI). Current multimodal large models encode rich human semantic knowledge, which is utilized for guidance in tasks such as remote sensing change detection. However, existing methods that use semantic guidance for detecting users' CRoI overly rely on explicit textual descriptions of CRoI, leading to the problem of near-complete performance failure when presented with implicit CRoI textual descriptions. This paper proposes a multimodal reasoning change detection model named ReasonCD, capable of mining users' implicit task intent. The model leverages the powerful reasoning capabilities of pre-trained large language models to mine users' implicit task intents and subsequently obtains different change detection results based on these intents. Experiments on public datasets demonstrate that the model achieves excellent change detection performance, with an F1 score of 92.1\% on the BCDD dataset. Furthermore, to validate its superior reasoning functionality, this paper annotates a subset of reasoning data based on the SECOND dataset. Experimental results show that the model not only excels at basic reasoning-based change detection tasks but can also explain the reasoning process to aid human decision-making. --- ### PENDULUM: A Benchmark for Assessing Sycophancy in Multimodal Large Language Models Authors: A. B. M. Ashikur Rahman, Saeed Anwar, Muhammad Usman, Irfan Ahmad, Ajmal Mian Published: 2025-12-22 Categories: cs.AI arXiv: https://arxiv.org/abs/2512.19350v1 Sycophancy, an excessive tendency of AI models to agree with user input at the expense of factual accuracy or in contradiction of visual evidence, poses a critical and underexplored challenge for multimodal large language models (MLLMs). While prior studies have examined this behavior in text-only settings of large language models, existing research on visual or multimodal counterparts remains limited in scope and depth of analysis. To address this gap, we introduce a comprehensive evaluation benchmark, \textit{PENDULUM}, comprising approximately 2,000 human-curated Visual Question Answering pairs specifically designed to elicit sycophantic responses. The benchmark spans six distinct image domains of varying complexity, enabling a systematic investigation of how image type and inherent challenges influence sycophantic tendencies. Through extensive evaluation of state-of-the-art MLLMs. we observe substantial variability in model robustness and a pronounced susceptibility to sycophantic and hallucinatory behavior. Furthermore, we propose novel metrics to quantify sycophancy in visual reasoning, offering deeper insights into its manifestations across different multimodal contexts. Our findings highlight the urgent need for developing sycophancy-resilient architectures and training strategies to enhance factual consistency and reliability in future MLLMs. Our proposed dataset with MLLMs response are available at https://github.com/ashikiut/pendulum/. --- ### VIGOR+: Iterative Confounder Generation and Validation via LLM-CEVAE Feedback Loop Authors: JiaWei Zhu, ZiHeng Liu Published: 2025-12-22 Categories: cs.AI, cs.LG arXiv: https://arxiv.org/abs/2512.19349v1 Hidden confounding remains a fundamental challenge in causal inference from observational data. Recent advances leverage Large Language Models (LLMs) to generate plausible hidden confounders based on domain knowledge, yet a critical gap exists: LLM-generated confounders often exhibit semantic plausibility without statistical utility. We propose VIGOR+ (Variational Information Gain for iterative cOnfounder Refinement), a novel framework that closes the loop between LLM-based confounder generation and CEVAE-based statistical validation. Unlike prior approaches that treat generation and validation as separate stages, VIGOR+ establishes an iterative feedback mechanism: validation signals from CEVAE (including information gain, latent consistency metrics, and diagnostic messages) are transformed into natural language feedback that guides subsequent LLM generation rounds. This iterative refinement continues until convergence criteria are met. We formalize the feedback mechanism, prove convergence properties under mild assumptions, and provide a complete algorithmic framework. --- ### Faster Distributed Inference-Only Recommender Systems via Bounded Lag Synchronous Collectives Authors: Kiril Dichev, Filip Pawlowski, Albert-Jan Yzelman Published: 2025-12-22 Categories: cs.DC, cs.LG arXiv: https://arxiv.org/abs/2512.19342v1 Recommender systems are enablers of personalized content delivery, and