# Research Papers — 2026-01 · bundle page 167 of 189

> Full-text markdown of papers #8301–8350 of 9403 published in 2026-01, newest first. AI-readable mirrors of the full 9403-paper 2026-01 archive (50 papers per file).

Canonical page: https://aipod.dev/papers/archive/2026-01/page/167
[← Previous bundle](https://aipod.dev/papers/archive/2026-01/page/166.md)
[Next bundle →](https://aipod.dev/papers/archive/2026-01/page/168.md)

### DNACHUNKER: Learnable Tokenization for DNA Language Models

Authors: Taewon Kim, Jihwan Shin, Hyomin Kim, Youngmok Jung, Jonhoon Lee, Won-Chul Lee, Insu Han, Sungsoo Ahn
Published: 2026-01-06
Categories: q-bio.GN, cs.CL
arXiv: https://arxiv.org/abs/2601.03019v1

DNA language models have emerged as powerful tools for decoding the complex language of DNA sequences. However, the performance of these models is heavily affected by their tokenization strategy, i.e., a method used to parse DNA sequences into a shorter sequence of chunks. In this work, we propose DNACHUNKER, which integrates a learnable dynamic DNA tokenization mechanism and is trained as a masked language model. Adopting the dynamic chunking procedure proposed by H-Net, our model learns to segment sequences into variable-length chunks. This dynamic chunking offers two key advantages: it's resilient to shifts and mutations in the DNA, and it allocates more detail to important functional areas. We demonstrate the performance of DNACHUNKER by training it on the human reference genome (HG38) and testing it on the Nucleotide Transformer and Genomic benchmarks. Further ablative experiments reveal that DNACHUNKER learns tokenization that grasps biological grammar and uses smaller chunks to preserve detail in important functional elements such as promoters and exons, while using larger chunks for repetitive, redundant regions.

---

### Dementia-R1: Reinforced Pretraining and Reasoning from Unstructured Clinical Notes for Real-World Dementia Prognosis

Authors: Choonghan Kim, Hyunmin Hwang, Hangeol Chang, Jaemin Kim, Jinse Park, Jae-Sung Lim, Jong Chul Ye
Published: 2026-01-06
Categories: cs.CL, cs.AI, cs.LG
arXiv: https://arxiv.org/abs/2601.03018v2

While Large Language Models (LLMs) have shown strong performance on clinical text understanding, they struggle with longitudinal prediction tasks such as dementia prognosis, which require reasoning over complex, non-monotonic symptom trajectories across multiple visits. Standard supervised training lacks explicit annotations for symptom evolution, while direct Reinforcement Learning (RL) is hindered by sparse binary rewards. To address this challenge, we introduce Dementia-R1, an RL-based framework for longitudinal dementia prognosis from unstructured clinical notes. Our approach adopts a Cold-Start RL strategy that pre-trains the model to predict verifiable clinical indices extracted from patient histories, enhancing the capability to reason about disease progression before determining the final clinical status. Extensive experiments show that Dementia-R1 achieves the best overall performance on the AMC real-world unstructured cohort, reaching an AUROC of 84.02% and outperforming models up to 10x larger. The framework also generalizes to Parkinson's disease dementia prediction in an independent hospital cohort, achieving an AUROC of 78.37%. On the ADNI benchmark, our 7B model attains the highest AUROC among all LLM baselines at 83.17%, demonstrating strong longitudinal reasoning over fluctuating cognitive trajectories. Code is available at https://anonymous.4open.science/r/dementiar1-CDB5.

---

### Dementia-R1: Reinforced Pretraining and Reasoning from Unstructured Clinical Notes for Real-World Dementia Prognosis

Authors: Choonghan Kim, Hyunmin Hwang, Hangeol Chang, Jaemin Kim, Jinse Park, Jae-Sung Lim, Jong Chul Ye
Published: 2026-01-06
Categories: cs.CL, cs.AI, cs.LG
arXiv: https://arxiv.org/abs/2601.03018v1

While Large Language Models (LLMs) have shown strong performance on clinical text understanding, they struggle with longitudinal prediction tasks such as dementia prognosis, which require reasoning over complex, non-monotonic symptom trajectories across multiple visits. Standard supervised training lacks explicit annotations for symptom evolution, while direct Reinforcement Learning (RL) is hindered by sparse binary rewards. To address this challenge, we introduce Dementia-R1, an RL-based framework for longitudinal dementia prognosis from unstructured clinical notes. Our approach adopts a Cold-Start RL strategy that pre-trains the model to predict verifiable clinical indices extracted from patient histories, enhancing the capability to reason about disease progression before determining the final clinical status. Extensive experiments demonstrate that Dementia-R1 achieves an F1 score of 77.03% on real-world unstructured clinical datasets. Notably, on the ADNI benchmark, our 7B model rivals GPT-4o, effectively capturing fluctuating cognitive trajectories. Code is available at https://anonymous.4open.science/r/dementiar1-CDB5

---

### MMFormalizer: Multimodal Autoformalization in the Wild

Authors: Jing Xiong, Qi Han, Yunta Hsieh, Hui Shen, Huajian Xin, Chaofan Tao, Chenyang Zhao, Hengyuan Zhang, Taiqiang Wu, Zhen Zhang, Haochen Wang, Zhongwei Wan et al.
Published: 2026-01-06
Categories: cs.CL
arXiv: https://arxiv.org/abs/2601.03017v1

Autoformalization, which translates natural language mathematics into formal statements to enable machine reasoning, faces fundamental challenges in the wild due to the multimodal nature of the physical world, where physics requires inferring hidden constraints (e.g., mass or energy) from visual elements. To address this, we propose MMFormalizer, which extends autoformalization beyond text by integrating adaptive grounding with entities from real-world mathematical and physical domains. MMFormalizer recursively constructs formal propositions from perceptually grounded primitives through recursive grounding and axiom composition, with adaptive recursive termination ensuring that every abstraction is supported by visual evidence and anchored in dimensional or axiomatic grounding. We evaluate MMFormalizer on a new benchmark, PhyX-AF, comprising 115 curated samples from MathVerse, PhyX, Synthetic Geometry, and Analytic Geometry, covering diverse multimodal autoformalization tasks. Results show that frontier models such as GPT-5 and Gemini-3-Pro achieve the highest compile and semantic accuracy, with GPT-5 excelling in physical reasoning, while geometry remains the most challenging domain. Overall, MMFormalizer provides a scalable framework for unified multimodal autoformalization, bridging perception and formal reasoning. To the best of our knowledge, this is the first multimodal autoformalization method capable of handling classical mechanics (derived from the Hamiltonian), as well as relativity, quantum mechanics, and thermodynamics. More details are available on our project page: MMFormalizer.github.io

---

### In-Context Reinforcement Learning through Bayesian Fusion of Context and Value Prior

Authors: Anaïs Berkes, Vincent Taboga, Donna Vakalis, David Rolnick, Yoshua Bengio
Published: 2026-01-06
Categories: cs.LG, cs.AI
arXiv: https://arxiv.org/abs/2601.03015v1

In-context reinforcement learning (ICRL) promises fast adaptation to unseen environments without parameter updates, but current methods either cannot improve beyond the training distribution or require near-optimal data, limiting practical adoption. We introduce SPICE, a Bayesian ICRL method that learns a prior over Q-values via deep ensemble and updates this prior at test-time using in-context information through Bayesian updates. To recover from poor priors resulting from training on sub-optimal data, our online inference follows an Upper-Confidence Bound rule that favours exploration and adaptation. We prove that SPICE achieves regret-optimal behaviour in both stochastic bandits and finite-horizon MDPs, even when pretrained only on suboptimal trajectories. We validate these findings empirically across bandit and control benchmarks. SPICE achieves near-optimal decisions on unseen tasks, substantially reduces regret compared to prior ICRL and meta-RL approaches while rapidly adapting to unseen tasks and remaining robust under distribution shift.

---

### SentGraph: Hierarchical Sentence Graph for Multi-hop Retrieval-Augmented Question Answering

Authors: Junli Liang, Pengfei Zhou, Wangqiu Zhou, Wenjie Qing, Qi Zhao, Ziwen Wang, Qi Song, Xiangyang Li
Published: 2026-01-06
Categories: cs.CL, cs.AI
arXiv: https://arxiv.org/abs/2601.03014v3

Traditional Retrieval-Augmented Generation (RAG) effectively supports single-hop question answering with large language models but faces significant limitations in multi-hop question answering tasks, which require combining evidence from multiple documents. Existing chunk-based retrieval often provides irrelevant and logically incoherent context, leading to incomplete evidence chains and incorrect reasoning during answer generation. To address these challenges, we propose SentGraph, a sentence-level graph-based RAG framework that explicitly models fine-grained logical relationships between sentences for multi-hop question answering. Specifically, we construct a hierarchical sentence graph offline by first adapting Rhetorical Structure Theory to distinguish nucleus and satellite sentences, and then organizing them into topic-level subgraphs with cross-document entity bridges. During online retrieval, SentGraph performs graph-guided evidence selection and path expansion to retrieve fine-grained sentence-level evidence. Extensive experiments on four multi-hop question answering benchmarks demonstrate the effectiveness of SentGraph, validating the importance of explicitly modeling sentence-level logical dependencies for multi-hop reasoning.

