# Research Papers — 2025-12 · bundle page 169 of 175 > Full-text markdown of papers #8401–8450 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/169 [← Previous bundle](https://aipod.dev/papers/archive/2025-12/page/168.md) [Next bundle →](https://aipod.dev/papers/archive/2025-12/page/170.md) ### AlignSAE: Concept-Aligned Sparse Autoencoders Authors: Minglai Yang, Xinyu Guo, Zhengliang Shi, Jinhe Bi, Steven Bethard, Mihai Surdeanu, Liangming Pan Published: 2025-12-01 Categories: cs.LG, cs.CL arXiv: https://arxiv.org/abs/2512.02004v4 Large Language Models (LLMs) encode factual knowledge within hidden parametric spaces that are difficult to inspect or control. While Sparse Autoencoders (SAEs) can decompose hidden activations into more fine-grained, interpretable features, they often struggle to reliably align these features with human-defined concepts, resulting in entangled and distributed feature representations. To address this, we introduce AlignSAE, a method that aligns SAE features with a predefined ontology through a "pre-train, then post-train" curriculum. After an initial unsupervised training phase, we apply supervised post-training to bind specific concepts to dedicated latent slots while preserving the remaining capacity for general reconstruction. This separation creates an interpretable interface where specific concepts can be inspected and controlled without interference from unrelated features. Empirical results on GPT-2 across synthetic biographical QA (1-hop, 6 concepts) and compositional relational reasoning (2-hop, 20 concepts) demonstrate that AlignSAE enables precise causal interventions, such as reliable concept swaps (94.8% success in 1-hop), by targeting single, semantically aligned slots. AlignSAE achieves disentanglement improvements of $9$--$\infty\times$ over standard SAEs (measured via Mutual Information Gap), and further supports multi-hop reasoning and a mechanistic probe of grokking-like generalization dynamics. Although our evaluation focuses on GPT-2, AlignSAE itself is model-agnostic. --- ### Learning Sim-to-Real Humanoid Locomotion in 15 Minutes Authors: Younggyo Seo, Carmelo Sferrazza, Juyue Chen, Guanya Shi, Rocky Duan, Pieter Abbeel Published: 2025-12-01 Categories: cs.RO, cs.AI, cs.LG arXiv: https://arxiv.org/abs/2512.01996v1 Massively parallel simulation has reduced reinforcement learning (RL) training time for robots from days to minutes. However, achieving fast and reliable sim-to-real RL for humanoid control remains difficult due to the challenges introduced by factors such as high dimensionality and domain randomization. In this work, we introduce a simple and practical recipe based on off-policy RL algorithms, i.e., FastSAC and FastTD3, that enables rapid training of humanoid locomotion policies in just 15 minutes with a single RTX 4090 GPU. Our simple recipe stabilizes off-policy RL algorithms at massive scale with thousands of parallel environments through carefully tuned design choices and minimalist reward functions. We demonstrate rapid end-to-end learning of humanoid locomotion controllers on Unitree G1 and Booster T1 robots under strong domain randomization, e.g., randomized dynamics, rough terrain, and push perturbations, as well as fast training of whole-body human-motion tracking policies. We provide videos and open-source implementation at: https://younggyo.me/fastsac-humanoid. --- ### RoaD: Rollouts as Demonstrations for Closed-Loop Supervised Fine-Tuning of Autonomous Driving Policies Authors: Guillermo Garcia-Cobo, Maximilian Igl, Peter Karkus, Zhejun Zhang, Michael Watson, Yuxiao Chen, Boris Ivanovic, Marco Pavone Published: 2025-12-01 Categories: cs.RO, cs.AI, cs.CV, cs.LG arXiv: https://arxiv.org/abs/2512.01993v1 Autonomous driving policies are typically trained via open-loop behavior cloning of human demonstrations. However, such policies suffer from covariate shift when deployed in closed loop, leading to compounding errors. We introduce Rollouts as Demonstrations (RoaD), a simple and efficient method to mitigate covariate shift by leveraging the policy's own closed-loop rollouts as additional training data. During rollout generation, RoaD incorporates expert guidance to bias trajectories toward high-quality behavior, producing informative yet realistic demonstrations for fine-tuning. This approach enables robust closed-loop adaptation with orders of magnitude less data than reinforcement learning, and avoids restrictive assumptions of prior closed-loop supervised fine-tuning (CL-SFT) methods, allowing broader applications domains including end-to-end driving. We demonstrate the effectiveness of RoaD on WOSAC, a large-scale traffic simulation benchmark, where it performs similar or better than the prior CL-SFT method; and in AlpaSim, a high-fidelity neural reconstruction-based simulator for end-to-end driving, where it improves driving score by 41\% and reduces collisions by 54\%. --- ### LLM CHESS: Benchmarking Reasoning and Instruction-Following in LLMs through Chess Authors: Sai Kolasani, Maxim Saplin, Nicholas Crispino, Kyle Montgomery, Jared Quincy Davis, Matei Zaharia, Chi Wang, Chenguang Wang Published: 2025-12-01 Categories: cs.AI, cs.CL arXiv: https://arxiv.org/abs/2512.01992v1 We introduce LLM CHESS, an evaluation framework designed to probe the generalization of reasoning and instruction-following abilities in large language models (LLMs) through extended agentic interaction in the domain of chess. We rank over 50 open and closed source models by playing against a random opponent using a range of behavioral metrics, including win and loss rates, move quality, move legality, hallucinated actions, and game duration. For a subset of top reasoning models, we derive an Elo estimate by playing against a chess engine with variably configured skill, which allows for comparisons between models in an easily understandable way. Despite the simplicity of the instruction-following task and the weakness of the opponent, many state-of-the-art models struggle to complete games or achieve consistent wins. Similar to other benchmarks on complex reasoning tasks, our experiments reveal a clear separation between reasoning and non-reasoning models. However, unlike existing static benchmarks, the stochastic and dynamic nature of LLM CHESS uniquely reduces overfitting and memorization while preventing benchmark saturation, proving difficult even for top reasoning models. To support future work on evaluating reasoning and instruction-following in LLMs, we release our experimental framework, a public leaderboard, and a dataset of associated games. --- ### PAI-Bench: A Comprehensive Benchmark For Physical AI Authors: Fengzhe Zhou, Jiannan Huang, Jialuo Li, Deva Ramanan, Humphrey Shi Published: 2025-12-01 Categories: cs.CV arXiv: https://arxiv.org/abs/2512.01989v1 Physical AI aims to develop models that can perceive and predict real-world dynamics; yet, the extent to which current multi-modal large language models and video generative models support these abilities is insufficiently understood. We introduce Physical AI Bench (PAI-Bench), a unified and comprehensive benchmark that evaluates perception and prediction capabilities across video generation, conditional video generation, and video understanding, comprising 2,808 real-world cases with task-aligned metrics designed to capture physical plausibility and domain-specific reasoning. Our study provides a systematic assessment of recent models and shows that video generative models, despite strong visual fidelity, often struggle to maintain physically coherent dynamics, while multi-modal large language models exhibit limited performance in forecasting and causal interpretation. These observations suggest that current systems are still at an early stage in handling the perceptual and predictive demands of Physical AI. In summary, PAI-Bench establishes a realistic foundation for evaluating Physical AI and highlights key gaps that future systems must address. --- ### Artemis: Structured Visual Reasoning for Perception Policy Learning Authors: Wei Tang, Yanpeng Sun, Shan Zhang, Weihao Bo, Xiaofan Li, Piotr Koniusz, Wei Li, Na Zhao, Zechao Li Published: 2025-12-01 Categories: cs.CV arXiv: https://arxiv.org/abs/2512.01988v2 Recent reinforcement-learning frameworks for visual perception policy usually incorporate intermediate reasoning chains expressed in natural language. Empirical observations indicate that such purely linguistic intermediate reasoning often reduces performance on perception tasks. We argue that the core issue lies not in reasoning per se but in the form of reasoning: while these chains perform semantic reasoning in an unstructured linguistic space, \textbf{visual perception requires reasoning in a spatial and object-centric space}. In response, we introduce \textbf{Artemis}, a perception-policy learning method that performs structured visual reasoning, where each intermediate step is represented as a (label, bounding-box) pair capturing a verifiable visual state. This design enables explicit tracking of intermediate states, direct supervision for proposal quality, and avoids ambiguity introduced by language-based reasoning. Building upon verifiable and spatially grounded reasoning chains, Artemis provides a unified architecture for diverse perceptual tasks, without requiring the task-specific designs relied upon by prior perceptual policy models. Trained using grounding and detection sampeles in natural image domains, Artemis generalizes to counting and geometric perception tasks. At its core, a spatially grounded, object-centric chain rule provides a principled foundation for scalable and general perceptual policies. --- ### Forecasting in Offline Reinforcement Learning for Non-stationary Environments Authors: Suzan Ece Ada, Georg Martius, Emre Ugur, Erhan Oztop Published: 2025-12-01 Categories: cs.LG, cs.AI, cs.RO arXiv: https://arxiv.org/abs/2512.01987v3 Offline Reinforcement Learning (RL) provides