therefore revenue, for many large companies. In the last decade, deep learning recommender models (DLRMs) are the de-facto standard in this field. The main bottleneck in DLRM inference is the lookup of sparse features across huge embedding tables, which are usually partitioned across the aggregate RAM of many nodes. In state-of-the-art recommender systems, the distributed lookup is implemented via irregular all-to-all (alltoallv) communication, and often presents the main bottleneck. Today, most related work sees this operation as a given; in addition, every collective is synchronous in nature. In this work, we propose a novel bounded lag synchronous (BLS) version of the alltoallv operation. The bound can be a parameter allowing slower processes to lag behind entire iterations before the fastest processes block. In special applications such as inference-only DLRM, the accuracy of the application is fully preserved. We implement BLS alltoallv in a new PyTorch Distributed backend and evaluate it with a BLS version of the reference DLRM code. We show that for well balanced, homogeneous-access DLRM runs our BLS technique does not offer notable advantages. But for unbalanced runs, e.g. runs with strongly irregular embedding table accesses or with delays across different processes, our BLS technique improves both the latency and throughput of inference-only DLRM. In the best-case scenario, the proposed reduced synchronisation can mask the delays across processes altogether. --- ### GANeXt: A Fully ConvNeXt-Enhanced Generative Adversarial Network for MRI- and CBCT-to-CT Synthesis Authors: Siyuan Mei, Yan Xia, Fuxin Fan, Andreas Maier Published: 2025-12-22 Categories: cs.CV arXiv: https://arxiv.org/abs/2512.19336v2 The synthesis of computed tomography (CT) from magnetic resonance imaging (MRI) and cone-beam CT (CBCT) plays a critical role in clinical treatment planning by enabling accurate anatomical representation in adaptive radiotherapy. In this work, we propose GANeXt, a 3D patch-based, fully ConvNeXt-powered generative adversarial network for unified CT synthesis across different modalities and anatomical regions. Specifically, GANeXt employs an efficient U-shaped generator constructed from stacked 3D ConvNeXt blocks with compact convolution kernels, while the discriminator adopts a conditional PatchGAN. To improve synthesis quality, we incorporate a combination of loss functions, including mean absolute error (MAE), perceptual loss, segmentation-based masked MAE, and adversarial loss and a combination of Dice loss and cross-entropy for multi-head segmentation discriminator. For both tasks, training is performed with a batch size of 8 using two separate AdamW optimizers for the generator and discriminator, each equipped with a warmup and cosine decay scheduler, with learning rates of $5\times10^{-4}$ and $1\times10^{-3}$, respectively. Data preprocessing includes deformable registration, foreground cropping, percentile normalization for the input modality, and linear normalization of the CT to the range $(-1024, 1000)$. Data augmentation involves random zooming within $(0.8, 1.3)$ (for MRI-to-CT only), fixed-size cropping to $32\times160\times192$ for MRI-to-CT and $32\times128\times128$ for CBCT-to-CT, and random flipping. During inference, we apply a sliding-window approach with $0.8$ overlap and average folding to reconstruct the full-size sCT, followed by inversion of the CT normalization. After joint training on all regions without any fine-tuning, the final models are selected at the end of 3000 epochs for MRI-to-CT and 1000 epochs for CBCT-to-CT using the full training dataset. --- ### Orthogonal Approximate Message Passing with Optimal Spectral Initializations for Rectangular Spiked Matrix Models Authors: Haohua Chen, Songbin Liu, Junjie Ma Published: 2025-12-22 Categories: cs.IT, cs.LG, math.ST arXiv: https://arxiv.org/abs/2512.19334v1 We propose an orthogonal approximate message passing (OAMP) algorithm for signal estimation in the rectangular spiked matrix model with general rotationally invariant (RI) noise. We establish a rigorous state evolution that precisely characterizes the algorithm's high-dimensional dynamics and enables the construction of iteration-wise optimal denoisers. Within this framework, we accommodate spectral initializations under minimal assumptions on the empirical noise spectrum. In the rectangular setting, where a single rank-one component typically generates multiple informative outliers, we further propose a procedure for combining these outliers