---

### SentGraph: Hierarchical Sentence Graph for Multi-hop Retrieval-Augmented Question Answering

Authors: Junli Liang, Pengfei Zhou, Wangqiu Zhou, Wenjie Qing, Qi Zhao, Ziwen Wang, Qi Song, Xiangyang Li
Published: 2026-01-06
Categories: cs.CL, cs.AI
arXiv: https://arxiv.org/abs/2601.03014v1

Traditional Retrieval-Augmented Generation (RAG) effectively supports single-hop question answering with large language models but faces significant limitations in multi-hop question answering tasks, which require combining evidence from multiple documents. Existing chunk-based retrieval often provides irrelevant and logically incoherent context, leading to incomplete evidence chains and incorrect reasoning during answer generation. To address these challenges, we propose SentGraph, a sentence-level graph-based RAG framework that explicitly models fine-grained logical relationships between sentences for multi-hop question answering. Specifically, we construct a hierarchical sentence graph offline by first adapting Rhetorical Structure Theory to distinguish nucleus and satellite sentences, and then organizing them into topic-level subgraphs with cross-document entity bridges. During online retrieval, SentGraph performs graph-guided evidence selection and path expansion to retrieve fine-grained sentence-level evidence. Extensive experiments on four multi-hop question answering benchmarks demonstrate the effectiveness of SentGraph, validating the importance of explicitly modeling sentence-level logical dependencies for multi-hop reasoning.

---

### ReCCur: A Recursive Corner-Case Curation Framework for Robust Vision-Language Understanding in Open and Edge Scenarios

Authors: Yihan Wei, Shenghai Yuan, Tianchen Deng, Boyang Lou, Enwen Hu
Published: 2026-01-06
Categories: cs.CV, cs.MA
arXiv: https://arxiv.org/abs/2601.03011v1

Corner cases are rare or extreme scenarios that drive real-world failures, but they are difficult to curate at scale: web data are noisy, labels are brittle, and edge deployments preclude large retraining. We present ReCCur (Recursive Corner-Case Curation), a low-compute framework that converts noisy web imagery into auditable fine-grained labels via a multi-agent recursive pipeline. First, large-scale data acquisition and filtering expands a domain vocabulary with a vision-language model (VLM), crawls the web, and enforces tri-modal (image, description, keyword) consistency with light human spot checks to yield refined candidates. Next, mixture-of-experts knowledge distillation uses complementary encoders (e.g., CLIP, DINOv2, BEiT) for kNN voting with dual-confidence activation and uncertainty sampling, converging to a high-precision set. Finally, region-evidence VLM adversarial labeling pairs a proposer (multi-granularity regions and semantic cues) with a validator (global and local chained consistency) to produce explainable labels and close the loop. On realistic corner-case scenarios (e.g., flooded-car inspection), ReCCur runs on consumer-grade GPUs, steadily improves purity and separability, and requires minimal human supervision, providing a practical substrate for downstream training and evaluation under resource constraints. Code and dataset will be released.

---

### JPU: Bridging Jailbreak Defense and Unlearning via On-Policy Path Rectification

Authors: Xi Wang, Songlei Jian, Shasha Li, Xiaopeng Li, Zhaoye Li, Bin Ji, Baosheng Wang, Jie Yu
Published: 2026-01-06
Categories: cs.CR, cs.AI
arXiv: https://arxiv.org/abs/2601.03005v1

Despite extensive safety alignment, Large Language Models (LLMs) often fail against jailbreak attacks. While machine unlearning has emerged as a promising defense by erasing specific harmful parameters, current methods remain vulnerable to diverse jailbreaks. We first conduct an empirical study and discover that this failure mechanism is caused by jailbreaks primarily activating non-erased parameters in the intermediate layers. Further, by probing the underlying mechanism through which these circumvented parameters reassemble into the prohibited output, we verify the persistent existence of dynamic $\textbf{jailbreak paths}$ and show that the inability to rectify them constitutes the fundamental gap in existing unlearning defenses. To bridge this gap, we propose $\textbf{J}$ailbreak $\textbf{P}$ath $\textbf{U}$nlearning (JPU), which is the first to rectify dynamic jailbreak paths towards safety anchors by dynamically mining on-policy adversarial samples to expose vulnerabilities and identify jailbreak paths. Extensive experiments demonstrate that JPU significantly enhances jailbreak resistance against dynamic attacks while preserving the model's utility.

---

### Towards Efficient 3D Object Detection for Vehicle-Infrastructure Collaboration via Risk-Intent Selection

Authors: Li Wang, Boqi Li, Hang Chen, Xingjian Wu, Yichen Wang, Jiewen Tan, Xinyu Zhang, Huaping Liu
Published: 2026-01-06
Categories: cs.CV
arXiv: https://arxiv.org/abs/2601.03001v1

Vehicle-Infrastructure Collaborative Perception (VICP) is pivotal for resolving occlusion in autonomous driving, yet the trade-off between communication bandwidth and feature redundancy remains a critical bottleneck. While intermediate fusion mitigates data volume compared to raw sharing, existing frameworks typically rely on spatial compression or static confidence maps, which inefficiently transmit spatially redundant features from non-critical background regions. To address this, we propose Risk-intent Selective detection (RiSe), an interaction-aware framework that shifts the paradigm from identifying visible regions to prioritizing risk-critical ones. Specifically, we introduce a Potential Field-Trajectory Correlation Model (PTCM) grounded in potential field theory to quantitatively assess kinematic risks. Complementing this, an Intention-Driven Area Prediction Module (IDAPM) leverages ego-motion priors to proactively predict and filter key Bird's-Eye-View (BEV) areas essential for decision-making. By integrating these components, RiSe implements a semantic-selective fusion scheme that transmits high-fidelity features only from high-interaction regions, effectively acting as a feature denoiser. Extensive experiments on the DeepAccident dataset demonstrate that our method reduces communication volume to 0.71\% of full feature sharing while maintaining state-of-the-art detection accuracy, establishing a competitive Pareto frontier between bandwidth efficiency and perception performance.

---

### Multi-Distribution Robust Conformal Prediction

Authors: Yuqi Yang, Ying Jin
Published: 2026-01-06
Categories: cs.LG, stat.ME, stat.ML
arXiv: https://arxiv.org/abs/2601.02998v2

In many fairness and distribution robustness problems, one has access to labeled data from multiple source distributions yet the test data may come from an arbitrary member or a mixture of them. We study the problem of constructing a conformal prediction set that is uniformly valid across multiple, heterogeneous distributions, in the sense that no matter which distribution the test point is from, the coverage of the prediction set is guaranteed to exceed a pre-specified level. We first propose a max-p aggregation scheme that delivers finite-sample, multi-distribution coverage given any conformity scores associated with each distribution. Upon studying several efficiency optimization programs subject to uniform coverage, we prove the optimality and tightness of our aggregation scheme, and propose a general algorithm to learn conformity scores that lead to efficient prediction sets after the aggregation under standard conditions. We discuss how our framework relates to group-wise distributionally robust optimization, sub-population shift, fairness, and multi-source learning. In synthetic and real-data experiments, our method delivers valid worst-case coverage across multiple distributions while greatly reducing the set size compared with naively applying max-p aggregation to single-source conformity scores, and can be comparable in size to single-source prediction sets with popular, standard conformity scores.

---

### Multi-Distribution Robust Conformal Prediction

Authors: Yuqi Yang, Ying Jin
Published: 2026-01-06
Categories: cs.LG, stat.ME, stat.ML
arXiv: https://arxiv.org/abs/2601.02998v1

In many fairness and distribution robustness problems, one has access to labeled data from multiple source distributions yet the test data may come from an arbitrary member or a mixture of them. We study the problem of constructing a conformal prediction set that is uniformly valid across multiple, heterogeneous distributions, in the sense that no matter which distribution the test point is from, the coverage of the prediction set is guaranteed to exceed a pre-specified level. We first propose a max-p aggregation scheme that delivers finite-sample, multi-distribution coverage given any conformity scores associated with each distribution. Upon studying several efficiency optimization programs subject to uniform coverage, we prove the optimality and tightness of our aggregation scheme, and propose a general algorithm to learn conformity scores that lead to efficient prediction sets after the aggregation under standard conditions. We discuss how our framework relates to group-wise distributionally robust optimization, sub-population shift, fairness, and multi-source learning. In synthetic and real-data experiments, our method delivers valid worst-case coverage across multiple distributions while greatly reducing the set size compared with naively applying max-p aggregation to single-source conformity scores, and can be comparable in size to single-source prediction sets with popular, standard conformity scores.

