# Research Papers — 2026-01 · bundle page 131 of 189 > Full-text markdown of papers #6501–6550 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/131 [← Previous bundle](https://aipod.dev/papers/archive/2026-01/page/130.md) [Next bundle →](https://aipod.dev/papers/archive/2026-01/page/132.md) ### A Visual Semantic Adaptive Watermark grounded by Prefix-Tuning for Large Vision-Language Model Authors: Qi Zheng, Shuliang Liu, Yu Huang, Sihang Jia, Jungang Li, Lyuhao Chen, Junhao Chen, Hanqian Li, Aiwei Liu, Yibo Yan, Xuming Hu Published: 2026-01-12 Categories: cs.CV, cs.AI arXiv: https://arxiv.org/abs/2601.07291v1 Watermarking has emerged as a pivotal solution for content traceability and intellectual property protection in Large Vision-Language Models (LVLMs). However, vision-agnostic watermarks introduce visually irrelevant tokens and disrupt visual grounding by enforcing indiscriminate pseudo-random biases, while some semantic-aware methods incur prohibitive inference latency due to rejection sampling. In this paper, we propose the VIsual Semantic Adaptive Watermark (VISA-Mark), a novel framework that embeds detectable signals while strictly preserving visual fidelity. Our approach employs a lightweight, efficiently trained prefix-tuner to extract dynamic Visual-Evidence Weights, which quantify the evidentiary support for candidate tokens based on the visual input. These weights guide an adaptive vocabulary partitioning and logits perturbation mechanism, concentrating watermark strength specifically on visually-supported tokens. By actively aligning the watermark with visual evidence, VISA-Mark effectively maintains visual fidelity. Empirical results confirm that VISA-Mark outperforms conventional methods with a 7.8% improvement in visual consistency (Chair-I) and superior semantic fidelity. The framework maintains highly competitive detection accuracy (96.88% AUC) and robust attack resilience (99.3%) without sacrificing inference efficiency, effectively establishing a new standard for reliability-preserving multimodal watermarking. --- ### VideoLoom: A Video Large Language Model for Joint Spatial-Temporal Understanding Authors: Jiapeng Shi, Junke Wang, Zuyao You, Bo He, Zuxuan Wu Published: 2026-01-12 Categories: cs.CV arXiv: https://arxiv.org/abs/2601.07290v1 This paper presents VideoLoom, a unified Video Large Language Model (Video LLM) for joint spatial-temporal understanding. To facilitate the development of fine-grained spatial and temporal localization capabilities, we curate LoomData-8.7k, a human-centric video dataset with temporally grounded and spatially localized captions. With this, VideoLoom achieves state-of-the-art or highly competitive performance across a variety of spatial and temporal benchmarks (e.g., 63.1 J&F on ReVOS for referring video object segmentation, and 48.3 R1@0.7 on Charades-STA for temporal grounding). In addition, we introduce LoomBench, a novel benchmark consisting of temporal, spatial, and compositional video-question pairs, enabling a comprehensive evaluation of Video LLMs from diverse aspects. Collectively, these contributions offer a universal and effective suite for joint spatial-temporal video understanding, setting a new standard in multimodal intelligence. --- ### Kernel Alignment-based Multi-view Unsupervised Feature Selection with Sample-level Adaptive Graph Learning Authors: Yalan Tan, Yanyong Huang, Zongxin Shen, Dongjie Wang, Fengmao Lv, Tianrui Li Published: 2026-01-12 Categories: cs.LG arXiv: https://arxiv.org/abs/2601.07288v2 Although multi-view unsupervised feature selection (MUFS) has demonstrated success in dimensionality reduction for unlabeled multi-view data, most existing methods reduce feature redundancy by focusing on linear correlations among features but often overlook complex nonlinear dependencies. This limits the effectiveness of feature selection. In addition, existing methods fuse similarity graphs from multiple views by employing sample-invariant weights to preserve local structure. However, this process fails to account for differences in local neighborhood clarity among samples within each view, thereby hindering accurate characterization of the intrinsic local structure of the data. In this paper, we propose a Kernel Alignment-based multi-view unsupervised FeatUre selection with Sample-level adaptive graph lEarning method (KAFUSE) to address these issues. Specifically, we first employ kernel alignment with an orthogonal constraint to reduce feature redundancy in both linear and nonlinear relationships. Then, a cross-view consistent similarity graph is learned by applying sample-level fusion to each slice of a tensor formed by stacking similarity graphs from different views, which automatically adjusts the view weights for each sample during fusion. These two steps are integrated into a unified model for feature selection, enabling mutual enhancement between them. Extensive experiments on real multi-view datasets demonstrate the superiority of KAFUSE over state-of-the-art methods. --- ### Focal Guidance: Unlocking Controllability from Semantic-Weak Layers in Video Diffusion Models Authors: Yuanyang Yin, Yufan Deng, Shenghai Yuan, Kaipeng Zhang, Xiao Yang, Feng Zhao Published: 2026-01-12 Categories: cs.CV arXiv: https://arxiv.org/abs/2601.07287v1 The task of Image-to-Video (I2V) generation aims to synthesize a video from a reference image and a text prompt. This requires diffusion models to reconcile high-frequency visual constraints and low-frequency textual guidance during the denoising process. However, while existing I2V models prioritize visual consistency, how to effectively couple this dual guidance to ensure strong adherence to the text prompt remains underexplored. In this work, we observe that in Diffusion Transformer (DiT)-based I2V models, certain intermediate layers exhibit weak semantic responses (termed Semantic-Weak Layers), as indicated by a measurable drop in text-visual similarity. We attribute this to a phenomenon called Condition Isolation, where attention to visual features becomes partially detached from text guidance and overly relies on learned visual priors. To address this, we propose Focal Guidance (FG), which enhances the controllability from Semantic-Weak Layers. FG comprises two mechanisms: (1) Fine-grained Semantic Guidance (FSG) leverages CLIP to identify key regions in the reference frame and uses them as anchors to guide Semantic-Weak Layers. (2) Attention Cache transfers attention maps from semantically responsive layers to Semantic-Weak Layers, injecting explicit semantic signals and alleviating their over-reliance on the model's learned visual priors, thereby enhancing adherence to textual instructions. To further validate our approach and address the lack of evaluation in this direction, we introduce a benchmark for assessing instruction following in I2V models. On this benchmark, Focal Guidance proves its effectiveness and generalizability, raising the total score on Wan2.1-I2V to 0.7250 (+3.97\%) and boosting the MMDiT-based HunyuanVideo-I2V to 0.5571 (+7.44\%). --- ### Minimum Wasserstein distance estimator under covariate shift: closed-form, super-efficiency and irregularity Authors: Junjun Lang, Qiong Zhang, Yukun Liu Published: 2026-01-12 Categories: stat.ME, stat.ML arXiv: https://arxiv.org/abs/2601.07282v1 Covariate shift arises when covariate distributions differ between source and target populations while the conditional distribution of the response remains invariant, and it underlies problems in missing data and causal inference. We propose a minimum Wasserstein distance estimation framework for inference under covariate shift that avoids explicit modeling of outcome regressions or importance weights. The resulting W-estimator admits a closed-form expression and is numerically equivalent to the classical 1-nearest neighbor estimator, yielding a new optimal transport interpretation of nearest neighbor methods. We establish root-$n$ asymptotic normality and show that the estimator is not asymptotically linear, leading to super-efficiency relative to the semiparametric efficient estimator under covariate shift in certain regimes, and uniformly in missing data problems. Numerical simulations, along with an analysis of a rainfall dataset, underscore the exceptional performance of