under mild non-Gaussian signal assumptions. For general RI noise models, the predicted performance of the proposed optimal OAMP algorithm agrees with replica-symmetric predictions for the associated Bayes-optimal estimator, and we conjecture that it is statistically optimal within a broad class of iterative estimation methods. --- ### A Logical View of GNN-Style Computation and the Role of Activation Functions Authors: Pablo Barceló, Floris Geerts, Matthias Lanzinger, Klara Pakhomenko, Jan Van den Bussche Published: 2025-12-22 Categories: cs.LG, cs.LO arXiv: https://arxiv.org/abs/2512.19332v2 We study the numerical and Boolean expressiveness of MPLang, a declarative language that captures the computation of graph neural networks (GNNs) through linear message passing and activation functions. We begin with A-MPLang, the fragment without activation functions, and give a characterization of its expressive power in terms of walk-summed features. For bounded activation functions, we show that (under mild conditions) all eventually constant activations yield the same expressive power - numerical and Boolean - and that it subsumes previously established logics for GNNs with eventually constant activation functions but without linear layers. Finally, we prove the first expressive separation between unbounded and bounded activations in the presence of linear layers: MPLang with ReLU is strictly more powerful for numerical queries than MPLang with eventually constant activation functions, e.g., truncated ReLU. This hinges on subtle interactions between linear aggregation and eventually constant non-linearities, and it establishes that GNNs using ReLU are more expressive than those restricted to eventually constant activations and linear layers. --- ### DeltaMIL: Gated Memory Integration for Efficient and Discriminative Whole Slide Image Analysis Authors: Yueting Zhu, Yuehao Song, Shuai Zhang, Wenyu Liu, Xinggang Wang Published: 2025-12-22 Categories: cs.CV arXiv: https://arxiv.org/abs/2512.19331v1 Whole Slide Images (WSIs) are typically analyzed using multiple instance learning (MIL) methods. However, the scale and heterogeneity of WSIs generate highly redundant and dispersed information, making it difficult to identify and integrate discriminative signals. Existing MIL methods either fail to discard uninformative cues effectively or have limited ability to consolidate relevant features from multiple patches, which restricts their performance on large and heterogeneous WSIs. To address this issue, we propose DeltaMIL, a novel MIL framework that explicitly selects semantically relevant regions and integrates the discriminative information from WSIs. Our method leverages the gated delta rule to efficiently filter and integrate information through a block combining forgetting and memory mechanisms. The delta mechanism dynamically updates the memory by removing old values and inserting new ones according to their correlation with the current patch. The gating mechanism further enables rapid forgetting of irrelevant signals. Additionally, DeltaMIL integrates a complementary local pattern mixing mechanism to retain fine-grained pathological locality. Our design enhances the extraction of meaningful cues and suppresses redundant or noisy information, which improves the model's robustness and discriminative power. Experiments demonstrate that DeltaMIL achieves state-of-the-art performance. Specifically, for survival prediction, DeltaMIL improves performance by 3.69\% using ResNet-50 features and 2.36\% using UNI features. For slide-level classification, it increases accuracy by 3.09\% with ResNet-50 features and 3.75\% with UNI features. These results demonstrate the strong and consistent performance of DeltaMIL across diverse WSI tasks. --- ### Extended OpenTT Games Dataset: A table tennis dataset for fine-grained shot type and point outcome Authors: Moamal Fadhil Abdul-Mahdi, Jonas Bruun Hubrechts, Thomas Martini Jørgensen, Emil Hovad Published: 2025-12-22 Categories: cs.CV arXiv: https://arxiv.org/abs/2512.19327v2 Automatically detecting and classifying strokes in table tennis video can streamline training workflows, enrich broadcast overlays, and enable fine-grained performance analytics. For this to be possible, annotated video data of table tennis is needed. We extend the public OpenTTGames dataset with highly detailed, frame-accurate shot type annotations (forehand, backhand with subtypes), player posture labels (body lean and leg stance), and rally outcome tags at point end. OpenTTGames is