---

### From Memorization to Creativity: LLM as a Designer of Novel Neural Architectures

Authors: Waleed Khalid, Dmitry Ignatov, Radu Timofte
Published: 2026-01-06
Categories: cs.LG, cs.CV
arXiv: https://arxiv.org/abs/2601.02997v2

Large language models (LLMs) excel in program synthesis, yet their capacity for neural architecture design -- balancing syntactic reliability, performance, and structural novelty -- remains underexplored. We present a closed-loop architecture synthesis pipeline within the NNGPT framework, in which a code-oriented LLM evolves over 22 supervised fine-tuning cycles. At each cycle, the LLM synthesizes PyTorch convolutional networks, validated via low-fidelity performance signals and filtered via a MinHash--Jaccard criterion to prevent structural redundancy before being incorporated into the LEMUR dataset. High-performing candidates with novel architectures are converted into prompt--code pairs for parameter-efficient LoRA fine-tuning. This feedback loop drives a measurable distributional shift, progressively internalizing empirical architectural priors such that valid and high-performing outputs evolve from scarce to dominant across cycles. On CIFAR-10, the valid generation rate stabilizes at 50.6% (peaking at 74.5%), mean first-epoch accuracy rises from 28.1% to 51.0%, and candidates exceeding 40% accuracy grow from 2.0% to 96.8%. Cross-dataset transfer to CIFAR-100 and SVHN confirms that improved validity, shifted accuracy distributions, and sustained novelty generalize across benchmarks of varying difficulty and visual domain. Across 22 cycles, 455 unique architectures absent from the original corpus are admitted under the novelty filter. By grounding synthesis in execution feedback and novelty filtering, we demonstrate that iterative self-supervised fine-tuning reshapes an LLM into a task-specialized architectural prior -- improving generation reliability, proxy performance, and structural diversity -- offering a reproducible, annotation-free alternative to hand-crafted search spaces.

---

### From Memorization to Creativity: LLM as a Designer of Novel Neural-Architectures

Authors: Waleed Khalid, Dmitry Ignatov, Radu Timofte
Published: 2026-01-06
Categories: cs.LG, cs.CV
arXiv: https://arxiv.org/abs/2601.02997v1

Large language models (LLMs) excel in program synthesis, yet their ability to autonomously navigate neural architecture design--balancing syntactic reliability, performance, and structural novelty--remains underexplored. We address this by placing a code-oriented LLM within a closed-loop synthesis framework, analyzing its evolution over 22 supervised fine-tuning cycles. The model synthesizes PyTorch convolutional networks which are validated, evaluated via low-fidelity performance signals (single-epoch accuracy), and filtered using a MinHash-Jaccard criterion to prevent structural redundancy. High-performing, novel architectures are converted into prompt-code pairs for iterative fine-tuning via parameter-efficient LoRA adaptation, initialized from the LEMUR dataset. Across cycles, the LLM internalizes empirical architectural priors, becoming a robust generator. The valid generation rate stabilizes at 50.6 percent (peaking at 74.5 percent), while mean first-epoch accuracy rises from 28.06 percent to 50.99 percent, and the fraction of candidates exceeding 40 percent accuracy grows from 2.04 percent to 96.81 percent. Analyses confirm the model moves beyond replicating existing motifs, synthesizing 455 high-performing architectures absent from the original corpus. By grounding code synthesis in execution feedback, this work provides a scalable blueprint for transforming stochastic generators into autonomous, performance-driven neural designers, establishing that LLMs can internalize empirical, non-textual rewards to transcend their training data.

---

### Large Reasoning Models Are (Not Yet) Multilingual Latent Reasoners

Authors: Yihong Liu, Raoyuan Zhao, Hinrich Schütze, Michael A. Hedderich
Published: 2026-01-06
Categories: cs.CL
arXiv: https://arxiv.org/abs/2601.02996v2

Large reasoning models (LRMs) achieve strong performance on mathematical reasoning tasks, often attributed to their capability to generate explicit chain-of-thought (CoT) explanations. However, recent work shows that LRMs often arrive at the correct answer before completing these textual reasoning steps, indicating the presence of latent reasoning -- internal, non-verbal computation encoded in hidden states. While this phenomenon has been explored in English, its multilingual behavior remains largely unknown. In this paper, we conduct a systematic investigation of multilingual latent reasoning in LRMs across 11 languages. Using a truncation-based strategy, we examine how the correct answer emerges as the model is given only partial reasoning traces, allowing us to measure stepwise latent prediction formation. Our results reveal clear evidence of multilingual latent reasoning, though unevenly: strong in resource-rich languages, weaker in low-resource ones, and broadly less observable on harder benchmarks. To understand whether these differences reflect distinct internal mechanisms, we further perform representational analyses. Despite surface-level disparities, we find that the internal evolution of predictions is highly consistent across languages and broadly aligns with English -- a pattern suggesting an English-centered latent reasoning pathway.

---

### Learning to Act Robustly with View-Invariant Latent Actions

Authors: Youngjoon Jeong, Junha Chun, Taesup Kim
Published: 2026-01-06
Categories: cs.RO, cs.AI, cs.LG
arXiv: https://arxiv.org/abs/2601.02994v1

Vision-based robotic policies often struggle with even minor viewpoint changes, underscoring the need for view-invariant visual representations. This challenge becomes more pronounced in real-world settings, where viewpoint variability is unavoidable and can significantly disrupt policy performance. Existing methods typically learn invariance from multi-view observations at the scene level, but such approaches rely on visual appearance and fail to incorporate the physical dynamics essential for robust generalization. We propose View-Invariant Latent Action (VILA), which models a latent action capturing transition patterns across trajectories to learn view-invariant representations grounded in physical dynamics. VILA aligns these latent actions across viewpoints using an action-guided objective based on ground-truth action sequences. Experiments in both simulation and the real world show that VILA-based policies generalize effectively to unseen viewpoints and transfer well to new tasks, establishing VILA as a strong pretraining framework that improves robustness and downstream learning performance.

---

### Stable-RAG: Mitigating Retrieval-Permutation-Induced Hallucinations in Retrieval-Augmented Generation

Authors: Qianchi Zhang, Hainan Zhang, Liang Pang, Hongwei Zheng, Zhiming Zheng
Published: 2026-01-06
Categories: cs.CL
arXiv: https://arxiv.org/abs/2601.02993v4

Retrieval-Augmented Generation (RAG) has become a key paradigm for reducing factual hallucinations in Large Language Models (LLMs), yet little is known about how the order of retrieved documents affects model behavior. We empirically show that under a Top-5 retrieval setting with the gold document included, LLM answers vary substantially across permutations of the retrieved set, even when the gold document is fixed in the first position. This reveals a previously underexplored sensitivity to retrieval permutations. Although existing robust RAG methods focus primarily on enhancing LLM robustness to low-quality retrieval and mitigating positional bias to distribute attention fairly over long contexts, neither approach directly addresses permutation sensitivity. In this paper, we propose Stable-RAG, which exploits permutation sensitivity estimation to mitigate permutation-induced hallucinations. Stable-RAG runs the generator under multiple retrieval orders, clusters hidden states, and decodes from a cluster-center representation that captures the dominant reasoning pattern. It then uses these reasoning results to align hallucinated outputs toward the correct answer, encouraging the model to produce consistent and accurate predictions across document permutations. Experiments on three QA datasets show that Stable-RAG improves answer accuracy, reasoning consistency, and generalization across datasets, retrievers, and input lengths compared with strong baselines.

---

### Towards Faithful Reasoning in Comics for Small MLLMs

Authors: Chengcheng Feng, Haojie Yin, Yucheng Jin, Kaizhu Huang
Published: 2026-01-06
Categories: cs.CV, cs.AI
arXiv: https://arxiv.org/abs/2601.02991v2

Comic understanding presents a significant challenge for Multimodal Large Language Models (MLLMs), as the intended meaning of a comic often emerges from the joint interpretation of visual, textual, and social cues. This naturally motivates Chain-of-Thought (CoT) prompting, since explicit intermediate reasoning appears promising for integrating such heterogeneous signals. However, existing CoT methods are poorly matched to this structure: they tend to force interpretation into a single reasoning path before multiple cues have been jointly considered, often degrading performance, especially for small MLLMs. Our key idea is to explicitly preserve multi-cue interpretation during supervision construction, rather than collapsing comic understanding into a single reasoning chain. To this end, we propose a two-stage framework for faithful comic reasoning in small MLLMs. First, we introduce MoCoT, a modular supervision construction framework that preserves multi-cue interpretation and turns it into more faithful supervision. Second, we propose VERA, a structured reward mechanism that turns such supervision into faithful reasoning behavior by aligning optimization with both reasoning faithfulness and answer correctness. Extensive experiments on five benchmarks spanning comic understanding and broader humor-centric and abstract visual reasoning tasks demonstrate that our framework achieves strong results in the $\leq$ 4B regime, surpasses several 7B baselines, improves four small MLLMs by an average of $\mathbf{12.1%}$ as a plug-in, and consistently enhances reasoning faithfulness while preserving inference efficiency.