our W-estimator. --- ### Covariance-Driven Regression Trees: Reducing Overfitting in CART Authors: Likun Zhang, Wei Ma Published: 2026-01-12 Categories: stat.ML, cs.LG, stat.ME arXiv: https://arxiv.org/abs/2601.07281v1 Decision trees are powerful machine learning algorithms, widely used in fields such as economics and medicine for their simplicity and interpretability. However, decision trees such as CART are prone to overfitting, especially when grown deep or the sample size is small. Conventional methods to reduce overfitting include pre-pruning and post-pruning, which constrain the growth of uninformative branches. In this paper, we propose a complementary approach by introducing a covariance-driven splitting criterion for regression trees (CovRT). This method is more robust to overfitting than the empirical risk minimization criterion used in CART, as it produces more balanced and stable splits and more effectively identifies covariates with true signals. We establish an oracle inequality of CovRT and prove that its predictive accuracy is comparable to that of CART in high-dimensional settings. We find that CovRT achieves superior prediction accuracy compared to CART in both simulations and real-world tasks. --- ### ReasonTabQA: A Comprehensive Benchmark for Table Question Answering from Real World Industrial Scenarios Authors: Changzai Pan, Jie Zhang, Kaiwen Wei, Chenshuo Pan, Yu Zhao, Jingwang Huang, Jian Yang, Zhenhe Wu, Haoyang Zeng, Xiaoyan Gu, Weichao Sun, Yanbo Zhai et al. Published: 2026-01-12 Categories: cs.CL arXiv: https://arxiv.org/abs/2601.07280v1 Recent advancements in Large Language Models (LLMs) have significantly catalyzed table-based question answering (TableQA). However, existing TableQA benchmarks often overlook the intricacies of industrial scenarios, which are characterized by multi-table structures, nested headers, and massive scales. These environments demand robust table reasoning through deep structured inference, presenting a significant challenge that remains inadequately addressed by current methodologies. To bridge this gap, we present ReasonTabQA, a large-scale bilingual benchmark encompassing 1,932 tables across 30 industry domains such as energy and automotive. ReasonTabQA provides high-quality annotations for both final answers and explicit reasoning chains, supporting both thinking and no-thinking paradigms. Furthermore, we introduce TabCodeRL, a reinforcement learning method that leverages table-aware verifiable rewards to guide the generation of logical reasoning paths. Extensive experiments on ReasonTabQA and 4 TableQA datasets demonstrate that while TabCodeRL yields substantial performance gains on open-source LLMs, the persistent performance gap on ReasonTabQA underscores the inherent complexity of real-world industrial TableQA. --- ### A High-Recall Cost-Sensitive Machine Learning Framework for Real-Time Online Banking Transaction Fraud Detection Authors: Karthikeyan V. R., Premnath S., Kavinraaj S., J. Sangeetha Published: 2026-01-12 Categories: cs.CR, cs.LG arXiv: https://arxiv.org/abs/2601.07276v2 Fraudulent activities on digital banking services are becoming more intricate by the day, challenging existing defenses. While older rule driven methods struggle to keep pace, even precision focused algorithms fall short when new scams are introduced. These tools typically overlook subtle shifts in criminal behavior, missing crucial signals. Because silent breaches cost institutions far more than flagged but legitimate actions, catching every possible case is crucial. High sensitivity to actual threats becomes essential when oversight leads to heavy losses. One key aim here involves reducing missed fraud cases without spiking incorrect alerts too much. This study builds a system using group learning methods adjusted through smart threshold choices. Using real world transaction records shared openly, where cheating acts rarely appear among normal activities, tests are run under practical skewed distributions. The outcomes reveal that approximately 98 percent of actual fraud is detected, outperforming standard setups that rely on unchanging rules when dealing with uneven examples across classes. When tested in live settings, the fraud detection system connects directly to an online banking transaction flow, stopping questionable activities before they are completed. Alongside this setup, a browser add on built for Chrome is designed to flag deceptive web links and reduce threats from harmful sites. These results show that adjusting decisions by cost impact and validating across entire systems makes deployment more stable and realistic for today's digital banking platforms. --- ### Towards Comprehensive Semantic Speech Embeddings for Chinese Dialects Authors: Kalvin Chang, Yiwen Shao, Jiahong Li, Dong Yu Published: 2026-01-12 Categories: cs.CL arXiv: https://arxiv.org/abs/2601.07274v1 Despite having hundreds of millions of speakers, Chinese dialects lag behind Mandarin in speech and language technologies. Most varieties are primarily spoken, making dialect-to-Mandarin speech-LLMs (large language models) more practical than dialect LLMs. Building dialect-to-Mandarin speech-LLMs requires speech representations with cross-dialect semantic alignment between Chinese dialects and Mandarin. In this paper, we achieve such a cross-dialect semantic alignment by training a speech encoder with ASR (automatic speech recognition)-only data, as demonstrated by speech-to-speech retrieval on a new benchmark of spoken Chinese varieties that we contribute. Our speech encoder further demonstrates state-of-the-art ASR performance on Chinese dialects. Together, our Chinese dialect benchmark, semantically aligned speech representations, and speech-to-speech retrieval evaluation lay the groundwork for future Chinese dialect speech-LLMs. We release the benchmark at https://github.com/kalvinchang/yubao. --- ### GenDet: Painting Colored Bounding Boxes on Images via Diffusion Model for Object Detection Authors: Chen Min, Chengyang Li, Fanjie Kong, Qi Zhu, Dawei Zhao, Liang Xiao Published: 2026-01-12 Categories: cs.CV arXiv: https://arxiv.org/abs/2601.07273v1 This paper presents GenDet, a novel framework that redefines object detection as an image generation task. In contrast to traditional approaches, GenDet adopts a pioneering approach by leveraging generative modeling: it conditions on the input image and directly generates bounding boxes with semantic annotations in the original image space. GenDet establishes a conditional generation architecture built upon the large-scale pre-trained Stable Diffusion model, formulating the detection task as semantic constraints within the latent space. It enables precise control over bounding box positions and category attributes, while preserving the flexibility of the generative model. This novel methodology effectively bridges the gap between generative models and discriminative tasks, providing a fresh perspective for constructing unified visual understanding systems. Systematic experiments demonstrate that GenDet achieves competitive accuracy compared to discriminative detectors, while retaining the flexibility characteristic of generative methods. --- ### PALUM: Part-based Attention Learning for Unified Motion Retargeting Authors: Siqi Liu, Maoyu Wang, Bo Dai, Cewu Lu Published: 2026-01-12 Categories: cs.CV arXiv: https://arxiv.org/abs/2601.07272v1 Retargeting motion between characters with different skeleton structures is a fundamental challenge in computer animation. When source and target characters have vastly different bone arrangements, maintaining the original motion's semantics and quality becomes increasingly difficult. We present PALUM, a novel approach that learns common motion representations across diverse skeleton topologies by partitioning joints into semantic body parts and applying attention mechanisms to capture spatio-temporal relationships. Our method transfers motion to target skeletons by leveraging these skeleton-agnostic representations alongside target-specific structural information. To ensure robust learning and preserve motion fidelity, we introduce a cycle consistency mechanism that maintains semantic coherence throughout the retargeting process. Extensive experiments demonstrate superior performance in handling