a promising avenue for training policies from pre-collected datasets when gathering additional interaction data is infeasible. However, existing offline RL methods often assume stationarity or only consider synthetic perturbations at test time, assumptions that often fail in real-world scenarios characterized by abrupt, time-varying offsets. These offsets can lead to partial observability, causing agents to misperceive their true state and degrade performance. To overcome this challenge, we introduce Forecasting in Non-stationary Offline RL (FORL), a framework that unifies (i) conditional diffusion-based candidate state generation, trained without presupposing any specific pattern of future non-stationarity, and (ii) zero-shot time-series foundation models. FORL targets environments prone to unexpected, potentially non-Markovian offsets, requiring robust agent performance from the onset of each episode. Empirical evaluations on offline RL benchmarks, augmented with real-world time-series data to simulate realistic non-stationarity, demonstrate that FORL consistently improves performance compared to competitive baselines. By integrating zero-shot forecasting with the agent's experience, we aim to bridge the gap between offline RL and the complexities of real-world, non-stationary environments. --- ### A robust generalizable device-agnostic deep learning model for sleep-wake determination from triaxial wrist accelerometry Authors: Nasim Montazeri, Stone Yang, Dominik Luszczynski, John Zhang, Dharmendra Gurve, Andrew Centen, Maged Goubran, Andrew Lim Published: 2025-12-01 Categories: q-bio.QM, cs.LG arXiv: https://arxiv.org/abs/2512.01986v1 Study Objectives: Wrist accelerometry is widely used for inferring sleep-wake state. Previous works demonstrated poor wake detection, without cross-device generalizability and validation in different age range and sleep disorders. We developed a robust deep learning model for to detect sleep-wakefulness from triaxial accelerometry and evaluated its validity across three devices and in a large adult population spanning a wide range of ages with and without sleep disorders. Methods: We collected wrist accelerometry simultaneous to polysomnography (PSG) in 453 adults undergoing clinical sleep testing at a tertiary care sleep laboratory, using three devices. We extracted features in 30-second epochs and trained a 3-class model to detect wake, sleep, and sleep with arousals, which was then collapsed into wake vs. sleep using a decision tree. To enhance wake detection, the model was specifically trained on randomly selected subjects with low sleep efficiency and/or high arousal index from one device recording and then tested on the remaining recordings. Results: The model showed high performance with F1 Score of 0.86, sensitivity (sleep) of 0.87, and specificity (wakefulness) of 0.78, and significant and moderate correlation to PSG in predicting total sleep time (R=0.69) and sleep efficiency (R=0.63). Model performance was robust to the presence of sleep disorders, including sleep apnea and periodic limb movements in sleep, and was consistent across all three models of accelerometer. Conclusions: We present a deep model to detect sleep-wakefulness from actigraphy in adults with relative robustness to the presence of sleep disorders and generalizability across diverse commonly used wrist accelerometers. --- ### ECO: Energy-Constrained Operator Learning for Chaotic Dynamics with Boundedness Guarantees Authors: Andrea Goertzen, Sunbochen Tang, Navid Azizan Published: 2025-12-01 Categories: eess.SY, cs.LG arXiv: https://arxiv.org/abs/2512.01984v1 Chaos is a fundamental feature of many complex dynamical systems, including weather systems and fluid turbulence. These systems are inherently difficult to predict due to their extreme sensitivity to initial conditions. Many chaotic systems are dissipative and ergodic, motivating data-driven models that aim to learn invariant statistical properties over long time horizons. While recent models have shown empirical success in preserving invariant statistics, they are prone to generating unbounded predictions, which prevent meaningful statistics evaluation. To overcome this, we introduce the Energy-Constrained Operator (ECO) that simultaneously learns the system dynamics while enforcing boundedness in predictions. We leverage concepts from control theory to develop algebraic conditions based on a learnable energy function, ensuring the learned dynamics is dissipative. ECO enforces these algebraic conditions through an efficient closed-form quadratic projection layer, which provides provable trajectory boundedness. To our knowledge, this is the first work establishing such formal guarantees for data-driven chaotic dynamics models. Additionally, the learned invariant level set provides an outer estimate for the strange attractor, a complex structure that is computationally intractable to characterize. We demonstrate empirical success in ECO's ability to generate stable long-horizon forecasts, capturing invariant statistics on systems governed by chaotic PDEs, including the Kuramoto--Sivashinsky and the Navier--Stokes equations. --- ### Feature-Based Semantics-Aware Scheduling for Energy-Harvesting Federated Learning Authors: Eunjeong Jeong, Giovanni Perin, Howard H. Yang, Nikolaos Pappas Published: 2025-12-01 Categories: cs.LG, cs.DC, cs.IT, cs.NI, eess.SP arXiv: https://arxiv.org/abs/2512.01983v1 Federated Learning (FL) on resource-constrained edge devices faces a critical challenge: The computational energy required for training Deep Neural Networks (DNNs) often dominates communication costs. However, most existing Energy-Harvesting FL (EHFL) strategies fail to account for this reality, resulting in wasted energy due to redundant local computations. For efficient and proactive resource management, algorithms that predict local update contributions must be devised. We propose a lightweight client scheduling framework using the Version Age of Information (VAoI), a semantics-aware metric that quantifies update timeliness and significance. Crucially, we overcome VAoI's typical prohibitive computational cost, which requires statistical distance over the entire parameter space, by introducing a feature-based proxy. This proxy estimates model redundancy using intermediate-layer extraction from a single forward pass, dramatically reducing computational complexity. Experiments conducted under extreme non-IID data distributions and scarce energy availability demonstrate superior learning performance while achieving energy reduction compared to existing baseline selection policies. Our framework establishes semantics-aware scheduling as a practical and vital solution for EHFL in realistic scenarios where training costs dominate transmission costs. --- ### Low-Rank Prehab: Preparing Neural Networks for SVD Compression Authors: Haoran Qin, Shansita Sharma, Ali Abbasi, Chayne Thrash, Soheil Kolouri Published: 2025-12-01 Categories: cs.LG arXiv: https://arxiv.org/abs/2512.01980v1 Low-rank approximation methods such as singular value decomposition (SVD) and its variants (e.g., Fisher-weighted SVD, Activation SVD) have recently emerged as effective tools for neural network compression. In this setting, decomposition acts as a "surgical" intervention, followed by fine-tuning that serves as "rehab" to recover accuracy. Inspired by prehabilitation in surgery, we introduce a pre-compression fine-tuning stage, Low-Rank Prehab, that explicitly encourages low-rank structure in weight matrices while preserving task performance. By conditioning the model before SVD, Prehab steers weights toward spectrally compact regions of the parameter space, enabling smoother low-rank approximation and improved recovery. Experiments on large language models (LLMs) and other Transformer-based architectures, including Vision Transformers (ViTs), show that Prehab substantially reduces the immediate accuracy drop after compression and consistently improves post-finetuning performance. Across a wide range of compression ratios, our method outperforms state-of-the-art SVD-based techniques such as SVD-LLM, highlighting the importance of preparing models for compression rather than only improving the compression and recovery stages. Source code is available at https://github.com/niqretnuh/PREHAB-SVD --- ### Chain-of-Ground: Improving GUI Grounding via Iterative Reasoning and Reference Feedback Authors: Aiden Yiliu Li, Bizhi Yu, Daoan Lei, Tianhe Ren, Shilong Liu Published: 2025-12-01 Categories: cs.AI, cs.CL, cs.CV arXiv: https://arxiv.org/abs/2512.01979v1 GUI grounding aims to align natural language instructions with precise regions in complex user interfaces. Advanced multimodal large language models show strong ability in visual GUI grounding but still struggle with small or visually similar targets and ambiguity in real world layouts. These limitations arise from limited grounding capacity and from underuse of existing reasoning potential. We present Chain of Ground CoG a training free multi step grounding framework that uses multimodal large language models for iterative visual reasoning and refinement. Instead of direct prediction the model progressively reflects and adjusts its hypotheses leading to more accurate and interpretable localization. Our approach achieves 68.4 accuracy on the ScreenSpot Pro benchmark an improvement of 4.8 points. To measure real world generalization we introduce TPanel UI a dataset of 420 labeled industrial control panels with visual distortions such as blur and masking. On TPanel UI Chain of Ground improves over the strong baseline Qwen3 VL 235B by 6.9 points showing the effectiveness of multi step training free grounding across real world and digital interfaces. These results highlight a direction for unlocking grounding potential through structured iterative refinement instead of additional training. --- ### AI-Driven Optimization under Uncertainty for Mineral