---

### Mechanistic Interpretability of Large-Scale Counting in LLMs through a System-2 Strategy

Authors: Hosein Hasani, Mohammadali Banayeeanzade, Ali Nafisi, Sadegh Mohammadian, Fatemeh Askari, Mobin Bagherian, Amirmohammad Izadi, Mahdieh Soleymani Baghshah
Published: 2026-01-06
Categories: cs.CL
arXiv: https://arxiv.org/abs/2601.02989v2

Large language models (LLMs), despite strong performance on complex mathematical problems, exhibit systematic limitations in counting tasks. This issue arises from the architectural limits of transformers, where counting is performed across layers, leading to degraded precision for larger counting problems due to depth constraints. To address this limitation, we propose a simple test-time strategy inspired by System-2 cognitive processes that decomposes large counting tasks into smaller, independent sub-problems that the model can reliably solve. We evaluate this approach using observational and causal mediation analyses to understand the underlying mechanism of this System-2-like strategy. Our mechanistic analysis identifies key components: latent counts are computed and stored in the final item representations of each part, transferred to intermediate steps via dedicated attention heads, and aggregated in the final stage to produce the total count. Experimental results demonstrate that this strategy enables LLMs to surpass architectural limitations and achieve higher accuracy on large-scale counting tasks. This work provides mechanistic insight into System-2 counting in LLMs and presents a generalizable approach for improving and understanding their reasoning behavior.

---

### ULS+: Data-driven Model Adaptation Enhances Lesion Segmentation

Authors: Rianne Weber, Niels Rocholl, Max de Grauw, Mathias Prokop, Ewoud Smit, Alessa Hering
Published: 2026-01-06
Categories: cs.CV, cs.AI
arXiv: https://arxiv.org/abs/2601.02988v1

In this study, we present ULS+, an enhanced version of the Universal Lesion Segmentation (ULS) model. The original ULS model segments lesions across the whole body in CT scans given volumes of interest (VOIs) centered around a click-point. Since its release, several new public datasets have become available that can further improve model performance. ULS+ incorporates these additional datasets and uses smaller input image sizes, resulting in higher accuracy and faster inference. We compared ULS and ULS+ using the Dice score and robustness to click-point location on the ULS23 Challenge test data and a subset of the Longitudinal-CT dataset. In all comparisons, ULS+ significantly outperformed ULS. Additionally, ULS+ ranks first on the ULS23 Challenge test-phase leaderboard. By maintaining a cycle of data-driven updates and clinical validation, ULS+ establishes a foundation for robust and clinically relevant lesion segmentation models.

---

### LAMS-Edit: Latent and Attention Mixing with Schedulers for Improved Content Preservation in Diffusion-Based Image and Style Editing

Authors: Wingwa Fu, Takayuki Okatani
Published: 2026-01-06
Categories: cs.CV, cs.AI
arXiv: https://arxiv.org/abs/2601.02987v1

Text-to-Image editing using diffusion models faces challenges in balancing content preservation with edit application and handling real-image editing. To address these, we propose LAMS-Edit, leveraging intermediate states from the inversion process--an essential step in real-image editing--during edited image generation. Specifically, latent representations and attention maps from both processes are combined at each step using weighted interpolation, controlled by a scheduler. This technique, Latent and Attention Mixing with Schedulers (LAMS), integrates with Prompt-to-Prompt (P2P) to form LAMS-Edit--an extensible framework that supports precise editing with region masks and enables style transfer via LoRA. Extensive experiments demonstrate that LAMS-Edit effectively balances content preservation and edit application.

---

### P-Check: Advancing Personalized Reward Model via Learning to Generate Dynamic Checklist

Authors: Kwangwook Seo, Dongha Lee
Published: 2026-01-06
Categories: cs.CL
arXiv: https://arxiv.org/abs/2601.02986v3

Recent approaches in personalized reward modeling have primarily focused on leveraging user interaction history to align model judgments with individual preferences. However, existing approaches largely treat user context as a static or implicit conditioning signal, failing to capture the dynamic and multi-faceted nature of human judgment. In this paper, we propose P-Check, a novel personalized reward modeling framework, designed to train a plug-and-play checklist generator that synthesizes dynamic evaluation criteria for guiding the reward prediction. To better align these checklists with personalized nuances, we introduce Preference-Contrastive Criterion Weighting, a training strategy that assigns saliency scores to criteria based on their discriminative power for personalized judgment. We conduct extensive experiments and demonstrate that P-Check not only improves reward accuracy but also enhances downstream personalized generation, and remains robust in OOD scenarios.

---

### Interpretable All-Type Audio Deepfake Detection with Audio LLMs via Frequency-Time Reinforcement Learning

Authors: Yuankun Xie, Xiaoxuan Guo, Jiayi Zhou, Tao Wang, Jian Liu, Ruibo Fu, Xiaopeng Wang, Haonan Cheng, Long Ye
Published: 2026-01-06
Categories: cs.SD, cs.AI
arXiv: https://arxiv.org/abs/2601.02983v1

Recent advances in audio large language models (ALLMs) have made high-quality synthetic audio widely accessible, increasing the risk of malicious audio deepfakes across speech, environmental sounds, singing voice, and music. Real-world audio deepfake detection (ADD) therefore requires all-type detectors that generalize across heterogeneous audio and provide interpretable decisions. Given the strong multi-task generalization ability of ALLMs, we first investigate their performance on all-type ADD under both supervised fine-tuning (SFT) and reinforcement fine-tuning (RFT). However, SFT using only binary real/fake labels tends to reduce the model to a black-box classifier, sacrificing interpretability. Meanwhile, vanilla RFT under sparse supervision is prone to reward hacking and can produce hallucinated, ungrounded rationales. To address this, we propose an automatic annotation and polishing pipeline that constructs Frequency-Time structured chain-of-thought (CoT) rationales, producing ~340K cold-start demonstrations. Building on CoT data, we propose Frequency Time-Group Relative Policy Optimization (FT-GRPO), a two-stage training paradigm that cold-starts ALLMs with SFT and then applies GRPO under rule-based frequency-time constraints. Experiments demonstrate that FT-GRPO achieves state-of-the-art performance on all-type ADD while producing interpretable, FT-grounded rationales. The data and code are available online.

---

### Mechanistic Knobs in LLMs: Retrieving and Steering High-Order Semantic Features via Sparse Autoencoders

Authors: Ruikang Zhang, Shuo Wang, Qi Su
Published: 2026-01-06
Categories: cs.CL, cs.AI
arXiv: https://arxiv.org/abs/2601.02978v3

Recent work in Mechanistic Interpretability (MI) has enabled the identification and intervention of internal features in Large Language Models (LLMs). However, a persistent challenge lies in linking such internal features to the reliable control of complex, behavior-level semantic attributes in language generation. In this paper, we propose a Sparse Autoencoder-based framework for retrieving and steering semantically interpretable internal features associated with high-level linguistic behaviors. Our method employs a contrastive feature retrieval pipeline based on controlled semantic oppositions, combing statistical activation analysis and generation-based validation to distill monosemantic functional features from sparse activation spaces. Using the Big Five personality traits as a case study, we demonstrate that our method enables precise, bidirectional steering of model behavior while maintaining superior stability and performance compared to existing activation steering methods like Contrastive Activation Addition (CAA). We further identify an empirical effect, which we term Functional Faithfulness, whereby intervening on a specific internal feature induces coherent and predictable shifts across multiple linguistic dimensions aligned with the target semantic attribute. Our findings suggest that LLMs internalize deeply integrated representations of high-order concepts, and provide a novel, robust mechanistic path for the regulation of complex AI behaviors.