diverse skeletal structures while maintaining motion realism and semantic fidelity, even when generalizing to previously unseen skeleton-motion combinations. We will make our implementation publicly available to support future research. --- ### Document-Level Zero-Shot Relation Extraction with Entity Side Information Authors: Mohan Raj Chanthran, Soon Lay Ki, Ong Huey Fang, Bhawani Selvaretnam Published: 2026-01-12 Categories: cs.CL arXiv: https://arxiv.org/abs/2601.07271v1 Document-Level Zero-Shot Relation Extraction (DocZSRE) aims to predict unseen relation labels in text documents without prior training on specific relations. Existing approaches rely on Large Language Models (LLMs) to generate synthetic data for unseen labels, which poses challenges for low-resource languages like Malaysian English. These challenges include the incorporation of local linguistic nuances and the risk of factual inaccuracies in LLM-generated data. This paper introduces Document-Level Zero-Shot Relation Extraction with Entity Side Information (DocZSRE-SI) to address limitations in the existing DocZSRE approach. The DocZSRE-SI framework leverages Entity Side Information, such as Entity Mention Descriptions and Entity Mention Hypernyms, to perform ZSRE without depending on LLM-generated synthetic data. The proposed low-complexity model achieves an average improvement of 11.6% in the macro F1-Score compared to baseline models and existing benchmarks. By utilizing Entity Side Information, DocZSRE-SI offers a robust and efficient alternative to error-prone, LLM-based methods, demonstrating significant advancements in handling low-resource languages and linguistic diversity in relation extraction tasks. This research provides a scalable and reliable solution for ZSRE, particularly in contexts like Malaysian English news articles, where traditional LLM-based approaches fall short. --- ### From Landslide Conditioning Factors to Satellite Embeddings: Evaluating the Utilisation of Google AlphaEarth for Landslide Susceptibility Mapping using Deep Learning Authors: Yusen Cheng, Qinfeng Zhu, Lei Fan Published: 2026-01-12 Categories: cs.CV arXiv: https://arxiv.org/abs/2601.07268v2 Data-driven landslide susceptibility mapping (LSM) typically relies on landslide conditioning factors (LCFs), whose availability, heterogeneity, and preprocessing-related uncertainties can constrain mapping reliability. Recently, Google AlphaEarth (AE) embeddings, derived from multi-source geospatial observations, have emerged as a unified representation of Earth surface conditions. This study evaluated the potential of AE embeddings as alternative predictors for LSM. Two AE representations, including retained principal components and the full set of 64 embedding bands, were systematically compared with conventional LCFs across three study areas (Nantou County, Taiwan; Hong Kong; and part of Emilia-Romagna, Italy) using three deep learning models (CNN1D, CNN2D, and Vision Transformer). Performance was assessed using multiple evaluation metrics, ROC-AUC analysis, error statistics, and spatial pattern assessment. Results showed that AE-based models consistently outperformed LCFs across all regions and models, yielding higher F1-scores, AUC values, and more stable error distributions. Such improvement was most pronounced when using the full 64-band AE representation, with F1-score improvements of approximately 4% to 15% and AUC increased ranging from 0.04 to 0.11, depending on the study area and model. AE-based susceptibility maps also exhibited clearer spatial correspondence with observed landslide occurrences and enhanced sensitivity to localised landslide-prone conditions. Performance improvements were more evident in Nantou and Emilia than in Hong Kong. SHAP analysis further revealed regional variability in the contribution of individual AE bands, while several bands showed consistently high importance across all three study areas. These findings highlight the strong potential of AE embeddings as a standardised and information-rich alternative to conventional LCFs for LSM. --- ### The Confidence Dichotomy: Analyzing and Mitigating Miscalibration in Tool-Use Agents Authors: Weihao Xuan, Qingcheng Zeng, Heli Qi, Yunze Xiao, Junjue Wang, Naoto Yokoya Published: 2026-01-12 Categories: cs.CL arXiv: https://arxiv.org/abs/2601.07264v1 Autonomous agents based on large language models (LLMs) are rapidly evolving to handle multi-turn tasks, but ensuring their trustworthiness remains a critical challenge. A fundamental pillar of this trustworthiness is calibration, which refers to an agent's ability to express confidence that reliably reflects its actual performance. While calibration is well-established for static models, its dynamics in tool-integrated agentic workflows remain underexplored. In this work, we systematically investigate verbalized calibration in tool-use agents, revealing a fundamental confidence dichotomy driven by tool type. Specifically, our pilot study identifies that evidence tools (e.g., web search) systematically induce severe overconfidence due to inherent noise in retrieved information, while verification tools (e.g., code interpreters) can ground reasoning through deterministic feedback and mitigate miscalibration. To robustly improve calibration across tool types, we propose a reinforcement learning (RL) fine-tuning framework that jointly optimizes task accuracy and calibration, supported by a holistic benchmark of reward designs. We demonstrate that our trained agents not only achieve superior calibration but also exhibit robust generalization from local training environments to noisy web settings and to distinct domains such as mathematical reasoning. Our results highlight the necessity of domain-specific calibration strategies for tool-use agents. More broadly, this work establishes a foundation for building self-aware agents that can reliably communicate uncertainty in high-stakes, real-world deployments. --- ### When Bots Take the Bait: Exposing and Mitigating the Emerging Social Engineering Attack in Web Automation Agent Authors: Xinyi Wu, Geng Hong, Yueyue Chen, MingXuan Liu, Feier Jin, Xudong Pan, Jiarun Dai, Baojun Liu Published: 2026-01-12 Categories: cs.CR, cs.AI arXiv: https://arxiv.org/abs/2601.07263v1 Web agents, powered by large language models (LLMs), are increasingly deployed to automate complex web interactions. The rise of open-source frameworks (e.g., Browser Use, Skyvern-AI) has accelerated adoption, but also broadened the attack surface. While prior research has focused on model threats such as prompt injection and backdoors, the risks of social engineering remain largely unexplored. We present the first systematic study of social engineering attacks against web automation agents and design a pluggable runtime mitigation solution. On the attack side, we introduce the AgentBait paradigm, which exploits intrinsic weaknesses in agent execution: inducement contexts can distort the agent's reasoning and steer it toward malicious objectives misaligned with the intended task. On the defense side, we propose SUPERVISOR, a lightweight runtime module that enforces environment and intention consistency alignment between webpage context and intended goals to mitigate unsafe operations before execution. Empirical results show that mainstream frameworks are highly vulnerable to AgentBait, with an average attack success rate of 67.5% and peaks above 80% under specific strategies (e.g., trusted identity forgery). Compared with existing lightweight defenses, our module can be seamlessly integrated across different web automation frameworks and reduces attack success rates by up to 78.1% on average while incurring only a 7.7% runtime overhead and preserving usability. This work reveals AgentBait as a critical new threat surface for web agents and establishes a practical, generalizable defense, advancing the security of this rapidly emerging ecosystem. We reported the details of this attack to the framework developers and received acknowledgment before submission. --- ### Pseudodata-guided Invariant Representation Learning Boosts the Out-of-Distribution Generalization in Enzymatic Kinetic Parameter Prediction Authors: Haomin Wu, Zhiwei Nie, Hongyu Zhang, Zhixiang Ren Published: 2026-01-12 Categories: cs.LG, cs.AI, q-bio.QM arXiv: https://arxiv.org/abs/2601.07261v1 Accurate prediction of enzyme kinetic parameters is essential for understanding catalytic mechanisms and guiding enzyme engineering.However, existing deep learning-based enzyme-substrate interaction (ESI) predictors often exhibit performance degradation on sequence-divergent, out-of-distribution (OOD) cases, limiting robustness under biologically relevant perturbations.We propose O$^2$DENet, a lightweight, plug-and-play module that enhances OOD generalization via biologically and chemically informed perturbation augmentation and invariant representation learning.O$^2$DENet introduces enzyme-substrate perturbations and enforces consistency between original and augmented enzyme-substrate-pair representations to encourage invariance to distributional shifts.When integrated with representative ESI models, O$^2$DENet consistently improves predictive performance for both $k_{cat}$ and $K_m$ across stringent sequence-identity-based OOD benchmarks, achieving state-of-the-art results among the evaluated methods in terms of accuracy and robustness metrics.Overall, O$^2$DENet provides a general and effective strategy to enhance the stability and deployability of data-driven enzyme kinetics predictors for real-world enzyme engineering applications. --- ### ActiShade: Activating Overshadowed Knowledge to Guide Multi-Hop Reasoning in Large Language Models Authors: Huipeng Ma, Luan Zhang, Dandan Song, Linmei Hu, Yuhang Tian, Jun Yang, Changzhi Zhou, Chenhao Li, Yizhou Jin, Xudong Li, Meng Lin, Mingxing Zhang et al. Published: 2026-01-12 Categories: cs.CL arXiv: https://arxiv.org/abs/2601.07260v1 In multi-hop reasoning, multi-round retrieval-augmented generation (RAG) methods typically rely on LLM-generated content as the retrieval query. However, these approaches are inherently vulnerable to knowledge overshadowing - a phenomenon where critical information is overshadowed during generation. As a result, the LLM-generated content may be incomplete or inaccurate, leading to irrelevant retrieval and causing error accumulation during the iteration process. To address this challenge, we propose ActiShade, which detects and activates overshadowed knowledge to guide large language models (LLMs) in multi-hop reasoning. Specifically, ActiShade iteratively detects the overshadowed keyphrase in the given query, retrieves documents relevant to both the query and the overshadowed keyphrase, and generates a new query based on the retrieved documents to guide the next-round iteration. By supplementing the overshadowed knowledge during the formulation of next-round queries while minimizing the introduction of irrelevant noise, ActiShade reduces the error accumulation caused by knowledge overshadowing. Extensive experiments show that ActiShade outperforms existing methods across multiple datasets and LLMs. --- ### Simulated Annealing-based Candidate Optimization for Batch Acquisition Functions Authors: Sk Md Ahnaf Akif Alvi, Raymundo Arróyave, Douglas Allaire Published: 2026-01-12 Categories: cs.LG, cond-mat.mtrl-sci, stat.ML arXiv: https://arxiv.org/abs/2601.07258v1 Bayesian Optimization with multi-objective acquisition functions such as q-Expected Hypervolume Improvement (qEHVI) requires efficient candidate optimization to maximize acquisition function values. Traditional approaches rely on continuous optimization methods like Sequential Least Squares Programming (SLSQP) for candidate selection. However, these gradient-based methods can become trapped in local optima, particularly in complex or high-dimensional objective landscapes. This paper presents a simulated annealing-based approach for candidate optimization in batch acquisition functions as an alternative to conventional continuous optimization methods. We evaluate our simulated annealing approach against SLSQP across four benchmark multi-objective optimization problems: ZDT1 (30D, 2 objectives), DTLZ2 (7D, 3 objectives), Kursawe (3D, 2 objectives), and Latent-Aware (4D, 2 objectives). Our results demonstrate that simulated annealing consistently achieves superior hypervolume performance compared to SLSQP in most test functions. The improvement is particularly pronounced for DTLZ2 and Latent-Aware problems, where simulated annealing reaches significantly higher hypervolume values and maintains better convergence characteristics. The histogram analysis of objective space coverage further reveals that simulated annealing explores more diverse and optimal regions of the Pareto front. These findings suggest that metaheuristic optimization approaches like simulated annealing can provide more robust and effective candidate optimization for multi-objective Bayesian optimization, offering a promising alternative to traditional gradient-based methods for batch acquisition function optimization. --- ### Innovation Capacity of Dynamical Learning Systems Authors: Anthony M. Polloreno Published: 2026-01-12 Categories: cs.LG, cs.IT arXiv: https://arxiv.org/abs/2601.07257v1 In noisy physical reservoirs, the classical information-processing capacity $C_{\mathrm{ip}}$ quantifies how well a linear readout can realize tasks measurable from the input history, yet $C_{\mathrm{ip}}$ can be far smaller than the observed rank of the readout covariance. We explain this ``missing capacity'' by introducing the innovation capacity $C_{\mathrm{i}}$, the total capacity allocated to readout components orthogonal to the input filtration (Doob innovations, including input-noise mixing). Using a basis-free Hilbert-space formulation of the predictable/innovation decomposition, we prove the conservation law $C_{\mathrm{ip}}+C_{\mathrm{i}}=\mathrm{rank}(Σ_{XX})\le d$, so predictable and innovation capacities exactly partition the rank of the observable readout dimension covariance $Σ_{XX}\in \mathbb{R}^{\rm d\times d}$. In linear-Gaussian Johnson-Nyquist regimes, $Σ_{XX}(T)=S+T N_0$, the split becomes a generalized-eigenvalue shrinkage rule and gives an explicit monotone tradeoff between temperature and predictable capacity. Geometrically, in whitened coordinates the predictable and innovation components correspond to complementary covariance ellipsoids, making $C_{\mathrm{i}}$ a trace-controlled innovation budget. A large $C_{\mathrm{i}}$ forces a high-dimensional innovation subspace with a variance floor and under mild mixing and anti-concentration assumptions this yields extensive innovation-block differential entropy and exponentially many distinguishable histories. Finally, we give an information-theoretic lower bound showing that learning the induced innovation-block law in total variation requires a number of samples that scales with the effective innovation dimension, supporting the generative utility of noisy physical reservoirs. --- ### Universal Adversarial Purification with DDIM Metric Loss for Stable Diffusion Authors: Li Zheng, Liangbin Xie, Jiantao Zhou, He YiMin Published: 2026-01-12 Categories: cs.CV, cs.CR arXiv: https://arxiv.org/abs/2601.07253v1 Stable Diffusion (SD) often produces degraded outputs when the training dataset contains adversarial noise. Adversarial purification offers a promising solution by removing adversarial noise from contaminated data. However, existing purification methods are primarily designed for classification tasks and fail to address SD-specific adversarial strategies, such as attacks targeting the VAE encoder, UNet denoiser, or both. To address the gap in SD security, we propose Universal Diffusion Adversarial Purification (UDAP), a novel framework tailored for defending adversarial attacks targeting SD models. UDAP leverages the distinct reconstruction behaviors of clean and adversarial images during Denoising Diffusion Implicit Models (DDIM) inversion to optimize the purification process. By minimizing the DDIM metric loss, UDAP can effectively remove adversarial noise. Additionally, we introduce a dynamic epoch adjustment strategy that adapts optimization iterations based on reconstruction errors, significantly improving efficiency without sacrificing purification quality. Experiments demonstrate UDAP's robustness against diverse adversarial methods, including PID (VAE-targeted), Anti-DreamBooth (UNet-targeted), MIST (hybrid), and robustness-enhanced variants like Anti-Diffusion (Anti-DF) and MetaCloak. UDAP also