Processing Operations Authors: William Xu, Amir Eskanlou, Mansur Arief, David Zhen Yin, Jef K. Caers Published: 2025-12-01 Categories: eess.SY, cs.AI arXiv: https://arxiv.org/abs/2512.01977v2 The global capacity for mineral processing must expand rapidly to meet the demand for critical minerals, which are essential for building the clean energy technologies necessary to mitigate climate change. However, the efficiency of mineral processing is severely limited by uncertainty, which arises from both the variability of feedstock and the complexity of process dynamics. To optimize mineral processing circuits under uncertainty, we introduce an AI-driven approach that formulates mineral processing as a Partially Observable Markov Decision Process (POMDP). We demonstrate the capabilities of this approach in handling both feedstock uncertainty and process model uncertainty to optimize the operation of a simulated, simplified flotation cell as an example. We show that by integrating the process of information gathering (i.e., uncertainty reduction) and process optimization, this approach has the potential to consistently perform better than traditional approaches at maximizing an overall objective, such as net present value (NPV). Our methodological demonstration of this optimization-under-uncertainty approach for a synthetic case provides a mathematical and computational framework for later real-world application, with the potential to improve both the laboratory-scale design of experiments and industrial-scale operation of mineral processing circuits without any additional hardware. --- ### SGDiff: Scene Graph Guided Diffusion Model for Image Collaborative SegCaptioning Authors: Xu Zhang, Jin Yuan, Hanwang Zhang, Guojin Zhong, Yongsheng Zang, Jiacheng Lin, Zhiyong Li Published: 2025-12-01 Categories: cs.CV arXiv: https://arxiv.org/abs/2512.01975v1 Controllable image semantic understanding tasks, such as captioning or segmentation, necessitate users to input a prompt (e.g., text or bounding boxes) to predict a unique outcome, presenting challenges such as high-cost prompt input or limited information output. This paper introduces a new task ``Image Collaborative Segmentation and Captioning'' (SegCaptioning), which aims to translate a straightforward prompt, like a bounding box around an object, into diverse semantic interpretations represented by (caption, masks) pairs, allowing flexible result selection by users. This task poses significant challenges, including accurately capturing a user's intention from a minimal prompt while simultaneously predicting multiple semantically aligned caption words and masks. Technically, we propose a novel Scene Graph Guided Diffusion Model that leverages structured scene graph features for correlated mask-caption prediction. Initially, we introduce a Prompt-Centric Scene Graph Adaptor to map a user's prompt to a scene graph, effectively capturing his intention. Subsequently, we employ a diffusion process incorporating a Scene Graph Guided Bimodal Transformer to predict correlated caption-mask pairs by uncovering intricate correlations between them. To ensure accurate alignment, we design a Multi-Entities Contrastive Learning loss to explicitly align visual and textual entities by considering inter-modal similarity, resulting in well-aligned caption-mask pairs. Extensive experiments conducted on two datasets demonstrate that SGDiff achieves superior performance in SegCaptioning, yielding promising results for both captioning and segmentation tasks with minimal prompt input. --- ### Atomic Skills are the Prerequisite: When Reinforcement Learning Synthesizes Compositional Reasoning, and When It Only Amplifies Authors: Sitao Cheng, Xunjian Yin, Ruiwen Zhou, Yuxuan Li, Xinyi Wang, Liangming Pan, William Yang Wang, Victor Zhong Published: 2025-12-01 Categories: cs.AI, cs.CL arXiv: https://arxiv.org/abs/2512.01970v3 Does Reinforcement Learning (RL) merely amplify existing skills, or synthesize novel skills? We investigate this question through the lens of Complementary Reasoning: the critical practical capability of integrating internal knowledge with external context, a prerequisite for reliable Continual Learning and Retrieval-Augmented Generation. To avoid pre-training contamination, we construct a controlled semanticsynthetic dataset of biographies and decompose this capability into two atomic skills: Parametric Reasoning (retrieving facts encoded in model weights) and Contextual Reasoning (processing novel in-context information). We present two findings. First, models supervised directly on the composite task reach high accuracy on seen facts and reasoning paths (90%) but collapse on novel facts and reasoning paths (18%), indicating that Supervised Fine-Tuning (SFT) relies on rote memorization rather than genuine skill integration. Second, RL bridges this generalization gap, acting as a skill synthesizer rather than a mere amplifier--but only under a strict prerequisite: it synthesizes new composite strategies only when the base model has first mastered the independent atomic skills via SFT. These results suggest that decoupled atomic training followed by RL offers a scalable path to complex novel reasoning. --- ### SpriteHand: Real-Time Versatile Hand-Object Interaction with Autoregressive Video Generation Authors: Zisu Li, Hengye Lyu, Jiaxin Shi, Yufeng Zeng, Mingming Fan, Hanwang Zhang, Chen Liang Published: 2025-12-01 Categories: cs.CV, cs.HC arXiv: https://arxiv.org/abs/2512.01960v1 Modeling and synthesizing complex hand-object interactions remains a significant challenge, even for state-of-the-art physics engines. Conventional simulation-based approaches rely on explicitly defined rigid object models and pre-scripted hand gestures, making them inadequate for capturing dynamic interactions with non-rigid or articulated entities such as deformable fabrics, elastic materials, hinge-based structures, furry surfaces, or even living creatures. In this paper, we present SpriteHand, an autoregressive video generation framework for real-time synthesis of versatile hand-object interaction videos across a wide range of object types and motion patterns. SpriteHand takes as input a static object image and a video stream in which the hands are imagined to interact with the virtual object embedded in a real-world scene, and generates corresponding hand-object interaction effects in real time. Our model employs a causal inference architecture for autoregressive generation and leverages a hybrid post-training approach to enhance visual realism and temporal coherence. Our 1.3B model supports real-time streaming generation at around 18 FPS and 640x368 resolution, with an approximate 150 ms latency on a single NVIDIA RTX 5090 GPU, and more than a minute of continuous output. Experiments demonstrate superior visual quality, physical plausibility, and interaction fidelity compared to both generative and engine-based baselines. --- ### Learned-Rule-Augmented Large Language Model Evaluators Authors: Jie Meng, Jin Mao Published: 2025-12-01 Categories: cs.AI arXiv: https://arxiv.org/abs/2512.01958v1 Large language models (LLMs) are predominantly used as evaluators for natural language generation (NLG) tasks, but their application to broader evaluation scenarios remains limited. In this work, we explore the potential of LLMs as general evaluators across diverse tasks. Although LLM-based evaluators have made progress in different areas, existing methods struggle to generalize due to their reliance on costly, human-designed evaluation principles, which are often misaligned with both annotated data and LLMs' understanding.To address these challenges, we propose a rule-augmented evaluation paradigm. First, we introduce a rule distillation method that automatically extracts scoring rules from data using an LLM-assisted Monte Carlo Tree Search (MCTS), alleviating scalability issues and improving alignment with data. Second, to enable LLMs to effectively apply the learned rules, we propose two strategies: (1) Chain-of-Rule (CoR), which guides LLM to follow distilled rules, and (2) training a rule-augmented LLM evaluator (RuAE) via reinforcement learning, further bridging the gap between rules and LLMs' reasoning. Extensive experiments on diverse tasks demonstrate the effectiveness and generalizability of our approach across various evaluation scenarios. --- ### KV Pareto: Systems-Level Optimization of KV Cache and Model Compression for Long Context Inference Authors: Sai Gokhale, Devleena Das, Rajeev Patwari, Ashish Sirasao, Elliott Delaye Published: 2025-12-01 Categories: cs.LG arXiv: https://arxiv.org/abs/2512.01953v1 Long-context Large Language Models (LLMs) face significant memory bottlenecks during inference due to the linear growth of key-value (KV) cache with sequence length. While individual optimization techniques like KV cache quantization, chunked prefill, and model weight quantization have shown promise, their joint effects and optimal configurations for edge deployment remain underexplored. We introduce KV Pareto, a systems-level framework that systematically maps the trade-off frontier between total memory consumption and task accuracy across these three complementary optimization techniques. Our framework evaluates multiple LLM architectures (Qwen, Llama, Mistral) with varying KV quantization schemes (int2/4/8, mixed-precision), granularities (per-token, per-tensor, per-block), and 4-bit weight quantization via AWQ. Our framework identifies model-specific Pareto-optimal configurations that achieve 68-78% total memory reduction with minimal (1-3%) accuracy degradation on long-context tasks. We additionally verify the selected frontiers on additional benchmarks of Needle-in-a-Haystack, GSM8k and MMLU as well as extended context lengths of up to 128k to demonstrate the practical need of joint optimization for efficient LLM inference. --- ### GrndCtrl: Grounding World Models via Self-Supervised Reward Alignment