---

### Correct, Concise and Complete: Multi-stage Training For Adaptive Reasoning

Authors: Nathanaël Carraz Rakotonirina, Ren Pang, Neha Anna John, Michael Bohlke-Schneider, Momchil Hardalov
Published: 2026-01-06
Categories: cs.CL, cs.AI
arXiv: https://arxiv.org/abs/2601.02972v1

The reasoning capabilities of large language models (LLMs) have improved substantially through increased test-time computation, typically in the form of intermediate tokens known as chain-of-thought (CoT). However, CoT often becomes unnecessarily long, increasing computation cost without actual accuracy gains or sometimes even degrading performance, a phenomenon known as ``overthinking''. We propose a multi-stage efficient reasoning method that combines supervised fine-tuning -- via rejection sampling or reasoning trace reformatting -- with reinforcement learning using an adaptive length penalty. We introduce a lightweight reward function that penalizes tokens generated after the first correct answer but encouraging self-verification only when beneficial. We conduct a holistic evaluation across seven diverse reasoning tasks, analyzing the accuracy-response length trade-off. Our approach reduces response length by an average of 28\% for 8B models and 40\% for 32B models, while incurring only minor performance drops of 1.6 and 2.5 points, respectively. Despite its conceptual simplicity, it achieves a superior trade-off compared to more complex state-of-the-art efficient reasoning methods, scoring 76.6, in terms of the area under the Overthinking-Adjusted Accuracy curve ($\text{AUC}_{\text{OAA}}$) -- 5 points above the base model and 2.5 points above the second-best approach.

---

### Reliability-Aware Adaptive Self-Consistency for Efficient Sampling in LLM Reasoning

Authors: Junseok Kim, Nakyeong Yang, Kyungmin Min, Kyomin Jung
Published: 2026-01-06
Categories: cs.CL, cs.LG
arXiv: https://arxiv.org/abs/2601.02970v2

Self-Consistency improves reasoning reliability through multi-sample aggregation, but incurs substantial inference cost. Adaptive self-consistency methods mitigate this issue by adjusting the sampling budget; however, they rely on count-based stopping rules that treat all responses equally, often leading to unnecessary sampling. We propose Reliability-Aware Adaptive Self-Consistency (ReASC), which addresses this limitation by reframing adaptive sampling from response counting to evidence sufficiency, leveraging response-level confidence for principled information aggregation. ReASC operates in two stages: a single-sample decision stage that resolves instances confidently answerable from a single response, and a reliability-aware accumulation stage that aggregates responses by jointly leveraging their frequency and confidence. Across five models and four datasets, ReASC consistently achieves the best accuracy-cost trade-off compared to existing baselines, yielding improved inference efficiency across model scales from 3B to 27B parameters. As a concrete example, ReASC reduces inference cost by up to 70\% relative to self-consistency while preserving accuracy on GSM8K using Gemma-3-4B-it.

---

### Rationale-Grounded In-Context Learning for Time Series Reasoning with Multimodal Large Language Models

Authors: Qingxiang Liu, Zhiqing Cui, Xiaoliang Luo, Yuqian Wu, Zhuoyang Jiang, Huaiyu Wan, Sheng Sun, Lvchun Wang, Wei Yu, Yuxuan Liang
Published: 2026-01-06
Categories: cs.AI
arXiv: https://arxiv.org/abs/2601.02968v1

The underperformance of existing multimodal large language models for time series reasoning lies in the absence of rationale priors that connect temporal observations to their downstream outcomes, which leads models to rely on superficial pattern matching rather than principled reasoning. We therefore propose the rationale-grounded in-context learning for time series reasoning, where rationales work as guiding reasoning units rather than post-hoc explanations, and develop the RationaleTS method. Specifically, we firstly induce label-conditioned rationales, composed of reasoning paths from observable evidence to the potential outcomes. Then, we design the hybrid retrieval by balancing temporal patterns and semantic contexts to retrieve correlated rationale priors for the final in-context inference on new samples. We conduct extensive experiments to demonstrate the effectiveness and efficiency of our proposed RationaleTS on three-domain time series reasoning tasks. We will release our code for reproduction.

---

### MoE Adapter for Large Audio Language Models: Sparsity, Disentanglement, and Gradient-Conflict-Free

Authors: Yishu Lei, Shuwei He, Jing Hu, Dan Zhang, Xianlong Luo, Danxiang Zhu, Shikun Feng, Rui Liu, Jingzhou He, Yu Sun, Hua Wu, Haifeng Wang
Published: 2026-01-06
Categories: cs.SD, cs.AI, eess.AS
arXiv: https://arxiv.org/abs/2601.02967v2

Extending the input modality of Large Language Models~(LLMs) to the audio domain is essential for achieving comprehensive multimodal perception. However, it is well-known that acoustic information is intrinsically \textit{heterogeneous}, entangling attributes such as speech, music, and environmental context. Existing research is limited to a dense, parameter-shared adapter to model these diverse patterns, which induces \textit{gradient conflict} during optimization, as parameter updates required for distinct attributes contradict each other. To address this limitation, we introduce the \textit{\textbf{MoE-Adapter}}, a sparse Mixture-of-Experts~(MoE) architecture designed to decouple acoustic information. Specifically, it employs a dynamic gating mechanism that routes audio tokens to specialized experts capturing complementary feature subspaces while retaining shared experts for global context, thereby mitigating gradient conflicts and enabling fine-grained feature learning. Comprehensive experiments show that the MoE-Adapter achieves superior performance on both audio semantic and paralinguistic tasks, consistently outperforming dense linear baselines with comparable computational costs. Furthermore, we will release the related code and models to facilitate future research.

---

### Low-Resource Heuristics for Bahnaric Optical Character Recognition Improvement

Authors: Phat Tran, Phuoc Pham, Hung Trinh, Tho Quan
Published: 2026-01-06
Categories: cs.CL, cs.CV, cs.LG
arXiv: https://arxiv.org/abs/2601.02965v1

Bahnar, a minority language spoken across Vietnam, Cambodia, and Laos, faces significant preservation challenges due to limited research and data availability. This study addresses the critical need for accurate digitization of Bahnar language documents through optical character recognition (OCR) technology. Digitizing scanned paper documents poses significant challenges, as degraded image quality from broken or blurred areas introduces considerable OCR errors that compromise information retrieval systems. We propose a comprehensive approach combining advanced table and non-table detection techniques with probability-based post-processing heuristics to enhance recognition accuracy. Our method first applies detection algorithms to improve input data quality, then employs probabilistic error correction on OCR output. Experimental results indicate a substantial improvement, with recognition accuracy increasing from 72.86% to 79.26%. This work contributes valuable resources for Bahnar language preservation and provides a framework applicable to other minority language digitization efforts.

---

### LLM-Augmented Changepoint Detection: A Framework for Ensemble Detection and Automated Explanation

Authors: Fabian Lukassen, Christoph Weisser, Michael Schlee, Manish Kumar, Anton Thielmann, Benjamin Saefken, Alexander Silbersdorff, Thomas Kneib
Published: 2026-01-06
Categories: cs.CL
arXiv: https://arxiv.org/abs/2601.02957v3

This paper introduces a novel changepoint detection framework that combines ensemble statistical methods with Large Language Models (LLMs) to enhance both detection accuracy and the interpretability of regime changes in time series data. Two critical limitations in the field are addressed. First, individual detection methods exhibit complementary strengths and weaknesses depending on data characteristics, making method selection non-trivial and prone to suboptimal results. Second, automated, contextual explanations for detected changes are largely absent. The proposed ensemble method aggregates results from ten distinct changepoint detection algorithms, achieving superior performance and robustness compared to individual methods. Additionally, an LLM-powered explanation pipeline automatically generates contextual narratives, linking detected changepoints to potential real-world historical events. For private or domain-specific data, a Retrieval-Augmented Generation (RAG) solution enables explanations grounded in user-provided documents. The open source Python framework demonstrates practical utility in diverse domains, including finance, political science, and environmental science, transforming raw statistical output into actionable insights for analysts and decision-makers.

---

### Enhancing Multilingual RAG Systems with Debiased Language Preference-Guided Query Fusion

Authors: Jeonghyun Park, Byeongjeong Kim, Seojin Hwang, Hwanhee Lee
Published: 2026-01-06
Categories: cs.CL
arXiv: https://arxiv.org/abs/2601.02956v3

Multilingual Retrieval-Augmented Generation (mRAG) systems often exhibit a perceived preference for high-resource languages, particularly English, resulting in the widespread adoption of English pivoting. While prior studies attribute this advantage to the superior English-centric capabilities of Large Language Models (LLMs), we find that such measurements are significantly distorted by structural priors inherent in evaluation benchmarks. Specifically, we identify exposure bias and a gold availability prior-both driven by the disproportionate concentration of resources in English-as well as cultural priors rooted in topic locality, as factors that hinder accurate assessment of genuine language preference. To address these biases, we propose DeLP (Debiased Language Preference), a calibrated metric designed to explicitly factor out these structural confounds. Our analysis using DeLP reveals that the previously reported English preference is largely a byproduct of evidence distribution rather than an inherent model bias. Instead, we find that retrievers fundamentally favor monolingual alignment between the query and the document language. Building on this insight, we introduce DELTA (DEbiased Language preference-guided Text Augmentation), a lightweight and efficient mRAG framework that strategically leverages monolingual alignment to optimize cross-lingual retrieval and generation. Experimental results demonstrate that DELTA consistently outperforms English pivoting and mRAG baselines across diverse languages.