generalizes well across SD versions and text prompts, showcasing its practical applicability in real-world scenarios. --- ### DDT: A Dual-Masking Dual-Expert Transformer for Energy Time-Series Forecasting Authors: Mingnan Zhu, Qixuan Zhang, Yixuan Cheng, Fangzhou Gu, Shiming Lin Published: 2026-01-12 Categories: cs.LG, cs.AI arXiv: https://arxiv.org/abs/2601.07250v1 Accurate energy time-series forecasting is crucial for ensuring grid stability and promoting the integration of renewable energy, yet it faces significant challenges from complex temporal dependencies and the heterogeneity of multi-source data. To address these issues, we propose DDT, a novel and robust deep learning framework for high-precision time-series forecasting. At its core, DDT introduces two key innovations. First, we design a dual-masking mechanism that synergistically combines a strict causal mask with a data-driven dynamic mask. This novel design ensures theoretical causal consistency while adaptively focusing on the most salient historical information, overcoming the rigidity of traditional masking techniques. Second, our architecture features a dual-expert system that decouples the modeling of temporal dynamics and cross-variable correlations into parallel, specialized pathways, which are then intelligently integrated through a dynamic gated fusion module. We conducted extensive experiments on 7 challenging energy benchmark datasets, including ETTh, Electricity, and Solar. The results demonstrate that DDT consistently outperforms strong state-of-the-art baselines across all prediction horizons, establishing a new benchmark for the task. --- ### Multi-environment Invariance Learning with Missing Data Authors: Yiran Jia, Jelena Bradic Published: 2026-01-12 Categories: stat.ML, cs.LG, math.ST, stat.ME arXiv: https://arxiv.org/abs/2601.07247v2 Learning models that can handle distribution shifts is a key challenge in domain generalization. Invariance learning, an approach that focuses on identifying features invariant across environments, improves model generalization by capturing stable relationships, which may represent causal effects when the data distribution is encoded within a structural equation model (SEM) and satisfies modularity conditions. This has led to a growing body of work that builds on invariance learning, leveraging the inherent heterogeneity across environments to develop methods that provide causal explanations while enhancing robust prediction. However, in many practical scenarios, obtaining complete outcome data from each environment is challenging due to the high cost or complexity of data collection. This limitation in available data hinders the development of models that fully leverage environmental heterogeneity, making it crucial to address missing outcomes to improve both causal insights and robust prediction. In this work, we derive an estimator from the invariance objective under missing outcomes. We establish non-asymptotic guarantees on variable selection property and $\ell_2$ error convergence rates, which are influenced by the proportion of missing data and the quality of imputation models across environments. We evaluate the performance of the new estimator through extensive simulations and demonstrate its application using the UCI Bike Sharing dataset to predict the count of bike rentals. The results show that despite relying on a biased imputation model, the estimator is efficient and achieves lower prediction error, provided the bias is within a reasonable range. --- ### Learning to Trust the Crowd: A Multi-Model Consensus Reasoning Engine for Large Language Models Authors: Pranav Kallem Published: 2026-01-12 Categories: cs.AI, cs.CL, cs.LG arXiv: https://arxiv.org/abs/2601.07245v1 Large language models (LLMs) achieve strong average performance yet remain unreliable at the instance level, with frequent hallucinations, brittle failures, and poorly calibrated confidence. We study reliability through the lens of multi-model consensus: given responses from several heterogeneous LLMs, can we learn which answer is most likely correct for a given query? We introduce a Multi-Model Consensus Reasoning Engine that treats the set of LLM outputs as input to a supervised meta-learner. The system maps natural language responses into structured features using semantic embeddings, pairwise similarity and clustering statistics, lexical and structural cues, reasoning-quality scores, confidence estimates, and model-specific priors, and then applies gradient-boosted trees, listwise ranking, and graph neural networks over similarity graphs of answers. Using three open-weight LLMs evaluated on compact, resource-constrained subsets of GSM8K, ARC-Challenge, HellaSwag, and TruthfulQA, our best graph-attention-based consensus model improves macro-average accuracy by 4.6 percentage points over the strongest single LLM and by 8.1 points over majority vote, while also yielding lower Brier scores and fewer TruthfulQA hallucinations. Ablation and feature-importance analyses show that semantic agreement and clustering features are most influential, with reasoning-quality and model-prior features providing complementary gains, suggesting supervised multi-model consensus is a practical route toward more reliable LLM behavior, even in a modest single-machine setup. --- ### HERE: Hierarchical Active Exploration of Radiance Field with Epistemic Uncertainty Minimization Authors: Taekbeom Lee, Dabin Kim, Youngseok Jang, H. Jin Kim Published: 2026-01-12 Categories: cs.RO, cs.CV arXiv: https://arxiv.org/abs/2601.07242v2 We present HERE, an active 3D scene reconstruction framework based on neural radiance fields, enabling high-fidelity implicit mapping. Our approach centers around an active learning strategy for camera trajectory generation, driven by accurate identification of unseen regions, which supports efficient data acquisition and precise scene reconstruction. The key to our approach is epistemic uncertainty quantification based on evidential deep learning, which directly captures data insufficiency and exhibits a strong correlation with reconstruction errors. This allows our framework to more reliably identify unexplored or poorly reconstructed regions compared to existing methods, leading to more informed and targeted exploration. Additionally, we design a hierarchical exploration strategy that leverages learned epistemic uncertainty, where local planning extracts target viewpoints from high-uncertainty voxels based on visibility for trajectory generation, and global planning uses uncertainty to guide large-scale coverage for efficient and comprehensive reconstruction. The effectiveness of the proposed method in active 3D reconstruction is demonstrated by achieving higher reconstruction completeness compared to previous approaches on photorealistic simulated scenes across varying scales, while a hardware demonstration further validates its real-world applicability. Project page: https://taekbum.github.io/here/ --- ### Stochastic CHAOS: Why Deterministic Inference Kills, and Distributional Variability Is the Heartbeat of Artifical Cognition Authors: Tanmay Joshi, Shourya Aggarwal, Anusa Saha, Aadi Pandey, Shreyash Dhoot, Vighnesh Rai, Raxit Goswami, Aman Chadha, Vinija Jain, Amitava Das Published: 2026-01-12 Categories: cs.AI arXiv: https://arxiv.org/abs/2601.07239v1 Deterministic inference is a comforting ideal in classical software: the same program on the same input should always produce the same output. As large language models move into real-world deployment, this ideal has been imported wholesale into inference stacks. Recent work from the Thinking Machines Lab has presented a detailed analysis of nondeterminism in LLM inference, showing how batch-invariant kernels and deterministic attention can enforce bitwise-identical outputs, positioning deterministic inference as a prerequisite for reproducibility and enterprise reliability. In this paper, we take the opposite stance. We argue that, for LLMs, deterministic inference kills. It kills the ability to model uncertainty, suppresses emergent abilities, collapses reasoning into a single brittle path, and weakens safety alignment by hiding tail risks. LLMs implement conditional distributions over outputs, not fixed functions. Collapsing these distributions to a single