Authors: Haoyang He, Jay Patrikar, Dong-Ki Kim, Max Smith, Daniel McGann, Ali-akbar Agha-mohammadi, Shayegan Omidshafiei, Sebastian Scherer Published: 2025-12-01 Categories: cs.CV, cs.AI, cs.LG, cs.RO arXiv: https://arxiv.org/abs/2512.01952v2 Recent advances in video world modeling have enabled large-scale generative models to simulate embodied environments with high visual fidelity, providing strong priors for prediction, planning, and control. Yet, despite their realism, these models often lack geometric grounding, limiting their use in navigation tasks that require spatial coherence and stability. We introduce Reinforcement Learning with World Grounding (RLWG), a self-supervised post-training framework that aligns pretrained world models with a physically verifiable structure through geometric and perceptual rewards. Analogous to reinforcement learning from verifiable feedback (RLVR) in language models, RLWG can use multiple rewards that measure pose cycle-consistency, depth reprojection, and temporal coherence. We instantiate this framework with GrndCtrl, a reward-aligned adaptation method based on Group Relative Policy Optimization (GRPO), yielding world models that maintain stable trajectories, consistent geometry, and reliable rollouts for embodied navigation. Like post-training alignment in large language models, GrndCtrl leverages verifiable rewards to bridge generative pretraining and grounded behavior, achieving superior spatial coherence and navigation stability over supervised fine-tuning in outdoor environments. --- ### Script: Graph-Structured and Query-Conditioned Semantic Token Pruning for Multimodal Large Language Models Authors: Zhongyu Yang, Dannong Xu, Wei Pang, Yingfang Yuan Published: 2025-12-01 Categories: cs.CV arXiv: https://arxiv.org/abs/2512.01949v1 The rapid growth of visual tokens in multimodal large language models (MLLMs) leads to excessive memory consumption and inference latency, especially when handling high-resolution images and videos. Token pruning is a technique used to mitigate this issue by removing redundancy, but existing methods often ignore relevance to the user query or suffer from the limitations of attention mechanisms, reducing their adaptability and effectiveness. To address these challenges, we propose Script, a plug-and-play pruning method that requires no retraining and generalizes across diverse MLLMs. Script comprises two modules: a graph-structured pruning module that removes visually redundant tokens, and a query-conditioned semantic pruning module that preserves query-relevant visual information. Together, they enhance performance on multimodal tasks. Experiments on fourteen benchmarks across image and video understanding tasks show that Script consistently achieves higher model efficiency and predictive accuracy compared to existing pruning methods. On LLaVA-NeXT-7B, it achieves up to 6.8x prefill speedup and 10x FLOP reduction, while retaining 96.88% of the original performance. --- ### How Far Are We from Genuinely Useful Deep Research Agents? Authors: Dingling Zhang, He Zhu, Jincheng Ren, Kangqi Song, Xinran Zhou, Boyu Feng, Shudong Liu, Jiabin Luo, Weihao Xie, Zhaohui Wang, Tianrui Qin, King Zhu et al. Published: 2025-12-01 Categories: cs.CL arXiv: https://arxiv.org/abs/2512.01948v2 Deep Research Agents (DRAs) aim to automatically produce analyst-level reports through iterative information retrieval and synthesis. However, most existing DRAs were validated on question-answering benchmarks, while research on generating comprehensive reports remains overlooked. Worse, current benchmarks for report synthesis suffer from task complexity and subjective metrics -- this fails to reflect user demands and limits the practical utility of generated reports. To address these gaps, we present Fine-grained DEepResearch bench (FINDER), an enhanced benchmark consisting of 100 human-curated research tasks with 419 structured checklist items that standardize report structure, analytical depth, and factual grounding. Based on approximately 1,000 reports produced by mainstream DRAs, we further propose Deep rEsearch Failure Taxonomy (DEFT), the first failure taxonomy for deep research agents. DEFT contains 14 fine-grained failure modes across reasoning, retrieval, and generation, and is built upon grounded theory with human-LLM co-annotating and inter-annotator reliability validation. Our experimental findings reveal that current DRAs struggle not with task comprehension but with evidence integration, verification, and reasoning-resilient planning. --- ### Guardian: Detecting Robotic Planning and Execution Errors with Vision-Language Models Authors: Paul Pacaud, Ricardo Garcia, Shizhe Chen, Cordelia Schmid Published: 2025-12-01 Categories: cs.RO, cs.CV arXiv: https://arxiv.org/abs/2512.01946v4 Robust robotic manipulation requires reliable failure detection and recovery. Although recent Vision-Language Models (VLMs) show promise in robot failure detection, their generalization is severely limited by the scarcity and narrow coverage of failure data. To address this bottleneck, we propose an automatic framework for generating diverse robotic planning and execution failures across both simulated and real-world environments. Our approach perturbs successful manipulation trajectories to synthesize failures that reflect realistic failure distributions, and leverages VLMs to produce structured step-by-step reasoning traces. This yields GuardianFail-36k, a large-scale failure reasoning dataset built upon the RLBench simulator and the BridgeDataV2 real-robot dataset. Using GuardianFail-36k, we train Guardian, a multi-view reasoning VLM for unified planning and execution verification. Guardian achieves state-of-the-art performance on three unseen real-world benchmarks: RoboFail, RoboVQA, and our newly introduced UR5-Fail. When integrated with a state-of-the-art LLM-based manipulation policy, it consistently boosts task success rates in both simulation and real-world deployment. These results demonstrate that scaling high-quality failure reasoning data is critical for improving generalization in robotic failure detection. Code, Data, and Models are available at https://www.di.ens.fr/willow/research/guardian/. --- ### Agentic Policy Optimization via Instruction-Policy Co-Evolution Authors: Han Zhou, Xingchen Wan, Ivan Vulić, Anna Korhonen Published: 2025-12-01 Categories: cs.LG, cs.AI, cs.CL arXiv: https://arxiv.org/abs/2512.01945v2 Reinforcement Learning with Verifiable Rewards (RLVR) has advanced the reasoning capability of large language models (LLMs), enabling autonomous agents that can conduct effective multi-turn and tool-integrated reasoning. While instructions serve as the primary protocol for defining agents, RLVR typically relies on static and manually designed instructions. However, those instructions may be suboptimal for the base model, and the optimal instruction may change as the agent's policy improves and explores the interaction with the environment. To bridge the gap, we introduce INSPO, a novel Instruction-Policy co-evolution framework that integrates instruction optimization as a dynamic component of the reinforcement learning (RL) loop. INSPO maintains a dynamic population of instruction candidates that are sampled with questions, where reward signals in RL loops are automatically attributed to each instruction, and low performers are periodically pruned. New instructions are generated and verified through an on-policy reflection mechanism, where an LLM-based optimizer analyzes past experience from a replay buffer and evolves more effective strategies given the current policy. We conduct extensive experiments on multi-turn retrieval and reasoning tasks, demonstrating that INSPO substantially outperforms strong baselines relying on static instructions. INSPO discovers innovative instructions that guide the agent toward more strategic reasoning paths, achieving substantial performance gains with only a marginal increase in computational overhead. --- ### An Empirical Study of Agent Developer Practices in AI Agent Frameworks Authors: Yanlin Wang, Xinyi Xu, Jiachi Chen, Tingting Bi, Wenchao Gu, Zibin Zheng Published: 2025-12-01 Categories: cs.SE, cs.AI arXiv: https://arxiv.org/abs/2512.01939v1 The rise of large language models (LLMs) has sparked a surge of interest in agents, leading to the rapid growth of agent frameworks. Agent frameworks are software toolkits and libraries that provide standardized components, abstractions, and orchestration mechanisms to simplify agent development. Despite widespread use of agent frameworks, their practical applications and how they influence the agent development process remain underexplored. Different agent frameworks encounter similar problems during use, indicating that these recurring issues deserve greater attention and call for further improvements in agent framework design. Meanwhile, as the number of agent frameworks continues to grow and evolve, more than 80% of developers report difficulties in identifying the frameworks that best meet their specific development requirements. In this paper, we conduct the first empirical study of LLM-based agent frameworks, exploring real-world experiences of developers in building AI agents. To compare how well the agent frameworks meet developer needs, we further collect developer discussions for the ten previously identified agent frameworks, resulting in a total of 11,910 discussions. Finally, by analyzing these discussions, we compare the frameworks across five dimensions: development efficiency, functional abstraction, learning cost, performance optimization, and maintainability, which refers to how easily developers can update and extend both the framework itself and the agents built upon it over time. Our comparative analysis reveals significant differences among frameworks in how they meet the needs of agent developers. Overall, we provide a set of findings and implications for