a set of recordings from the side of the table with official labels for bounces, when the ball is above the net, or hitting the net. The dataset already contains ball coordinates near events, which are either "bounce", "net", or "empty_event" in the original OpenTTGames dataset, and semantic masks (humans, table, scoreboard). Our extension adds the types of stroke to the events and a per-player taxonomy so models can move beyond event spotting toward tactical understanding (e.g., whether a stroke is likely to win the point or set up an advantage). We provide a compact coding scheme and code-assisted labeling procedure to support reproducible annotations and baselines for fine-grained stroke understanding in racket sports. This fills a practical gap in the community, where many prior video resources are either not publicly released or carry restrictive/unclear licenses that hinder reuse and benchmarking. Our annotations are released under the same CC BY-NC-SA 4.0 license as OpenTTGames, allowing free non-commercial use, modification, and redistribution, with appropriate attribution. --- ### Alternative positional encoding functions for neural transformers Authors: Ezequiel Lopez-Rubio, Macoris Decena-Gimenez, Rafael Marcos Luque-Baena Published: 2025-12-22 Categories: cs.LG, cs.AI arXiv: https://arxiv.org/abs/2512.19323v1 A key module in neural transformer-based deep architectures is positional encoding. This module enables a suitable way to encode positional information as input for transformer neural layers. This success has been rooted in the use of sinusoidal functions of various frequencies, in order to capture recurrent patterns of differing typical periods. In this work, an alternative set of periodic functions is proposed for positional encoding. These functions preserve some key properties of sinusoidal ones, while they depart from them in fundamental ways. Some tentative experiments are reported, where the original sinusoidal version is substantially outperformed. This strongly suggests that the alternative functions may have a wider use in other transformer architectures. --- ### Learning to Sample in Variable Neighborhood Search Algorithm for Urban Cable Routing Optimization Authors: Wei Liu, Rui Wang, Chenhui Lin, Kaiwen Li, Wenhua Li, Tao Zhang Published: 2025-12-22 Categories: cs.NE arXiv: https://arxiv.org/abs/2512.19321v2 Urban underground cable construction is essential for enhancing power grid reliability, yet the high construction costs demand systematic optimization. Constrained by road network infrastructure, this problem requires consideration of both connectivity relationships between substations and specific routing strategies along road networks, constituting a large-scale bilevel combinatorial optimization problem. Insufficient attention to routing subproblems in traditional research and simplistic algorithmic designs ill-equipped for large-scale optimization leave substantial room for advancement. To navigate the enormous combinatorial search space, we propose a learning-assisted variable neighborhood search (L-VNS) algorithm integrating four key components. First, an auxiliary task focusing on the upper-level connectivity subproblem generates high-quality initial solutions by employing hybrid genetic search for connection optimization and A* for detailed path routing. Subsequently, the algorithm iteratively refines the connectivity topology using variable neighborhood search with three complementary operators. A multi-agent deep reinforcement learning module adaptively guides probabilistic neighborhood sampling by jointly encoding upper-level connectivity patterns and lower-level routing structures, effectively exploiting problem structure. Finally, a modified A* operator re-plans lower-level paths affected by neighborhood modifications to ensure feasibility and solution completeness. Comprehensive experiments on 12 benchmark instances and 3 GIS-derived instances demonstrate the superiority of L-VNS, achieving total construction cost reductions of 0.92% to 73.72% compared to representative approaches. Ablation studies and sensitivity analyses further validate the effectiveness and robustness of the proposed algorithm. --- ### MAGIC: Achieving Superior Model Merging via Magnitude Calibration Authors: Yayuan Li, Jian Zhang, Jintao Guo, Zihan Cheng, Lei Qi, Yinghuan Shi, Yang Gao Published: 2025-12-22 Categories: cs.LG, cs.AI, cs.CL, cs.CV arXiv: https://arxiv.org/abs/2512.19320v1 The proliferation of pre-trained models has given rise to a wide array of