canonical completion may appear reassuring, but it systematically conceals properties central to artificial cognition. We instead advocate Stochastic CHAOS, treating distributional variability as a signal to be measured and controlled. Empirically, we show that deterministic inference is systematically misleading. Single-sample deterministic evaluation underestimates both capability and fragility, masking failure probability under paraphrases and noise. Phase-like transitions associated with emergent abilities disappear under greedy decoding. Multi-path reasoning degrades when forced onto deterministic backbones, reducing accuracy and diagnostic insight. Finally, deterministic evaluation underestimates safety risk by hiding rare but dangerous behaviors that appear only under multi-sample evaluation. --- ### Group Pattern Selection Optimization: Let LRMs Pick the Right Pattern for Reasoning Authors: Hanbin Wang, Jingwei Song, Jinpeng Li, Fei Mi, Lifeng Shang Published: 2026-01-12 Categories: cs.AI arXiv: https://arxiv.org/abs/2601.07238v1 Large reasoning models (LRMs) exhibit diverse high-level reasoning patterns (e.g., direct solution, reflection-and-verification, and exploring multiple solutions), yet prevailing training recipes implicitly bias models toward a limited set of dominant patterns. Through a systematic analysis, we identify substantial accuracy variance across these patterns on mathematics and science benchmarks, revealing that a model's default reasoning pattern is often sub-optimal for a given problem. To address this, we introduce Group Pattern Selection Optimization (GPSO), a reinforcement learning framework that extends GRPO by incorporating multi-pattern rollouts, verifier-guided optimal pattern selection per problem, and attention masking during optimization to prevent the leakage of explicit pattern suffixes into the learned policy. By exploring a portfolio of diverse reasoning strategies and optimizing the policy on the most effective ones, GPSO enables the model to internalize the mapping from problem characteristics to optimal reasoning patterns. Extensive experiments demonstrate that GPSO delivers consistent and substantial performance gains across various model backbones and benchmarks, effectively mitigating pattern sub-optimality and fostering more robust, adaptable reasoning. All data and codes are available at https://github.com/wanghanbinpanda/GPSO. --- ### From "Thinking" to "Justifying": Aligning High-Stakes Explainability with Professional Communication Standards Authors: Chen Qian, Yimeng Wang, Yu Chen, Lingfei Wu, Andreas Stathopoulos Published: 2026-01-12 Categories: cs.AI arXiv: https://arxiv.org/abs/2601.07233v1 Explainable AI (XAI) in high-stakes domains should help stakeholders trust and verify system outputs. Yet Chain-of-Thought methods reason before concluding, and logical gaps or hallucinations can yield conclusions that do not reliably align with their rationale. Thus, we propose "Result -> Justify", which constrains the output communication to present a conclusion before its structured justification. We introduce SEF (Structured Explainability Framework), operationalizing professional conventions (e.g., CREAC, BLUF) via six metrics for structure and grounding. Experiments across four tasks in three domains validate this approach: all six metrics correlate with correctness (r=0.20-0.42; p<0.001), and SEF achieves 83.9% accuracy (+5.3 over CoT). These results suggest structured justification can improve verifiability and may also improve reliability. --- ### Yes FLoReNce, I Will Do Better Next Time! Agentic Feedback Reasoning for Humorous Meme Detection Authors: Olivia Shanhong Liu, Pai Chet Ng, De Wen Soh, Konstantinos N. Plataniotis Published: 2026-01-12 Categories: cs.AI arXiv: https://arxiv.org/abs/2601.07232v1 Humorous memes blend visual and textual cues to convey irony, satire, or social commentary, posing unique challenges for AI systems that must interpret intent rather than surface correlations. Existing multimodal or prompting-based models generate explanations for humor but operate in an open loop,lacking the ability to critique or refine their reasoning once a prediction is made. We propose FLoReNce, an agentic feedback reasoning framework that treats meme understanding as a closed-loop process during learning and an open-loop process during inference. In the closed loop, a reasoning agent is critiqued by a judge; the error and semantic feedback are converted into control signals and stored in a feedback-informed, non-parametric knowledge base. At inference, the model retrieves similar judged experiences from this KB and uses them to modulate its prompt, enabling better, self-aligned reasoning without finetuning. On the PrideMM dataset, FLoReNce improves both predictive performance and explanation quality over static multimodal baselines, showing that feedback-regulated prompting is a viable path to adaptive meme humor understanding. --- ### DiSCo: Making Absence Visible in Intelligent Summarization Interfaces Authors: Eran Fainman, Hagit Ben Shoshan, Adir Solomon, Osnat Mokryn Published: 2026-01-12 Categories: cs.HC, cs.AI, cs.IR arXiv: https://arxiv.org/abs/2601.07229v2 Intelligent interfaces increasingly use large language models to summarize user-generated content, yet these summaries emphasize what is mentioned while overlooking what is missing. This presence bias can mislead users who rely on summaries to make decisions. We present Domain Informed Summarization through Contrast (DiSCo), an expectation-based computational approach that makes absences visible by comparing each entity's content with domain topical expectations captured in reference distributions of aspects typically discussed in comparable accommodations. This comparison identifies aspects that are either unusually emphasized or missing relative to domain norms and integrates them into the generated text. In a user study across three accommodation domains, namely ski, beach, and city center, DiSCo summaries were rated as more detailed and useful for decision making than baseline large language model summaries, although slightly harder to read. The findings show that modeling expectations reduces presence bias and improves both transparency and decision support in intelligent summarization interfaces. --- ### Lost in the Noise: How Reasoning Models Fail with Contextual Distractors Authors: Seongyun Lee, Yongrae Jo, Minju Seo, Moontae Lee, Minjoon Seo Published: 2026-01-12 Categories: cs.AI, cs.CL arXiv: https://arxiv.org/abs/2601.07226v1 Recent advances in reasoning models and agentic AI systems have led to an increased reliance on diverse external information. However, this shift introduces input contexts that are inherently noisy, a reality that current sanitized benchmarks fail to capture. We introduce NoisyBench, a comprehensive benchmark that systematically evaluates model robustness across 11 datasets in RAG, reasoning, alignment, and tool-use tasks against diverse noise types, including random documents, irrelevant chat histories, and hard negative distractors. Our evaluation reveals a catastrophic performance drop of up to 80% in state-of-the-art models when faced with contextual distractors. Crucially, we find that agentic workflows often amplify these errors by over-trusting noisy tool outputs, and distractors can trigger emergent misalignment even without adversarial intent. We find that prompting, context engineering, SFT, and outcome-reward only RL fail to ensure robustness; in contrast, our proposed Rationale-Aware Reward (RARE) significantly strengthens resilience by incentivizing the identification of helpful information within noise. Finally, we uncover an inverse scaling trend where increased test-time computation leads to worse performance in noisy settings and demonstrate via attention visualization that models disproportionately focus on distractor tokens, providing vital insights for building the next generation of robust, reasoning-capable agents. --- ### Consolidation or Adaptation? PRISM: Disentangling SFT and RL Data via Gradient Concentration Authors: Yang Zhao, Yangou Ouyang, Xiao Ding, Hepeng Wang, Bibo Cai, Kai Xiong, Jinglong Gao, Zhouhao Sun, Li Du, Bing Qin, Ting Liu Published: 2026-01-12 Categories: cs.AI, cs.LG arXiv: https://arxiv.org/abs/2601.07224v2 While Hybrid Supervised Fine-Tuning (SFT) followed by Reinforcement Learning (RL) has become the standard paradigm for training LLM agents, effective mechanisms for data allocation between these stages remain largely underexplored. Current data arbitration strategies often rely on surface-level heuristics that fail to diagnose intrinsic learning needs. Since SFT targets pattern consolidation through imitation while RL drives structural adaptation via exploration, misaligning data with these functional roles causes severe optimization interference. We propose PRISM, a dynamics-aware framework grounded in Schema Theory that arbitrates data based on its degree of cognitive conflict with the model's existing knowledge. By analyzing the spatial geometric structure of gradients, PRISM identifies data triggering high spatial concentration as high-conflict signals that require RL for structural restructuring. In contrast, data yielding diffuse updates is routed to SFT for efficient consolidation. Extensive experiments on WebShop and ALFWorld demonstrate that PRISM achieves a Pareto improvement, outperforming state-of-the-art hybrid methods while reducing computational costs by up to 3.22$\times$. Our findings suggest that disentangling data based on internal optimization regimes is crucial for scalable and robust agent alignment. --- ### Language-Grounded Multi-Domain Image Translation via Semantic Difference Guidance Authors: Jongwon Ryu, Joonhyung Park, Jaeho Han, Yeong-Seok Kim, Hye-rin Kim, Sunjae Yoon, Junyeong Kim Published: 2026-01-12 Categories: cs.CV arXiv: https://arxiv.org/abs/2601.07221v1 Multi-domain image-to-image translation re quires grounding semantic differences ex pressed in natural language prompts into corresponding visual transformations, while preserving unrelated structural and seman tic content. Existing methods struggle to maintain structural integrity and provide fine grained, attribute-specific control, especially when multiple domains are involved. We propose LACE (Language-grounded Attribute Controllable Translation), built on two compo nents: (1) a GLIP-Adapter that fuses global semantics with local structural features to pre serve consistency, and (2) a Multi-Domain Control Guidance mechanism that explicitly grounds the semantic delta between source and target prompts into per-attribute translation vec tors, aligning linguistic semantics with domain level visual changes. Together, these modules enable compositional multi-domain control with independent strength modulation for each attribute. Experiments on CelebA(Dialog) and BDD100K demonstrate that LACE achieves high visual fidelity, structural preservation, and interpretable domain-specific control, surpass ing prior baselines. This positions LACE as a cross-modal content generation framework bridging language semantics and controllable visual translation. --- ### The Roots of Performance Disparity in Multilingual Language Models: Intrinsic Modeling Difficulty or Design Choices? Authors: Chen Shani, Yuval Reif, Nathan Roll, Dan Jurafsky, Ekaterina Shutova Published: 2026-01-12 Categories: cs.CL arXiv: https://arxiv.org/abs/2601.07220v3 Multilingual language models (LMs) promise broader NLP access, yet current systems deliver uneven performance across the world's languages. This survey examines why these gaps persist and whether they reflect intrinsic linguistic difficulty or modeling artifacts. We organize the literature around two questions: do linguistic disparities arise from representation and allocation choices (e.g., tokenization, encoding, data exposure, parameter sharing) rather than inherent complexity; and which design choices mitigate inequities across typologically diverse languages. We review linguistic features, such as orthography, morphology, lexical diversity, syntax, information density, and typological distance, linking each to concrete modeling mechanisms. Gaps often shrink when segmentation, encoding, and data exposure are normalized, suggesting much apparent difficulty stems from current modeling choices. We synthesize these insights into design recommendations for tokenization, sampling, architectures, and evaluation to support more balanced multilingual LMs. --- ### VENUS: Visual Editing with Noise Inversion Using Scene Graphs Authors: Thanh-Nhan Vo, Trong-Thuan Nguyen, Tam V. Nguyen, Minh-Triet Tran Published: 2026-01-12 Categories: cs.CV arXiv: https://arxiv.org/abs/2601.07219v1 State-of-the-art text-based image editing models often struggle to balance background preservation with semantic consistency, frequently resulting either in the synthesis of entirely new images or in outputs that fail to realize the intended edits. In contrast, scene graph-based image editing addresses this limitation by providing a structured representation of semantic entities and their relations, thereby offering improved controllability. However, existing scene graph editing methods typically depend on model fine-tuning, which incurs high computational cost and limits scalability. To this end, we introduce VENUS (Visual Editing with Noise inversion Using Scene graphs), a training-free framework for scene graph-guided image editing. Specifically, VENUS employs a split prompt conditioning strategy that disentangles the target object of the edit from its background context, while simultaneously leveraging noise inversion to preserve fidelity in unedited regions. Moreover, our proposed approach integrates scene graphs extracted from multimodal large language models with diffusion backbones, without requiring any additional training. Empirically, VENUS substantially improves both background preservation and semantic alignment on PIE-Bench, increasing PSNR from 22.45 to 24.80, SSIM from 0.79 to 0.84, and reducing LPIPS from 0.100 to 0.070 relative to the state-of-the-art scene graph editing model (SGEdit). In addition, VENUS enhances semantic consistency as measured by CLIP similarity (24.97 vs. 24.19). On EditVal, VENUS achieves the highest fidelity with a 0.87 DINO score and, crucially, reduces per-image runtime from 6-10 minutes to only 20-30 seconds. Beyond scene graph-based editing, VENUS also surpasses strong text-based editing baselines such as LEDIT++ and P2P+DirInv, thereby demonstrating consistent improvements across both paradigms. --- ### SceneNAT: Masked Generative Modeling for Language-Guided Indoor Scene Synthesis Authors: Jeongjun Choi, Yeonsoo Park, H. Jin Kim Published: 2026-01-12 Categories: cs.CV arXiv: https://arxiv.org/abs/2601.07218v2 We present SceneNAT, a masked non-autoregressive Transformer for 3D indoor scene synthesis from natural language instructions. It generates complete scenes in a few parallel decoding passes, improving both quality and efficiency over prior methods. SceneNAT is trained via masked modeling over fully discretized representations of both semantic and spatial attributes. By applying a masking strategy at both the attribute level and the instance level, the model can better capture intra-object and inter-object structure. To boost relational reasoning, SceneNAT employs a relational reasoning module (RRM) that captures implicit spatial constraints. By formulating relation modeling as a set prediction task, it extracts structure-aware features to guide the layout generation without explicit sequential parsing. Extensive experiments on 3D-FRONT show that SceneNAT outperforms state-of-the-art autoregressive and diffusion baselines in both semantic compliance and spatial arrangement accuracy while using substantially lower computational cost, enabling high-throughput generation of diverse scenes at scale. --- ### BlindU: Blind Machine Unlearning without Revealing Erasing Data Authors: Weiqi Wang, Zhiyi Tian, Chenhan Zhang, Shui Yu Published: 2026-01-12 Categories: cs.CR, cs.AI, cs.CV arXiv: https://arxiv.org/abs/2601.07214v1 Machine unlearning enables data holders to remove the contribution of their specified samples from trained models to protect their privacy. However, it is paradoxical that most unlearning methods require the unlearning requesters to firstly upload their data to the server as a prerequisite for unlearning. These methods are infeasible in many privacy-preserving scenarios where servers are prohibited from accessing users' data, such as federated