the LLM-driven AI agent framework ecosystem and offer insights for the design of future LLM-based agent frameworks and agent developers. --- ### Physical ID-Transfer Attacks against Multi-Object Tracking via Adversarial Trajectory Authors: Chenyi Wang, Yanmao Man, Raymond Muller, Ming Li, Z. Berkay Celik, Ryan Gerdes, Jonathan Petit Published: 2025-12-01 Categories: cs.CV arXiv: https://arxiv.org/abs/2512.01934v1 Multi-Object Tracking (MOT) is a critical task in computer vision, with applications ranging from surveillance systems to autonomous driving. However, threats to MOT algorithms have yet been widely studied. In particular, incorrect association between the tracked objects and their assigned IDs can lead to severe consequences, such as wrong trajectory predictions. Previous attacks against MOT either focused on hijacking the trackers of individual objects, or manipulating the tracker IDs in MOT by attacking the integrated object detection (OD) module in the digital domain, which are model-specific, non-robust, and only able to affect specific samples in offline datasets. In this paper, we present AdvTraj, the first online and physical ID-manipulation attack against tracking-by-detection MOT, in which an attacker uses adversarial trajectories to transfer its ID to a targeted object to confuse the tracking system, without attacking OD. Our simulation results in CARLA show that AdvTraj can fool ID assignments with 100% success rate in various scenarios for white-box attacks against SORT, which also have high attack transferability (up to 93% attack success rate) against state-of-the-art (SOTA) MOT algorithms due to their common design principles. We characterize the patterns of trajectories generated by AdvTraj and propose two universal adversarial maneuvers that can be performed by a human walker/driver in daily scenarios. Our work reveals under-explored weaknesses in the object association phase of SOTA MOT systems, and provides insights into enhancing the robustness of such systems. --- ### SVRG and Beyond via Posterior Correction Authors: Nico Daheim, Thomas Möllenhoff, Ming Liang Ang, Mohammad Emtiyaz Khan Published: 2025-12-01 Categories: cs.LG, cs.AI arXiv: https://arxiv.org/abs/2512.01930v2 Stochastic Variance Reduced Gradient (SVRG) and its variants aim to speed-up training by using gradient corrections. Originally proposed over a decade ago, these methods have never been connected to any Bayesian method at a fundamental level. Here, we fill this gap and derive surprising new connections of SVRG to a recently proposed Bayesian method called `posterior correction'. Our main contribution is to show that SVRG can be recovered as a special case of posterior-correction over isotropic-Gaussian posteriors. Novel extensions of SVRG are automatically obtained by using more flexible exponential-family posteriors. We derive two new such extensions by using Gaussian families: a Newton-like variant with novel Hessian corrections, and an Adam-like extension that scales to large problems. Our work is the first to connect SVRG to Bayes and use it to speed-up training. --- ### Rectifying LLM Thought from Lens of Optimization Authors: Junnan Liu, Hongwei Liu, Songyang Zhang, Kai Chen Published: 2025-12-01 Categories: cs.CL, cs.AI arXiv: https://arxiv.org/abs/2512.01925v2 Recent advancements in large language models (LLMs) have been driven by their emergent reasoning capabilities, particularly through long chain-of-thought (CoT) prompting, which enables thorough exploration and deliberation. Despite these advances, long-CoT LLMs often exhibit suboptimal reasoning behaviors, such as overthinking and excessively protracted reasoning chains, which can impair performance. In this paper, we analyze reasoning processes through an optimization lens, framing CoT as a gradient descent procedure where each reasoning step constitutes an update toward problem resolution. Building on this perspective, we introduce RePro (Rectifying Process-level Reward), a novel approach to refine LLM reasoning during post-training. RePro defines a surrogate objective function to assess the optimization process underlying CoT, utilizing a dual scoring mechanism to quantify its intensity and stability. These scores are aggregated into a composite process-level reward, seamlessly integrated into reinforcement learning with verifiable rewards (RLVR) pipelines to optimize LLMs. Extensive experiments across multiple reinforcement learning algorithms and diverse LLMs, evaluated on benchmarks spanning mathematics, science, and coding, demonstrate that RePro consistently enhances reasoning performance and mitigates suboptimal reasoning behaviors. --- ### Real-World Robot Control by Deep Active Inference With a Temporally Hierarchical World Model Authors: Kentaro Fujii, Shingo Murata Published: 2025-12-01 Categories: cs.RO, cs.AI, cs.LG arXiv: https://arxiv.org/abs/2512.01924v1 Robots in uncertain real-world environments must perform both goal-directed and exploratory actions. However, most deep learning-based control methods neglect exploration and struggle under uncertainty. To address this, we adopt deep active inference, a framework that accounts for human goal-directed and exploratory actions. Yet, conventional deep active inference approaches face challenges due to limited environmental representation capacity and high computational cost in action selection. We propose a novel deep active inference framework that consists of a world model, an action model, and an abstract world model. The world model encodes environmental dynamics into hidden state representations at slow and fast timescales. The action model compresses action sequences into abstract actions using vector quantization, and the abstract world model predicts future slow states conditioned on the abstract action, enabling low-cost action selection. We evaluate the framework on object-manipulation tasks with a real-world robot. Results show that it achieves high success rates across diverse manipulation tasks and switches between goal-directed and exploratory actions in uncertain settings, while making action selection computationally tractable. These findings highlight the importance of modeling multiple timescale dynamics and abstracting actions and state transitions. --- ### Med-VCD: Mitigating Hallucination for Medical Large Vision Language Models through Visual Contrastive Decoding Authors: Zahra Mahdavi, Zahra Khodakaramimaghsoud, Hooman Khaloo, Sina Bakhshandeh Taleshani, Erfan Hashemi, Javad Mirzapour Kaleybar, Omid Nejati Manzari Published: 2025-12-01 Categories: cs.CV arXiv: https://arxiv.org/abs/2512.01922v1 Large vision-language models (LVLMs) are now central to healthcare applications such as medical visual question answering and imaging report generation. Yet, these models remain vulnerable to hallucination outputs that appear plausible but are in fact incorrect. In the natural image domain, several decoding strategies have been proposed to mitigate hallucinations by reinforcing visual evidence, but most rely on secondary decoding or rollback procedures that substantially slow inference. Moreover, existing solutions are often domain-specific and may introduce misalignment between modalities or between generated and ground-truth content. We introduce Med-VCD, a sparse visual-contrastive decoding method that mitigates hallucinations in medical LVLMs without the time overhead of secondary decoding. Med-VCD incorporates a novel token-sparsification strategy that selects visually informed tokens on the fly, trimming redundancy while retaining critical visual context and thus balancing efficiency with reliability. Evaluations on eight medical datasets, spanning ophthalmology, radiology, and pathology tasks in visual question answering, report generation, and dedicated hallucination benchmarks, show that Med-VCD raises factual accuracy by an average of 13\% and improves hallucination accuracy by 6\% relative to baseline medical LVLMs. --- ### A Footprint-Aware, High-Resolution Approach for Carbon Flux Prediction Across Diverse Ecosystems Authors: Jacob Searcy, Anish Dulal, Courtney Mathers, Scott Bridgham, Ashley Cordes, Lillian Aoki, Brendan Bohannan, Qing Zhu, Lucas C. R. Silva Published: 2025-12-01 Categories: cs.LG arXiv: https://arxiv.org/abs/2512.01917v2 Eddy-covariance (EC) flux towers provide in situ measurements of $CO_2$ flux and serve as the ground-truth data for predictive `upscaling' models derived from satellite products. However, many satellites now resolve spatial scales smaller than an EC tower's footprint. We show theoretically that upscaling models trained on high-resolution data in heterogeneous landscapes must account for an EC tower's footprint to avoid bias in pixel-level predictors. To address this problem, we introduce Footprint-Aware Regression (FAR), a deep-learning framework that simultaneously predicts spatial footprints and pixel-level estimates of $CO_2$ flux, and show it yields unbiased pixel-level predictions given sufficient training data. We demonstrate FAR on our AMERI-FAR25 dataset, which combines 205 site-years of tower data with corresponding Landsat scenes, and show that FAR outperforms non-footprint-aware models. FAR increased half-hourly $R^2$ from approximately 0.575 to 0.634 and reduced RMSE by about 7% relative to the best fixed-footprint baseline on a dataset of withheld sites. Gains are larger for monthly and yearly averages relative to a coarser 990 m baseline. Site-level analyses show that these gains extend across multiple ecosystem types. These performance gains are observed whether footprints are learned jointly or estimated independently using an established footprint model, despite substantial variation in footprint size between methods. In a regional comparison over the Western Cascades, FAR produces flux estimates with a distribution comparable to that of existing high-resolution process-based models. --- ### Disentangling Progress