specialised, fine-tuned models. Model merging aims to merge the distinct capabilities of these specialised models into a unified model, requiring minimal or even no additional training. A core objective of model merging is to ensure the merged model retains the behavioural characteristics of the specialised models, typically achieved through feature alignment. We identify that features consist of two critical components: direction and magnitude. Prior research has predominantly focused on directional alignment, while the influence of magnitude remains largely neglected, despite its pronounced vulnerability to perturbations introduced by common merging operations (e.g., parameter fusion and sparsification). Such perturbations to magnitude inevitably lead to feature deviations in the merged model from the specialised models, resulting in subsequent performance degradation. To address this, we propose MAGnItude Calibration (MAGIC), a plug-and-play framework that rectifies layer-wise magnitudes in feature and weight spaces, with three variants. Specifically, our Feature Space Calibration (FSC) realigns the merged model's features using a small set of unlabelled data, while Weight Space Calibration (WSC) extends this calibration to the weight space without requiring additional data. Combining these yields Dual Space Calibration (DSC). Comprehensive experiments demonstrate that MAGIC consistently boosts performance across diverse Computer Vision tasks (+4.3% on eight datasets) and NLP tasks (+8.0% on Llama) without additional training. Our code is available at: https://github.com/lyymuwu/MAGIC --- ### SafeMed-R1: Adversarial Reinforcement Learning for Generalizable and Robust Medical Reasoning in Vision-Language Models Authors: A. A. Gde Yogi Pramana, Jason Ray, Anthony Jaya, Michael Wijaya Published: 2025-12-22 Categories: cs.AI arXiv: https://arxiv.org/abs/2512.19317v1 Vision--Language Models (VLMs) show significant promise for Medical Visual Question Answering (VQA), yet their deployment in clinical settings is hindered by severe vulnerability to adversarial attacks. Standard adversarial training, while effective for simpler tasks, often degrades both generalization performance and the quality of generated clinical reasoning. We introduce SafeMed-R1, a hybrid defense framework that ensures robust performance while preserving high-quality, interpretable medical reasoning. SafeMed-R1 employs a two-stage approach: at training time, we integrate Adversarial Training with Group Relative Policy Optimization (AT-GRPO) to explicitly robustify the reasoning process against worst-case perturbations; at inference time, we augment the model with Randomized Smoothing to provide certified $L_2$-norm robustness guarantees. We evaluate SafeMed-R1 on the OmniMedVQA benchmark across eight medical imaging modalities comprising over 88,000 samples. Our experiments reveal that standard fine-tuned VLMs, despite achieving 95\% accuracy on clean inputs, collapse to approximately 25\% under PGD attacks. In contrast, SafeMed-R1 maintains 84.45\% accuracy under the same adversarial conditions, representing a 59 percentage point improvement in robustness. Furthermore, we demonstrate that models trained with explicit chain-of-thought reasoning exhibit superior adversarial robustness compared to instruction-only variants, suggesting a synergy between interpretability and security in medical AI systems. --- ### Neural Implicit Heart Coordinates: 3D cardiac shape reconstruction from sparse segmentations Authors: Marica Muffoletto, Uxio Hermida, Charlène Mauger, Avan Suinesiaputra, Yiyang Xu, Richard Burns, Lisa Pankewitz, Andrew D McCulloch, Steffen E Petersen, Daniel Rueckert, Alistair A Young Published: 2025-12-22 Categories: cs.CV, eess.IV arXiv: https://arxiv.org/abs/2512.19316v2 Accurate reconstruction of cardiac anatomy from sparse clinical images remains a major challenge in patient-specific modeling. While neural implicit functions have previously been applied to this task, their application to mapping anatomical consistency across subjects has been limited. In this work, we introduce Neural Implicit Heart Coordinates (NIHCs), a standardized implicit coordinate system, based on universal ventricular coordinates, that provides a common anatomical reference frame for the human heart. Our method predicts NIHCs directly from a limited number of 2D segmentations (sparse acquisition) and subsequently decodes them into dense 3D segmentations and high-resolution meshes at arbitrary output resolution. Trained on a large