---

### The World is Not Mono: Enabling Spatial Understanding in Large Audio-Language Models

Authors: Yuhuan You, Lai Wei, Xihong Wu, Tianshu Qu
Published: 2026-01-06
Categories: cs.SD, cs.AI
arXiv: https://arxiv.org/abs/2601.02954v3

Large audio-language models have made rapid progress in recognizing what is present in an audio clip, but spatial audio-language understanding still lacks a clear task interface. A model must also decide where sound events occur, which semantic and spatial attributes belong to the same auditory object, how multiple objects are arranged, and whether a scene-level answer is physically plausible. We formalize this capability as audio scene analysis (ASA), a three-level problem spanning atomic perception, relational integration, and cognitive reasoning. We propose The World is Not Mono (TWNM), a framework that equips audio-language models with explicit spatial evidence. TWNM uses physically grounded First-Order Ambisonics (FOA) simulation for controllable supervision, learns slot-regularized spatial representations from multichannel audio, fuses them with semantic audio features, and trains with a progressive curriculum ending in preference optimization over metadata-derived answers and auxiliary format/evidence rewards. To operationalize ASA, we build a controlled benchmark from scene metadata, covering localization, attribute binding, spatial comparison, scene abduction, and counterfactual reasoning. On this benchmark, TWNM achieves 70.8% overall accuracy, 66.4% on spatial-family tasks, and 79.76% on mixed L3 scene-level multiple-choice QA. We also audit monaural and binaural reference systems as diagnostic references with explicit audit labels, since they differ in spatial input, training interface, and output format. The supported claim is that a clearly defined ASA hierarchy, FOA-conditioned spatial representations, and metadata-grounded training enable controlled, auditable spatial audio-language reasoning, with STARSS23 providing a limited real-recording diagnostic.

---

### Batch-of-Thought: Cross-Instance Learning for Enhanced LLM Reasoning

Authors: Xuan Yang, Furong Jia, Roy Xie, Xiong Xi, Hengwei Bian, Jian Li, Monica Agrawal
Published: 2026-01-06
Categories: cs.AI
arXiv: https://arxiv.org/abs/2601.02950v3

Current Large Language Model reasoning systems process queries independently, discarding valuable cross-instance signals such as shared reasoning patterns and consistency constraints. We introduce Batch-of-Thought (BoT), a training-free method that processes related queries jointly to enable cross-instance learning. By performing comparative analysis across batches, BoT identifies high-quality reasoning templates, detects errors through consistency checks, and amortizes computational costs. We instantiate BoT within a multi-agent reflection architecture (BoT-R), where a Reflector performs joint evaluation to unlock mutual information gain unavailable in isolated processing. Experiments across three model families and six benchmarks demonstrate that BoT-R consistently improves accuracy and confidence calibration while reducing inference costs by up to 61%. Our theoretical and experimental analysis reveals when and why batch-aware reasoning benefits LLM systems. Our code is available at https://github.com/xuanyang19/BoT

---

### VTONQA: A Multi-Dimensional Quality Assessment Dataset for Virtual Try-on

Authors: Xinyi Wei, Sijing Wu, Zitong Xu, Yunhao Li, Huiyu Duan, Jia Wang, Ning Liu
Published: 2026-01-06
Categories: cs.CV
arXiv: https://arxiv.org/abs/2601.02945v2

With the rapid development of e-commerce and digital fashion, image-based virtual try-on (VTON) has attracted increasing attention. However, existing VTON models often suffer from artifacts such as garment distortion and body inconsistency, highlighting the need for reliable quality evaluation of VTON-generated images. To this end, we construct \textbf{VTONQA}, the first multi-dimensional quality assessment dataset specifically designed for VTON, which contains 8,132 images generated by 11 representative VTON models, along with 24,396 mean opinion scores (MOSs) across three evaluation dimensions (\textit{i.e.}, clothing fit, body compatibility, and overall quality). Based on VTONQA, we benchmark both VTON models and a diverse set of image quality assessment (IQA) metrics, revealing the limitations of existing methods and highlighting the value of the proposed dataset. We believe that the VTONQA dataset and corresponding benchmarks will provide a solid foundation for perceptually aligned evaluation, benefiting both the development of quality assessment methods and the advancement of VTON models. The dataset we proposed in this paper is publicly available at: https://huggingface.co/datasets/weixiny0408/vtonqa.

---

### MixTTE: Multi-Level Mixture-of-Experts for Scalable and Adaptive Travel Time Estimation

Authors: Wenzhao Jiang, Jindong Han, Ruiqian Han, Hao Liu
Published: 2026-01-06
Categories: cs.LG, cs.MA
arXiv: https://arxiv.org/abs/2601.02943v1

Accurate Travel Time Estimation (TTE) is critical for ride-hailing platforms, where errors directly impact user experience and operational efficiency. While existing production systems excel at holistic route-level dependency modeling, they struggle to capture city-scale traffic dynamics and long-tail scenarios, leading to unreliable predictions in large urban networks. In this paper, we propose \model, a scalable and adaptive framework that synergistically integrates link-level modeling with industrial route-level TTE systems. Specifically, we propose a spatio-temporal external attention module to capture global traffic dynamic dependencies across million-scale road networks efficiently. Moreover, we construct a stabilized graph mixture-of-experts network to handle heterogeneous traffic patterns while maintaining inference efficiency. Furthermore, an asynchronous incremental learning strategy is tailored to enable real-time and stable adaptation to dynamic traffic distribution shifts. Experiments on real-world datasets validate MixTTE significantly reduces prediction errors compared to seven baselines. MixTTE has been deployed in DiDi, substantially improving the accuracy and stability of the TTE service.

---

### SastBench: A Benchmark for Testing Agentic SAST Triage

Authors: Jake Feiglin, Guy Dar
Published: 2026-01-06
Categories: cs.CR, cs.AI, cs.CL
arXiv: https://arxiv.org/abs/2601.02941v1

SAST (Static Application Security Testing) tools are among the most widely used techniques in defensive cybersecurity, employed by commercial and non-commercial organizations to identify potential vulnerabilities in software. Despite their great utility, they generate numerous false positives, requiring costly manual filtering (aka triage). While LLM-powered agents show promise for automating cybersecurity tasks, existing benchmarks fail to emulate real-world SAST finding distributions. We introduce SastBench, a benchmark for evaluating SAST triage agents that combines real CVEs as true positives with filtered SAST tool findings as approximate false positives. SastBench features an agent-agnostic design. We evaluate different agents on the benchmark and present a comparative analysis of their performance, provide a detailed analysis of the dataset, and discuss the implications for future development.

---

### Pearmut: Human Evaluation of Translation Made Trivial

Authors: Vilém Zouhar, Tom Kocmi
Published: 2026-01-06
Categories: cs.CL, cs.HC
arXiv: https://arxiv.org/abs/2601.02933v4

Human evaluation is the gold standard for multilingual NLP, but is often skipped in practice and substituted with automatic metrics because it is notoriously complex and slow to set up with existing tools with substantial engineering and operational overhead. We introduce Pearmut, a lightweight yet feature-rich platform that makes end-to-end human evaluation as easy to run as automatic evaluation. Pearmut removes common entry barriers and provides support for evaluating multilingual tasks, with a particular focus on machine translation. The platform implements standard evaluation protocols, including DA, ESA, and MQM, and is extensible to support new protocols. It features document-level context, absolute and contrastive evaluation, attention checks, ESAAI pre-annotations and both static and dynamic assignment strategies. Pearmut enables reliable human evaluation to become a practical, routine component of model development and diagnosis rather than an occasional effort.

---

### Memorization, Emergence, and Explaining Reversal Failures: A Controlled Study of Relational Semantics in LLMs

Authors: Yihua Zhu, Qianying Liu, Jiaxin Wang, Fei Cheng, Chaoran Liu, Akiko Aizawa, Sadao Kurohashi, Hidetoshi Shimodaira
Published: 2026-01-06
Categories: cs.CL
arXiv: https://arxiv.org/abs/2601.02931v2

Autoregressive LLMs perform well on relational tasks that require linking entities via relational words (e.g., father/son, friend), but it is unclear whether they learn the logical semantics of such relations (e.g., symmetry and inversion logic) and, if so, whether reversal-type failures arise from missing relational semantics or left-to-right order bias. We propose a controlled Knowledge Graph-based synthetic framework that generates text from symmetric/inverse triples, train GPT-style autoregressive models from scratch, and evaluate memorization, logical inference, and in-context generalization to unseen entities to address these questions. We find a sharp phase transition in which relational semantics emerge with sufficient logic-bearing supervision, even in shallow (2-3 layer) models, and that successful generalization aligns with stable intermediate-layer signals. Finally, order-matched forward/reverse tests and a diffusion baseline indicate that reversal failures are primarily driven by autoregressive order bias rather than deficient inversion semantics.