in Medical Image Registration: Beyond Trend-Driven Architectures towards Domain-Specific Strategies Authors: Bailiang Jian, Jiazhen Pan, Rohit Jena, Morteza Ghahremani, Hongwei Bran Li, Daniel Rueckert, Christian Wachinger, Benedikt Wiestler Published: 2025-12-01 Categories: eess.IV, cs.CV arXiv: https://arxiv.org/abs/2512.01913v1 Medical image registration drives quantitative analysis across organs, modalities, and patient populations. Recent deep learning methods often combine low-level "trend-driven" computational blocks from computer vision, such as large-kernel CNNs, Transformers, and state-space models, with high-level registration-specific designs like motion pyramids, correlation layers, and iterative refinement. Yet, their relative contributions remain unclear and entangled. This raises a central question: should future advances in registration focus on importing generic architectural trends or on refining domain-specific design principles? Through a modular framework spanning brain, lung, cardiac, and abdominal registration, we systematically disentangle the influence of these two paradigms. Our evaluation reveals that low-level "trend-driven" computational blocks offer only marginal or inconsistent gains, while high-level registration-specific designs consistently deliver more accurate, smoother, and more robust deformations. These domain priors significantly elevate the performance of a standard U-Net baseline, far more than variants incorporating "trend-driven" blocks, achieving an average relative improvement of $\sim3\%$. All models and experiments are released within a transparent, modular benchmark that enables plug-and-play comparison for new architectures and registration tasks (https://github.com/BailiangJ/rethink-reg). This dynamic and extensible platform establishes a common ground for reproducible and fair evaluation, inviting the community to isolate genuine methodological contributions from domain priors. Our findings advocate a shift in research emphasis: from following architectural trends to embracing domain-specific design principles as the true drivers of progress in learning-based medical image registration. --- ### Latent Debate: A Surrogate Framework for Interpreting LLM Thinking Authors: Lihu Chen, Xiang Yin, Francesca Toni Published: 2025-12-01 Categories: cs.CL arXiv: https://arxiv.org/abs/2512.01909v2 Understanding the internal thinking process of Large Language Models (LLMs) and the cause of hallucinations remains a key challenge. To this end, we introduce latent debate, a novel framework for interpreting model predictions through the lens of implicit internal arguments. Unlike the current work of self-consistency and multi-agent debate, which relies on explicit debates among multiple answers or multiple models, latent debate captures the hidden supporting and attacking signals that arise within a single model during a single inference. We first present a model- and task-agnostic conceptual framework, and then instantiate it symbolically to approximate the thinking process of LLMs on True/False prediction tasks. Empirical studies demonstrate that latent debate is a faithful structured surrogate model that has highly consistent predictions with the original LLM. Beyond interpretability, we demonstrate that latent debate provides a strong baseline for hallucination detection. Further analysis reveals strong correlations between hallucinations and debate patterns, such as a high degree of latent debates in the middle layers is linked to a higher risk of hallucinations. These findings position latent debate as a potential framework for understanding internal mechanisms of LLMs, especially for scenarios where internal (dis)agreements appear during the inference steps. --- ### SARL: Spatially-Aware Self-Supervised Representation Learning for Visuo-Tactile Perception Authors: Gurmeher Khurana, Lan Wei, Dandan Zhang Published: 2025-12-01 Categories: cs.CV arXiv: https://arxiv.org/abs/2512.01908v1 Contact-rich robotic manipulation requires representations that encode local geometry. Vision provides global context but lacks direct measurements of properties such as texture and hardness, whereas touch supplies these cues. Modern visuo-tactile sensors capture both modalities in a single fused image, yielding intrinsically aligned inputs that are well suited to manipulation tasks requiring visual and tactile information. Most self-supervised learning (SSL) frameworks, however, compress feature maps into a global vector, discarding spatial structure and misaligning with the needs of manipulation. To address this, we propose SARL, a spatially-aware SSL framework that augments the Bootstrap Your Own Latent (BYOL) architecture with three map-level objectives, including Saliency Alignment (SAL), Patch-Prototype Distribution Alignment (PPDA), and Region Affinity Matching (RAM), to keep attentional focus, part composition, and geometric relations consistent across views. These losses act on intermediate feature maps, complementing the global objective. SARL consistently outperforms nine SSL baselines across six downstream tasks with fused visual-tactile data. On the geometry-sensitive edge-pose regression task, SARL achieves a Mean Absolute Error (MAE) of 0.3955, a 30% relative improvement over the next-best SSL method (0.5682 MAE) and approaching the supervised upper bound. These findings indicate that, for fused visual-tactile data, the most effective signal is structured spatial equivariance, in which features vary predictably with object geometry, which enables more capable robotic perception. --- ### Delays in Spiking Neural Networks: A State Space Model Approach Authors: Sanja Karilanova, Subhrakanti Dey, Ayça Özçelikkale Published: 2025-12-01 Categories: cs.LG arXiv: https://arxiv.org/abs/2512.01906v3 Spiking neural networks (SNNs) are biologically inspired, event-driven models suited for temporal data processing and energy-efficient neuromorphic computing. In SNNs, richer neuronal dynamic allows capturing more complex temporal dependencies, with delays playing a crucial role by allowing past inputs to directly influence present spiking behavior. We propose a general framework for incorporating delays into SNNs through additional state variables. The proposed mechanism enables each neuron to access a finite temporal input history. The framework is agnostic to neuron models and hence can be seamlessly integrated into standard spiking neuron models such as Leaky Integrate-and-Fire (LIF) and Adaptive LIF (adLIF). We analyze how the duration of the delays and the learnable parameters associated with them affect the performance. We investigate the trade-offs in the network architecture due to additional state variables introduced by the delay mechanism. Experiments on the Spiking Heidelberg Digits (SHD) dataset show that the proposed mechanism matches existing delay-based SNNs in performance while remaining computationally efficient, with particular gains in smaller networks. --- ### Provably Safe Model Updates Authors: Leo Elmecker-Plakolm, Pierre Fasterling, Philip Sosnin, Calvin Tsay, Matthew Wicker Published: 2025-12-01 Categories: cs.LG, stat.ML arXiv: https://arxiv.org/abs/2512.01899v2 Safety-critical environments are inherently dynamic. Distribution shifts, emerging vulnerabilities, and evolving requirements demand continuous updates to machine learning models. Yet even benign parameter updates can have unintended consequences, such as catastrophic forgetting in classical models or alignment drift in foundation models. Existing heuristic approaches (e.g., regularization, parameter isolation) can mitigate these effects but cannot certify that updated models continue to satisfy required performance specifications. We address this problem by introducing a framework for provably safe model updates. Our approach first formalizes the problem as computing the largest locally invariant domain (LID): a connected region in parameter space where all points are certified to satisfy a given specification. While exact maximal LID computation is intractable, we show that relaxing the problem to parameterized abstract domains (orthotopes, zonotopes) yields a tractable primal-dual formulation. This enables efficient certification of updates - independent of the data or algorithm used - by projecting them onto the safe domain. Our formulation further allows computation of multiple approximately optimal LIDs, incorporation of regularization-inspired biases, and use of lookahead data buffers. Across continual learning and foundation model fine-tuning benchmarks, our method matches or exceeds heuristic baselines for avoiding forgetting while providing formal safety guarantees. --- ### OPOR-Bench: Evaluating Large Language Models on Online Public Opinion Report Generation Authors: Jinzheng Yu, Yang Xu, Haozhen Li, Junqi Li, Yifan Feng, Ligu Zhu, Hao Shen, Lei Shi Published: 2025-12-01 Categories: cs.CL arXiv: https://arxiv.org/abs/2512.01896v1 Online Public Opinion Reports consolidate news and social media for timely crisis management by governments and enterprises. While large language models have made automated report generation technically feasible, systematic research in this specific area remains notably absent, particularly lacking formal task definitions and corresponding benchmarks. To bridge this gap, we define the Automated Online Public Opinion Report Generation (OPOR-GEN) task and construct OPOR-BENCH, an event-centric dataset covering 463 crisis events with their corresponding news articles, social media posts, and a reference summary. To evaluate report quality, we propose OPOR-EVAL, a novel agent-based framework that simulates human expert evaluation by analyzing generated reports in context. Experiments with frontier models demonstrate that our framework achieves high correlation with human judgments. Our comprehensive task definition, benchmark dataset, and evaluation framework provide a solid