dataset of 5,000 cardiac meshes, the model achieves high reconstruction accuracy on clinical contours, with mean Euclidean surface errors of 2.51$\pm$0.33 mm in a diseased cohort (n=4549) and 2.3$\pm$0.36 mm in a healthy cohort (n=5576). The NIHC representation enables anatomically coherent reconstruction even under severe slice sparsity and segmentation noise, faithfully recovering complex structures such as the valve planes. Compared with traditional pipelines, inference time is reduced from over 60 s to 5-15 s. These results demonstrate that NIHCs constitute a robust and efficient anatomical representation for patient-specific 3D cardiac reconstruction from minimal input data. --- ### MixFlow Training: Alleviating Exposure Bias with Slowed Interpolation Mixture Authors: Hui Li, Fu-Yun Wang, Haoyuan Xia, Jiayue Lyu, Kaihui Cheng, Siyu Zhu, Jingdong Wang Published: 2025-12-22 Categories: cs.CV, cs.AI arXiv: https://arxiv.org/abs/2512.19311v2 This paper studies the training-testing discrepancy (a.k.a. exposure bias) problem for improving the diffusion models. During training, the input of a prediction network at one training timestep is the corresponding ground-truth noisy data that is an interpolation of the noise and the data, and during testing, the input is the generated noisy data. We present a novel training approach, named MixFlow, for improving the performance. Our approach is motivated by the Slow Flow phenomenon: the ground-truth interpolation that is the nearest to the generated noisy data at a given sampling timestep is observed to correspond to a higher-noise timestep (termed slowed timestep), i.e., the corresponding ground-truth timestep is slower than the sampling timestep. MixFlow leverages the interpolations at the slowed timesteps, named slowed interpolation mixture, for post-training the prediction network for each training timestep. Experiments over class-conditional image generation (including SiT, REPA, and RAE) and text-to-image generation validate the effectiveness of our approach. Our approach MixFlow over the RAE models achieve strong generation results on ImageNet: 1.43 FID (without guidance) and 1.10 (with guidance) at 256 x 256, and 1.55 FID (without guidance) and 1.10 (with guidance) at 512 x 512. --- ### Time-Vertex Machine Learning for Optimal Sensor Placement in Temporal Graph Signals: Applications in Structural Health Monitoring Authors: Keivan Faghih Niresi, Jun Qing, Mengjie Zhao, Olga Fink Published: 2025-12-22 Categories: cs.LG, eess.SP arXiv: https://arxiv.org/abs/2512.19309v1 Structural Health Monitoring (SHM) plays a crucial role in maintaining the safety and resilience of infrastructure. As sensor networks grow in scale and complexity, identifying the most informative sensors becomes essential to reduce deployment costs without compromising monitoring quality. While Graph Signal Processing (GSP) has shown promise by leveraging spatial correlations among sensor nodes, conventional approaches often overlook the temporal dynamics of structural behavior. To overcome this limitation, we propose Time-Vertex Machine Learning (TVML), a novel framework that integrates GSP, time-domain analysis, and machine learning to enable interpretable and efficient sensor placement by identifying representative nodes that minimize redundancy while preserving critical information. We evaluate the proposed approach on two bridge datasets for damage detection and time-varying graph signal reconstruction tasks. The results demonstrate the effectiveness of our approach in enhancing SHM systems by providing a robust, adaptive, and efficient solution for sensor placement. --- ### CienaLLM: Generative Climate-Impact Extraction from News Articles with Autoregressive LLMs Authors: Javier Vela-Tambo, Jorge Gracia, Fernando Dominguez-Castro Published: 2025-12-22 Categories: cs.CL arXiv: https://arxiv.org/abs/2512.19305v1 Understanding and monitoring the socio-economic impacts of climate hazards requires extracting structured information from heterogeneous news articles on a large scale. To that end, we have developed CienaLLM, a modular framework based on schema-guided Generative Information Extraction. CienaLLM uses open-weight Large Language Models for zero-shot information extraction from news articles, and supports configurable prompts and output schemas, multi-step pipelines, and cloud or on-premise inference. To systematically assess how the choice of LLM family, size, precision regime, and prompting strategy affect performance, we run a large factorial study in