---

### HybridSolarNet: A Lightweight and Explainable EfficientNet-CBAM Architecture for Real-Time Solar Panel Fault Detection

Authors: Md. Asif Hossain, G M Mota-Tahrin Tayef, Nabil Subhan
Published: 2026-01-06
Categories: cs.CV
arXiv: https://arxiv.org/abs/2601.02928v1

Manual inspections for solar panel systems are a tedious, costly, and error-prone task, making it desirable for Unmanned Aerial Vehicle (UAV) based monitoring. Though deep learning models have excellent fault detection capabilities, almost all methods either are too large and heavy for edge computing devices or involve biased estimation of accuracy due to ineffective learning techniques. We propose a new solar panel fault detection model called HybridSolarNet. It integrates EfficientNet-B0 with Convolutional Block Attention Module (CBAM). We implemented it on the Kaggle Solar Panel Images competition dataset with a tight split-before-augmentation protocol. It avoids leakage in accuracy estimation. We introduced focal loss and cosine annealing. Ablation analysis validates that accuracy boosts due to added benefits from CBAM (+1.53%) and that there are benefits from recognition of classes with imbalanced samples via focal loss. Overall average accuracy on 5-fold stratified cross-validation experiments on the given competition dataset topped 92.37% +/- 0.41 and an F1-score of 0.9226 +/- 0.39 compared to baselines like VGG19, requiring merely 16.3 MB storage, i.e., 32 times less. Its inference speed measured at 54.9 FPS with GPU support makes it a successful candidate for real-time UAV implementation. Moreover, visualization obtained from Grad-CAM illustrates that HybridSolarNet focuses on actual locations instead of irrelevant ones.

---

### PrismVAU: Prompt-Refined Inference System for Multimodal Video Anomaly Understanding

Authors: Iñaki Erregue, Kamal Nasrollahi, Sergio Escalera
Published: 2026-01-06
Categories: cs.CV, cs.AI
arXiv: https://arxiv.org/abs/2601.02927v2

Video Anomaly Understanding (VAU) extends traditional Video Anomaly Detection (VAD) by not only localizing anomalies but also describing and reasoning about their context. Existing VAU approaches often rely on fine-tuned multimodal large language models (MLLMs) or external modules such as video captioners, which introduce costly annotations, complex training pipelines, and high inference overhead. In this work, we introduce PrismVAU, a lightweight yet effective system for real-time VAU that leverages a single off-the-shelf MLLM for anomaly scoring, explanation, and prompt optimization. PrismVAU operates in two complementary stages: (1) a coarse anomaly scoring module that computes frame-level anomaly scores via similarity to textual anchors, and (2) an MLLM-based refinement module that contextualizes anomalies through system and user prompts. Both textual anchors and prompts are optimized with a weakly supervised Automatic Prompt Engineering (APE) framework. Extensive experiments on standard VAD benchmarks demonstrate that PrismVAU delivers competitive detection performance and interpretable anomaly explanations -- without relying on instruction tuning, frame-level annotations, and external modules or dense processing -- making it an efficient and practical solution for real-world applications.

---

### DCG ReID: Disentangling Collaboration and Guidance Fusion Representations for Multi-modal Vehicle Re-Identification

Authors: Aihua Zheng, Ya Gao, Shihao Li, Chenglong Li, Jin Tang
Published: 2026-01-06
Categories: cs.CV, cs.AI
arXiv: https://arxiv.org/abs/2601.02924v1

Multi-modal vehicle Re-Identification (ReID) aims to leverage complementary information from RGB, Near Infrared (NIR), and Thermal Infrared (TIR) modalities to retrieve the same vehicle. The challenges of multi-modal vehicle ReID arise from the uncertainty of modality quality distribution induced by inherent discrepancies across modalities, resulting in distinct conflicting fusion requirements for data with balanced and unbalanced quality distributions. Existing methods handle all multi-modal data within a single fusion model, overlooking the different needs of the two data types and making it difficult to decouple the conflict between intra-class consistency and inter-modal heterogeneity. To this end, we propose Disentangle Collaboration and Guidance Fusion Representations for Multi-modal Vehicle ReID (DCG-ReID). Specifically, to disentangle heterogeneous quality-distributed modal data without mutual interference, we first design the Dynamic Confidence-based Disentangling Weighting (DCDW) mechanism: dynamically reweighting three-modal contributions via interaction-derived modal confidence to build a disentangled fusion framework. Building on DCDW, we develop two scenario-specific fusion strategies: (1) for balanced quality distributions, Collaboration Fusion Module (CFM) mines pairwise consensus features to capture shared discriminative information and boost intra-class consistency; (2) for unbalanced distributions, Guidance Fusion Module (GFM) implements differential amplification of modal discriminative disparities to reinforce dominant modality advantages, guide auxiliary modalities to mine complementary discriminative info, and mitigate inter-modal divergence to boost multi-modal joint decision performance. Extensive experiments on three multi-modal ReID benchmarks (WMVeID863, MSVR310, RGBNT100) validate the effectiveness of our method. Code will be released upon acceptance.

---

### Zoom-IQA: Image Quality Assessment with Reliable Region-Aware Reasoning

Authors: Guoqiang Liang, Jianyi Wang, Zhonghua Wu, Shangchen Zhou, Chen Change Loy
Published: 2026-01-06
Categories: cs.CV
arXiv: https://arxiv.org/abs/2601.02918v5

Image Quality Assessment (IQA) is a long-standing problem in computer vision. Previous methods typically focus on predicting numerical scores without explanation or providing low-level descriptions lacking precise scores. Recent reasoning-based vision language models (VLMs) have shown strong potential for IQA by jointly generating quality descriptions and scores. However, existing VLM-based IQA methods often suffer from unreliable reasoning due to their limited capability of integrating visual and textual cues. In this work, we introduce Zoom-IQA, a VLM-based IQA model to explicitly emulate key cognitive behaviors: uncertainty awareness, region reasoning, and iterative refinement. Specifically, we present a two-stage training pipeline: 1) supervised fine-tuning (SFT) on our Grounded-Rationale-IQA (GR-IQA) dataset to teach the model to ground its assessments in key regions, and 2) reinforcement learning (RL) for dynamic policy exploration, stabilized by our KL-Coverage regularizer to prevent reasoning and scoring diversity collapse, with a Progressive Re-sampling Strategy for mitigating annotation bias. Extensive experiments show that Zoom-IQA achieves improved robustness, explainability, and generalization. The application to downstream tasks, such as image restoration, further demonstrates the effectiveness of Zoom-IQA.

---

### RAL2M: Retrieval Augmented Learning-To-Match Against Hallucination in Compliance-Guaranteed Service Systems

Authors: Mengze Hong, Di Jiang, Jiangtao Wen, Zhiyang Su, Yawen Li, Yanjie Sun, Guan Wang, Chen Jason Zhang
Published: 2026-01-06
Categories: cs.CL, cs.AI
arXiv: https://arxiv.org/abs/2601.02917v1

Hallucination is a major concern in LLM-driven service systems, necessitating explicit knowledge grounding for compliance-guaranteed responses. In this paper, we introduce Retrieval-Augmented Learning-to-Match (RAL2M), a novel framework that eliminates generation hallucination by repositioning LLMs as query-response matching judges within a retrieval-based system, providing a robust alternative to purely generative approaches. To further mitigate judgment hallucination, we propose a query-adaptive latent ensemble strategy that explicitly models heterogeneous model competence and interdependencies among LLMs, deriving a calibrated consensus decision. Extensive experiments on large-scale benchmarks demonstrate that the proposed method effectively leverages the "wisdom of the crowd" and significantly outperforms strong baselines. Finally, we discuss best practices and promising directions for further exploiting latent representations in future work.

---

### ChemBART: A Pre-trained BART Model Assisting Organic Chemistry Analysis

Authors: Kenan Li, Yijian Zhang, Jin Wang, Haipeng Gan, Zeying Sun, Xiaoguang Lei, Hao Dong
Published: 2026-01-06
Categories: cs.LG
arXiv: https://arxiv.org/abs/2601.02915v1

Recent advances in large language models (LLMs) have demonstrated transformative potential across diverse fields. While LLMs have been applied to molecular simplified molecular input line entry system (SMILES) in computer-aided synthesis planning (CASP), existing methodologies typically address single tasks, such as precursor prediction. We introduce ChemBART, a SMILES-based LLM pre-trained on chemical reactions, which enables a unified model for multiple downstream chemical tasks--achieving the paradigm of "one model, one pre-training, multiple tasks." By leveraging outputs from a mask-filling pre-training task on reaction expressions, ChemBART effectively solves a variety of chemical problems, including precursor/reagent generation, temperature-yield regression, molecular property classification, and optimizing the policy and value functions within a reinforcement learning framework, integrated with Monte Carlo tree search for multi-step synthesis route design. Unlike single-molecule pre-trained LLMs constrained to specific applications, ChemBART addresses broader chemical challenges and integrates them for comprehensive synthesis planning. Crucially, ChemBART-designed multi-step synthesis routes and reaction conditions directly inspired wet-lab validation, which confirmed shorter pathways with ~30% yield improvement over literature benchmarks. Our work validates the power of reaction-focused pre-training and showcases the broad utility of ChemBART in advancing the complete synthesis planning cycle.