foundation for future research in this critical domain. --- ### StyleYourSmile: Cross-Domain Face Retargeting Without Paired Multi-Style Data Authors: Avirup Dey, Vinay Namboodiri Published: 2025-12-01 Categories: cs.CV arXiv: https://arxiv.org/abs/2512.01895v1 Cross-domain face retargeting requires disentangled control over identity, expressions, and domain-specific stylistic attributes. Existing methods, typically trained on real-world faces, either fail to generalize across domains, need test-time optimizations, or require fine-tuning with carefully curated multi-style datasets to achieve domain-invariant identity representations. In this work, we introduce \textit{StyleYourSmile}, a novel one-shot cross-domain face retargeting method that eliminates the need for curated multi-style paired data. We propose an efficient data augmentation strategy alongside a dual-encoder framework, for extracting domain-invariant identity cues and capturing domain-specific stylistic variations. Leveraging these disentangled control signals, we condition a diffusion model to retarget facial expressions across domains. Extensive experiments demonstrate that \textit{StyleYourSmile} achieves superior identity preservation and retargeting fidelity across a wide range of visual domains. --- ### Exploring Human Perceptions of AI Responses: Insights from a Mixed-Methods Study on Risk Mitigation in Generative Models Authors: Heloisa Candello, Muneeza Azmat, Uma Sushmitha Gunturi, Raya Horesh, Rogerio Abreu de Paula, Heloisa Pimentel, Marcelo Carpinette Grave, Aminat Adebiyi, Tiago Machado, Maysa Malfiza Garcia de Macedo Published: 2025-12-01 Categories: cs.CL, cs.AI, cs.HC arXiv: https://arxiv.org/abs/2512.01892v1 With the rapid uptake of generative AI, investigating human perceptions of generated responses has become crucial. A major challenge is their `aptitude' for hallucinating and generating harmful contents. Despite major efforts for implementing guardrails, human perceptions of these mitigation strategies are largely unknown. We conducted a mixed-method experiment for evaluating the responses of a mitigation strategy across multiple-dimensions: faithfulness, fairness, harm-removal capacity, and relevance. In a within-subject study design, 57 participants assessed the responses under two conditions: harmful response plus its mitigation and solely mitigated response. Results revealed that participants' native language, AI work experience, and annotation familiarity significantly influenced evaluations. Participants showed high sensitivity to linguistic and contextual attributes, penalizing minor grammar errors while rewarding preserved semantic contexts. This contrasts with how language is often treated in the quantitative evaluation of LLMs. We also introduced new metrics for training and evaluating mitigation strategies and insights for human-AI evaluation studies. --- ### Elastic Weight Consolidation for Knowledge Graph Continual Learning: An Empirical Evaluation Authors: Gaganpreet Jhajj, Fuhua Lin Published: 2025-12-01 Categories: cs.LG arXiv: https://arxiv.org/abs/2512.01890v1 Knowledge graphs (KGs) require continual updates as new information emerges, but neural embedding models suffer from catastrophic forgetting when learning new tasks sequentially. We evaluate Elastic Weight Consolidation (EWC), a regularization-based continual learning method, on KG link prediction using TransE embeddings on FB15k-237. Across multiple experiments with five random seeds, we find that EWC reduces catastrophic forgetting from 12.62% to 6.85%, a 45.7% reduction compared to naive sequential training. We observe that the task partitioning strategy affects the magnitude of forgetting: relation-based partitioning (grouping triples by relation type) exhibits 9.8 percentage points higher forgetting than randomly partitioned tasks (12.62% vs 2.81%), suggesting that task construction influences evaluation outcomes. While focused on a single embedding model and dataset, our results demonstrate that EWC effectively mitigates catastrophic forgetting in KG continual learning and highlight the importance of evaluation protocol design. --- ### KM-ViPE: Online Tightly Coupled Vision-Language-Geometry Fusion for Open-Vocabulary Semantic SLAM Authors: Zaid Nasser, Mikhail Iumanov, Tianhao Li, Maxim Popov, Jaafar Mahmoud, Malik Mohrat, Ilya Obrubov, Ekaterina Derevyanka, Ivan Sosin, Sergey Kolyubin Published: 2025-12-01 Categories: cs.CV arXiv: https://arxiv.org/abs/2512.01889v1 We present KM-ViPE (Knowledge Mapping Video Pose Engine), a real-time open-vocabulary SLAM framework for uncalibrated monocular cameras in dynamic environments. Unlike systems requiring depth sensors and offline calibration, KM-ViPE operates directly on raw RGB streams, making it ideal for ego-centric applications and harvesting internet-scale video data for training. KM-ViPE tightly couples DINO visual features with geometric constraints through a high-level features based adaptive robust kernel that handles both moving objects and movable static objects (e.g., moving furniture in ego-centric views). The system performs simultaneous online localization and open-vocabulary semantic mapping by fusing geometric and deep visual features aligned with language embeddings. Our results are competitive with state-of-the-art approaches, while existing solutions either operate offline, need depth data and/or odometry estimation, or lack dynamic scene robustness. KM-ViPE benefits from internet-scale training and uniquely combines online operation, uncalibrated monocular input, and robust handling of dynamic scenes, which makes it a good fit for autonomous robotics and AR/VR applications and advances practical spatial intelligence capabilities for embodied AI. --- ### Domain-Decomposed Graph Neural Network Surrogate Modeling for Ice Sheets Authors: Adrienne M. Propp, Mauro Perego, Eric C. Cyr, Anthony Gruber, Amanda A. Howard, Alexander Heinlein, Panos Stinis, Daniel M. Tartakovsky Published: 2025-12-01 Categories: cs.LG, math-ph, math.NA, physics.comp-ph arXiv: https://arxiv.org/abs/2512.01888v1 Accurate yet efficient surrogate models are essential for large-scale simulations of partial differential equations (PDEs), particularly for uncertainty quantification (UQ) tasks that demand hundreds or thousands of evaluations. We develop a physics-inspired graph neural network (GNN) surrogate that operates directly on unstructured meshes and leverages the flexibility of graph attention. To improve both training efficiency and generalization properties of the model, we introduce a domain decomposition (DD) strategy that partitions the mesh into subdomains, trains local GNN surrogates in parallel, and aggregates their predictions. We then employ transfer learning to fine-tune models across subdomains, accelerating training and improving accuracy in data-limited settings. Applied to ice sheet simulations, our approach accurately predicts full-field velocities on high-resolution meshes, substantially reduces training time relative to training a single global surrogate model, and provides a ripe foundation for UQ objectives. Our results demonstrate that graph-based DD, combined with transfer learning, provides a scalable and reliable pathway for training GNN surrogates on massive PDE-governed systems, with broad potential for application beyond ice sheet dynamics. --- ### TransientTrack: Advanced Multi-Object Tracking and Classification of Cancer Cells with Transient Fluorescent Signals Authors: Florian Bürger, Martim Dias Gomes, Nica Gutu, Adrián E. Granada, Noémie Moreau, Katarzyna Bozek Published: 2025-12-01 Categories: cs.CV, q-bio.CB, q-bio.QM arXiv: https://arxiv.org/abs/2512.01885v2 Tracking cells in time-lapse videos is an essential technique for monitoring cell population dynamics at a single-cell level. Current methods for cell tracking are developed on videos with mostly single, constant signals and do not detect pivotal events such as cell death. Here, we present TransientTrack, a deep learning-based framework for cell tracking in multi-channel microscopy video data with transient fluorescent signals that fluctuate over time following processes such as the circadian rhythm of cells. By identifying key cellular events - mitosis (cell division) and apoptosis (cell death) our method allows us to build complete trajectories, including cell lineage information. TransientTrack is lightweight and performs matching on cell detection embeddings directly, without the need for quantification of tracking-specific cell features. Furthermore, our approach integrates Transformer Networks, multi-stage matching using all detection boxes, and the interpolation of missing tracklets with the Kalman Filter. This unified framework achieves strong performance across diverse conditions, effectively tracking cells and capturing cell division and death. We demonstrate the use of TransientTrack in an analysis of the efficacy of a chemotherapeutic drug at a single-cell level. The proposed framework could further advance quantitative studies of cancer cell dynamics, enabling detailed characterization of treatment response and resistance mechanisms. The code is available at https://github.com/bozeklab/TransientTrack. --- ### New Spiking Architecture for Multi-Modal Decision-Making in Autonomous Vehicles Authors: Aref Ghoreishee, Abhishek Mishra, Lifeng Zhou, John Walsh, Nagarajan Kandasamy Published: 2025-12-01 Categories: cs.LG arXiv: https://arxiv.org/abs/2512.01882v1 This work proposes an end-to-end multi-modal reinforcement learning framework for high-level decision-making in autonomous vehicles. The framework integrates heterogeneous sensory input, including camera images, LiDAR point clouds, and vehicle heading information, through a cross-attention transformer-based perception module. Although transformers have become the backbone of modern multi-modal architectures, their high computational cost limits their deployment