models, precisions, and prompt engineering techniques. An additional response parsing step nearly eliminates format errors while preserving accuracy; larger models deliver the strongest and most stable performance, while quantization offers substantial efficiency gains with modest accuracy trade-offs; and prompt strategies show heterogeneous, model-specific effects. CienaLLM matches or outperforms the supervised baseline in accuracy for extracting drought impacts from Spanish news, although at a higher inference cost. While evaluated in droughts, the schema-driven and model-agnostic design is suitable for adapting to related information extraction tasks (e.g., other hazards, sectors, or languages) by editing prompts and schemas rather than retraining. We release code, configurations, and schemas to support reproducible use. --- ### Bridging Semantics and Geometry: A Decoupled LVLM-SAM Framework for Reasoning Segmentation in Optical Remote Sensing Authors: Xu Zhang, Junyao Ge, Yang Zheng, Kaitai Guo, Jimin Liang Published: 2025-12-22 Categories: cs.CV arXiv: https://arxiv.org/abs/2512.19302v2 Large Vision--Language Models (LVLMs) hold great promise for advancing optical remote sensing (RS) analysis, yet existing reasoning segmentation frameworks couple linguistic reasoning and pixel prediction through end-to-end supervised fine-tuning, leading to weak geometric grounding and limited generalization across tasks. To address this, we developed Think2Seg-RS, a decoupled framework that trains an LVLM prompter to control a frozen Segment Anything Model (SAM) via structured geometric prompts. Through a mask-only Group Relative Policy Optimization (GRPO) reinforcement learning objective driven strictly by final mask IoU, the LVLM learns to translate abstract semantic reasoning into spatially grounded actions, achieving state-of-the-art performance on the EarthReason dataset. Notably, Think2Seg-RS outperforms leading approaches such as RemoteReasoner and SegEarth-R1 on the EarthReason dataset by reaching a test cIoU of 75.60% and gIoU of 73.36%, yielding absolute improvements of 6.47% and 2.40% over the strongest baseline, respectively. Zero-shot evaluations across three referring segmentation benchmarks reveal a fundamental distinction in task inductive bias, exposing a distinct divide between semantic-level grounding -- which aggregates all regions matching a conceptual intent -- and instance-level tasks that demand discrete object separation. We further found that compact segmenters outperform larger ones under semantic-level supervision by mitigating textural over-segmentation, and that unconstrained negative prompting is unstable in heterogeneous aerial backgrounds. Together, these findings demonstrate that optimizing LVLMs through direct segmentation feedback offers a scalable framework for complex geospatial reasoning, effectively bridging the gap between abstract language understanding and precise pixel-level execution. --- ### RMLer: Synthesizing Novel Objects across Diverse Categories via Reinforcement Mixing Learning Authors: Jun Li, Zikun Chen, Haibo Chen, Shuo Chen, Jian Yang Published: 2025-12-22 Categories: cs.CV arXiv: https://arxiv.org/abs/2512.19300v1 Novel object synthesis by integrating distinct textual concepts from diverse categories remains a significant challenge in Text-to-Image (T2I) generation. Existing methods often suffer from insufficient concept mixing, lack of rigorous evaluation, and suboptimal outputs-manifesting as conceptual imbalance, superficial combinations, or mere juxtapositions. To address these limitations, we propose Reinforcement Mixing Learning (RMLer), a framework that formulates cross-category concept fusion as a reinforcement learning problem: mixed features serve as states, mixing strategies as actions, and visual outcomes as rewards. Specifically, we design an MLP-policy network to predict dynamic coefficients for blending cross-category text embeddings. We further introduce visual rewards based on (1) semantic similarity and (2) compositional balance between the fused object and its constituent concepts, optimizing the policy via proximal policy optimization. At inference, a selection strategy leverages these rewards to curate the highest-quality fused objects. Extensive experiments demonstrate RMLer's superiority in synthesizing coherent, high-fidelity objects from diverse categories, outperforming existing methods. Our work provides a robust framework for generating novel visual concepts, with promising applications in film, gaming, and design. ---