---

### Image, Word and Thought: A More Challenging Language Task for the Iterated Learning Model

Authors: Hyoyeon Lee, Seth Bullock, Conor Houghton
Published: 2026-01-06
Categories: cs.CL, cs.LG, cs.MA
arXiv: https://arxiv.org/abs/2601.02911v1

The iterated learning model simulates the transmission of language from generation to generation in order to explore how the constraints imposed by language transmission facilitate the emergence of language structure. Despite each modelled language learner starting from a blank slate, the presence of a bottleneck limiting the number of utterances to which the learner is exposed can lead to the emergence of language that lacks ambiguity, is governed by grammatical rules, and is consistent over successive generations, that is, one that is expressive, compositional and stable. The recent introduction of a more computationally tractable and ecologically valid semi supervised iterated learning model, combining supervised and unsupervised learning within an autoencoder architecture, has enabled exploration of language transmission dynamics for much larger meaning-signal spaces. Here, for the first time, the model has been successfully applied to a language learning task involving the communication of much more complex meanings: seven-segment display images. Agents in this model are able to learn and transmit a language that is expressive: distinct codes are employed for all 128 glyphs; compositional: signal components consistently map to meaning components, and stable: the language does not change from generation to generation.

---

### TA-Prompting: Enhancing Video Large Language Models for Dense Video Captioning via Temporal Anchors

Authors: Wei-Yuan Cheng, Kai-Po Chang, Chi-Pin Huang, Fu-En Yang, Yu-Chiang Frank Wang
Published: 2026-01-06
Categories: cs.CV, cs.AI, cs.LG
arXiv: https://arxiv.org/abs/2601.02908v1

Dense video captioning aims to interpret and describe all temporally localized events throughout an input video. Recent state-of-the-art methods leverage large language models (LLMs) to provide detailed moment descriptions for video data. However, existing VideoLLMs remain challenging in identifying precise event boundaries in untrimmed videos, causing the generated captions to be not properly grounded. In this paper, we propose TA-Prompting, which enhances VideoLLMs via Temporal Anchors that learn to precisely localize events and prompt the VideoLLMs to perform temporal-aware video event understanding. During inference, in order to properly determine the output caption sequence from an arbitrary number of events presented within a video, we introduce an event coherent sampling strategy to select event captions with sufficient coherence across temporal events and cross-modal similarity with the given video. Through extensive experiments on benchmark datasets, we show that our TA-Prompting is favorable against state-of-the-art VideoLLMs, yielding superior performance on dense video captioning and temporal understanding tasks including moment retrieval and temporalQA.

---

### Beyond the Black Box: A Survey on the Theory and Mechanism of Large Language Models

Authors: Zeyu Gan, Ruifeng Ren, Wei Yao, Xiaolin Hu, Gengze Xu, Chen Qian, Huayi Tang, Zixuan Gong, Xinhao Yao, Pengwei Tang, Zhenxing Dou, Yong Liu
Published: 2026-01-06
Categories: cs.CL
arXiv: https://arxiv.org/abs/2601.02907v2

The rapid emergence of Large Language Models (LLMs) has precipitated a profound paradigm shift in Artificial Intelligence, delivering monumental engineering successes that increasingly impact modern society. However, a critical paradox persists within the current field: despite the empirical efficacy, our theoretical understanding of LLMs remains disproportionately nascent, forcing these systems to be treated largely as ``black boxes''. To address this theoretical fragmentation, this survey proposes a unified lifecycle-based taxonomy that organizes the research landscape into six distinct stages: Data Preparation, Model Preparation, Training, Alignment, Inference, and Evaluation. Within this framework, we provide a systematic review of the foundational theories and internal mechanisms driving LLM performance. Specifically, we analyze core theoretical issues such as the mathematical justification for data mixtures, the representational limits of various architectures, and the optimization dynamics of alignment algorithms. Moving beyond current best practices, we identify critical frontier challenges, including the theoretical limits of synthetic data self-improvement, the mathematical bounds of safety guarantees, and the mechanistic origins of emergent intelligence. By connecting empirical observations with rigorous scientific inquiry, this work provides a structured roadmap for transitioning LLM development from engineering heuristics toward a principled scientific discipline.

---

### Linear Script Representations in Speech Foundation Models Enable Zero-Shot Transliteration

Authors: Ryan Soh-Eun Shim, Kwanghee Choi, Kalvin Chang, Ming-Hao Hsu, Florian Eichin, Zhizheng Wu, Alane Suhr, Michael A. Hedderich, David Harwath, David R. Mortensen, Barbara Plank
Published: 2026-01-06
Categories: cs.CL
arXiv: https://arxiv.org/abs/2601.02906v1

Multilingual speech foundation models such as Whisper are trained on web-scale data, where data for each language consists of a myriad of regional varieties. However, different regional varieties often employ different scripts to write the same language, rendering speech recognition output also subject to non-determinism in the output script. To mitigate this problem, we show that script is linearly encoded in the activation space of multilingual speech models, and that modifying activations at inference time enables direct control over output script. We find the addition of such script vectors to activations at test time can induce script change even in unconventional language-script pairings (e.g. Italian in Cyrillic and Japanese in Latin script). We apply this approach to inducing post-hoc control over the script of speech recognition output, where we observe competitive performance across all model sizes of Whisper.

---

### LOST-3DSG: Lightweight Open-Vocabulary 3D Scene Graphs with Semantic Tracking in Dynamic Environments

Authors: Sara Micol Ferraina, Michele Brienza, Francesco Argenziano, Emanuele Musumeci, Vincenzo Suriani, Domenico D. Bloisi, Daniele Nardi
Published: 2026-01-06
Categories: cs.RO, cs.AI
arXiv: https://arxiv.org/abs/2601.02905v2

Tracking objects that move within dynamic environments is a core challenge in robotics. Recent research has advanced this topic significantly; however, many existing approaches remain inefficient due to their reliance on heavy foundation models. To address this limitation, we propose LOST-3DSG, a lightweight open-vocabulary 3D scene graph designed to track dynamic objects in real-world environments. Our method adopts a semantic approach to entity tracking based on word2vec and sentence embeddings, enabling an open-vocabulary representation while avoiding the necessity of storing dense CLIP visual features. As a result, LOST-3DSG achieves superior performance compared to approaches that rely on high-dimensional visual embeddings. We evaluate our method through qualitative and quantitative experiments conducted in a real 3D environment using a TIAGo robot. The results demonstrate the effectiveness and efficiency of LOST-3DSG in dynamic object tracking. Code and supplementary material are publicly available on the project website at https://lab-rococo-sapienza.github.io/lost-3dsg/.

---

### Logical Phase Transitions: Understanding Collapse in LLM Logical Reasoning

Authors: Xinglang Zhang, Yunyao Zhang, ZeLiang Chen, Junqing Yu, Wei Yang, Zikai Song
Published: 2026-01-06
Categories: cs.AI, cs.CL, cs.LO
arXiv: https://arxiv.org/abs/2601.02902v2

Symbolic logical reasoning is a critical yet underexplored capability of large language models (LLMs), providing reliable and verifiable decision-making in high-stakes domains such as mathematical reasoning and legal judgment. In this study, we present a systematic analysis of logical reasoning under controlled increases in logical complexity, and reveal a previously unrecognized phenomenon, which we term Logical Phase Transitions: rather than degrading smoothly, logical reasoning performance remains stable within a regime but collapses abruptly beyond a critical logical depth, mirroring physical phase transitions such as water freezing beyond a critical temperature threshold. Building on this insight, we propose Neuro-Symbolic Curriculum Tuning, a principled framework that adaptively aligns natural language with logical symbols to establish a shared representation, and reshapes training dynamics around phase-transition boundaries to progressively strengthen reasoning at increasing logical depths. Experiments on five benchmarks show that our approach effectively mitigates logical reasoning collapse at high complexity, yielding average accuracy gains of +1.26 in naive prompting and +3.95 in CoT, while improving generalization to unseen logical compositions. Code and data are available at https://github.com/AI4SS/Logical-Phase-Transitions.

---