in resource-constrained edge environments. To overcome this challenge, we propose a spiking temporal-aware transformer-like architecture that uses ternary spiking neurons for computationally efficient multi-modal fusion. Comprehensive evaluations across multiple tasks in the Highway Environment demonstrate the effectiveness and efficiency of the proposed approach for real-time autonomous decision-making. --- ### Unifying Sign and Magnitude for Optimizing Deep Vision Networks via ThermoLion Authors: Ahmed Nebli Published: 2025-12-01 Categories: cs.LG, cs.AI arXiv: https://arxiv.org/abs/2512.01881v2 The training of deep vision models is fundamentally a signal recovery problem amidst high-dimensional stochastic noise. Current optimization paradigms impose a static compromise on information channel capacity. For instance, magnitude-based methods, such as AdamW, operate on the assumption that gradient norms are high-fidelity curvature signals. While this allows for precision in smooth regimes, it leads to catastrophic noise amplification when applied to rugged, non-convex landscapes. Conversely, sign-based methods (e.g., Lion) perform a radical 1-bit quantization of the gradient, which aims to provide robust regularization at the cost of discarding fine-grained descent information. We propose that optimal convergence requires neither static prior, but rather a dynamic modulation of the update bitrate. We introduce ThermoLion, a vision-centric framework that utilizes local Signal-to-Noise Ratio (SNR) gating to autonomously transition parameters between a "low-bit" exploration phase and a "high-precision" exploitation phase. Furthermore, we introduce a Momentum Alignment mechanism that detects constructive interference between historical drift and instantaneous gradients to accelerate convergence during stable trajectories. Empirical benchmarks across 12 diverse vision datasets (including CIFAR, SVHN, and GTSRB) demonstrate that ThermoLion surpasses state-of-the-art optimizers, such as AdamW and Lion, in convergence speed and terminal accuracy. --- ### Predicting Human Chess Moves: An AI Assisted Analysis of Chess Games Using Skill-group Specific n-gram Language Models Authors: Daren Zhong, Dingcheng Huang, Clayton Greenberg Published: 2025-12-01 Categories: cs.AI arXiv: https://arxiv.org/abs/2512.01880v1 Chess, a deterministic game with perfect information, has long served as a benchmark for studying strategic decision-making and artificial intelligence. Traditional chess engines or tools for analysis primarily focus on calculating optimal moves, often neglecting the variability inherent in human chess playing, particularly across different skill levels. To overcome this limitation, we propose a novel and computationally efficient move prediction framework that approaches chess move prediction as a behavioral analysis task. The framework employs n-gram language models to capture move patterns characteristic of specific player skill levels. By dividing players into seven distinct skill groups, from novice to expert, we trained separate models using data from the open-source chess platform Lichess. The framework dynamically selects the most suitable model for prediction tasks and generates player moves based on preceding sequences. Evaluation on real-world game data demonstrates that the model selector module within the framework can classify skill levels with an accuracy of up to 31.7\% when utilizing early game information (16 half-moves). The move prediction framework also shows substantial accuracy improvements, with our Selector Assisted Accuracy being up to 39.1\% more accurate than our benchmark accuracy. The computational efficiency of the framework further enhances its suitability for real-time chess analysis. --- ### Graph Distance as Surprise: Free Energy Minimization in Knowledge Graph Reasoning Authors: Gaganpreet Jhajj, Fuhua Lin Published: 2025-12-01 Categories: cs.AI arXiv: https://arxiv.org/abs/2512.01878v1 In this work, we propose that reasoning in knowledge graph (KG) networks can be guided by surprise minimization. Entities that are close in graph distance will have lower surprise than those farther apart. This connects the Free Energy Principle (FEP) from neuroscience to KG systems, where the KG serves as the agent's generative model. We formalize surprise using the shortest-path distance in directed graphs and provide a framework for KG-based agents. Graph distance appears in graph neural networks as message passing depth and in model-based reinforcement learning as world model trajectories. This work-in-progress study explores whether distance-based surprise can extend recent work showing that syntax minimizes surprise and free energy via tree structures. --- ### Testing Transformer Learnability on the Arithmetic Sequence of Rooted Trees Authors: Alessandro Breccia, Federica Gerace, Marco Lippi, Gabriele Sicuro, Pierluigi Contucci Published: 2025-12-01 Categories: cs.AI, cond-mat.dis-nn, math-ph, math.NT arXiv: https://arxiv.org/abs/2512.01870v2 We study whether a transformer network can learn the deterministic sequence of trees generated by the iterated prime factorization of the natural numbers. Each integer is mapped into a rooted planar tree and the resulting sequence $\mathbb{N}\mathscr{T}$ defines an arithmetic text with measurable statistical structure. A transformer network (the GPT-2 architecture) is trained from scratch on the first $10^{11}$ elements and evaluated on Next-Token and masked-word prediction tasks, with a Hidden Markov Model as baseline and a scaling analysis over context window, dataset size, vocabulary size and model size. The model reaches a word accuracy of about $0.4$, well above the baseline, and its performance remains stable on test blocks located at $10^{13}$--$10^{15}$, far beyond the training interval. Moreover, the likelihood assigned by the model separates the arithmetic text from two controls: synthetic sequences reproducing its word frequencies exactly but carrying no sequential organization, with a separation that widens as the evaluated context grows; and sequences containing more than three consecutive square-free integers, a configuration that arithmetic forbids. These results indicate that the transformer captures regularities of the arithmetic text that go beyond its frequency profile. --- ### The Mean-Field Dynamics of Transformers Authors: Philippe Rigollet Published: 2025-12-01 Categories: cs.LG, math-ph, math.DS, math.PR arXiv: https://arxiv.org/abs/2512.01868v4 We develop a mathematical framework that interprets Transformer attention as an interacting particle system and studies its continuum (mean-field) limits. By idealizing attention on the sphere, we connect Transformer dynamics to Wasserstein gradient flows, synchronization models (Kuramoto), and mean-shift clustering. Central to our results is a global clustering phenomenon whereby tokens cluster asymptotically after long metastable states where they are arranged into multiple clusters. We further analyze a tractable equiangular reduction to obtain exact clustering rates, show how commonly used normalization schemes alter contraction speeds, and identify a phase transition for long-context attention. The results highlight both the mechanisms that drive representation collapse and the regimes that preserve expressive, multi-cluster structure in deep attention architectures. --- ### Cross-Lingual Interleaving for Speech Language Models Authors: Adel Moumen, Guangzhi Sun, Philip C. Woodland Published: 2025-12-01 Categories: cs.CL, cs.AI arXiv: https://arxiv.org/abs/2512.01865v2 Spoken Language Models (SLMs) aim to learn linguistic competence directly from speech using discrete units, widening access to Natural Language Processing (NLP) technologies for languages with limited written resources. However, progress has been largely English-centric due to scarce spoken evaluation benchmarks and training data, making cross-lingual learning difficult. We present a cross-lingual interleaving method that mixes speech tokens across languages without textual supervision. We also release an EN-FR training dataset, TinyStories (~42k hours), together with EN-FR spoken StoryCloze and TopicCloze benchmarks for cross-lingual semantic evaluation, both synthetically generated using GPT-4. On 360M and 1B SLMs under matched training-token budgets, interleaving improves monolingual semantic accuracy, enables robust cross-lingual continuation, and strengthens cross-lingual hidden-state alignment. Taken together, these results indicate that cross-lingual interleaving is a simple, scalable route to building multilingual SLMs that understand and converse across languages. All resources will be made open-source to support reproducibility. --- ### Topological Order in Neural Wavefunctions Authors: Ahmed Abouelkomsan, Max Geier, Liang Fu Published: 2025-12-01 Categories: cond-mat.mes-hall, cond-mat.str-el, cs.AI arXiv: https://arxiv.org/abs/2512.01863v2 Topologically ordered states are among the most interesting quantum phases of matter that host emergent quasi-particles having fractional charge and obeying fractional quantum statistics. Theoretical study of such states is however challenging owing to their strong-coupling nature that prevents conventional mean-field treatment. Here, we demonstrate that an attention-based deep neural network provides an expressive variational wavefunction that discovers fractional Chern insulator ground states purely through energy minimization without prior knowledge and achieves remarkable accuracy. We introduce an efficient method to extract ground state topological degeneracy -- a hallmark of topological order -- from a single optimized real-space wavefunction in translation-invariant systems by decomposing it into different many-body momentum sectors. Our results establish neural network variational Monte Carlo as a versatile tool for discovering strongly correlated topological phases. ---