learning (FL). In this paper, we explore how to implement unlearning under the condition of not uncovering the erasing data to the server. We propose \textbf{Blind Unlearning (BlindU)}, which carries out unlearning using compressed representations instead of original inputs. BlindU only involves the server and the unlearning user: the user locally generates privacy-preserving representations, and the server performs unlearning solely on these representations and their labels. For the FL model training, we employ the information bottleneck (IB) mechanism. The encoder of the IB-based FL model learns representations that distort maximum task-irrelevant information from inputs, allowing FL users to generate compressed representations locally. For effective unlearning using compressed representation, BlindU integrates two dedicated unlearning modules tailored explicitly for IB-based models and uses a multiple gradient descent algorithm to balance forgetting and utility retaining. While IB compression already provides protection for task-irrelevant information of inputs, to further enhance the privacy protection, we introduce a noise-free differential privacy (DP) masking method to deal with the raw erasing data before compressing. Theoretical analysis and extensive experimental results illustrate the superiority of BlindU in privacy protection and unlearning effectiveness compared with the best existing privacy-preserving unlearning benchmarks. --- ### MI-PRUN: Optimize Large Language Model Pruning via Mutual Information Authors: Hao Zhang, Zhibin Zhang, Guangxin Wu, He Chen, Jiafeng Guo, Xueqi Cheng Published: 2026-01-12 Categories: cs.CL arXiv: https://arxiv.org/abs/2601.07212v1 Large Language Models (LLMs) have become indispensable across various domains, but this comes at the cost of substantial computational and memory resources. Model pruning addresses this by removing redundant components from models. In particular, block pruning can achieve significant compression and inference acceleration. However, existing block pruning methods are often unstable and struggle to attain globally optimal solutions. In this paper, we propose a mutual information based pruning method MI-PRUN for LLMs. Specifically, we leverages mutual information to identify redundant blocks by evaluating transitions in hidden states. Additionally, we incorporate the Data Processing Inequality (DPI) to reveal the relationship between the importance of entire contiguous blocks and that of individual blocks. Moreover, we develop the Fast-Block-Select algorithm, which iteratively updates block combinations to achieve a globally optimal solution while significantly improving the efficiency. Extensive experiments across various models and datasets demonstrate the stability and effectiveness of our method. --- ### SIRR-LMM: Single-image Reflection Removal via Large Multimodal Model Authors: Yu Guo, Zhiqiang Lao, Xiyun Song, Yubin Zhou, Heather Yu Published: 2026-01-12 Categories: cs.CV, cs.AI, cs.GR arXiv: https://arxiv.org/abs/2601.07209v2 Glass surfaces create complex interactions of reflected and transmitted light, making single-image reflection removal (SIRR) challenging. Existing datasets suffer from limited physical realism in synthetic data or insufficient scale in real captures. We introduce a synthetic dataset generation framework that path-traces 3D glass models over real background imagery to create physically accurate reflection scenarios with varied glass properties, camera settings, and post-processing effects. To leverage the capabilities of Large Multimodal Model (LMM), we concatenate the image layers into a single composite input, apply joint captioning, and fine-tune the model using task-specific LoRA rather than full-parameter training. This enables our approach to achieve improved reflection removal and separation performance compared to state-of-the-art methods. --- ### MAESTRO: Meta-learning Adaptive Estimation of Scalarization Trade-offs for Reward Optimization Authors: Yang Zhao, Hepeng Wang, Xiao Ding, Yangou Ouyang, Bibo Cai, Kai Xiong, Jinglong Gao, Zhouhao Sun, Li Du, Bing Qin, Ting Liu Published: 2026-01-12 Categories: cs.LG, cs.CL arXiv: https://arxiv.org/abs/2601.07208v2 Group-Relative Policy Optimization (GRPO) has emerged as an efficient paradigm for aligning Large Language Models (LLMs), yet its efficacy is primarily confined to domains with verifiable ground truths. Extending GRPO to open-domain settings remains a critical challenge, as unconstrained generation entails multi-faceted and often conflicting objectives - such as creativity versus factuality - where rigid, static reward scalarization is inherently suboptimal. To address this, we propose MAESTRO (Meta-learning Adaptive Estimation of Scalarization Trade-offs for Reward Optimization), which introduces a meta-cognitive orchestration layer that treats reward scalarization as a dynamic latent policy, leveraging the model's terminal hidden states as a semantic bottleneck to perceive task-specific priorities. We formulate this as a contextual bandit problem within a bi-level optimization framework, where a lightweight Conductor network co-evolves with the policy by utilizing group-relative advantages as a meta-reward signal. Across seven benchmarks, MAESTRO consistently outperforms single-reward and static multi-objective baselines, while preserving the efficiency advantages of GRPO, and in some settings even reducing redundant generation. --- ### LLMRouterBench: A Massive Benchmark and Unified Framework for LLM Routing Authors: Hao Li, Yiqun Zhang, Zhaoyan Guo, Chenxu Wang, Shengji Tang, Qiaosheng Zhang, Yang Chen, Biqing Qi, Peng Ye, Lei Bai, Zhen Wang, Shuyue Hu Published: 2026-01-12 Categories: cs.AI arXiv: https://arxiv.org/abs/2601.07206v1 Large language model (LLM) routing assigns each query to the most suitable model from an ensemble. We introduce LLMRouterBench, a large-scale benchmark and unified framework for LLM routing. It comprises over 400K instances from 21 datasets and 33 models. Moreover, it provides comprehensive metrics for both performance-oriented routing and performance-cost trade-off routing, and integrates 10 representative routing baselines. Using LLMRouterBench, we systematically re-evaluate the field. While confirming strong model complementarity-the central premise of LLM routing-we find that many routing methods exhibit similar performance under unified evaluation, and several recent approaches, including commercial routers, fail to reliably outperform a simple baseline. Meanwhile, a substantial gap remains to the Oracle, driven primarily by persistent model-recall failures. We further show that backbone embedding models have limited impact, that larger ensembles exhibit diminishing returns compared to careful model curation, and that the benchmark also enables latency-aware analysis. All code and data are available at https://github.com/ynulihao/LLMRouterBench. --- ### CalPro: Prior-Aware Evidential--Conformal Prediction with Structure-Aware Guarantees for Protein Structures Authors: Ibne Farabi Shihab, Sanjeda Akter, Anuj Sharma Published: 2026-01-12 Categories: cs.LG, cs.AI arXiv: https://arxiv.org/abs/2601.07201v1 Deep protein structure predictors such as AlphaFold provide confidence estimates (e.g., pLDDT) that are often miscalibrated and degrade under distribution shifts across experimental modalities, temporal changes, and intrinsically disordered regions. We introduce CalPro, a prior-aware evidential-conformal framework for shift-robust uncertainty quantification. CalPro combines (i) a geometric evidential head that outputs Normal-Inverse-Gamma predictive distributions via a graph-based architecture; (ii) a differentiable conformal layer that enables end-to-end training with finite-sample coverage guarantees; and (iii) domain priors (disorder, flexibility) encoded as soft constraints. We derive structure-aware coverage guarantees under distribution shift using PAC-Bayesian bounds over ambiguity sets, and show that CalPro maintains near-nominal coverage while producing tighter intervals than standard conformal methods in regions where priors are informative. Empirically, CalPro exhibits at most 5% coverage degradation across modalities (vs. 15-25% for baselines), reduces calibration error by 30-50%, and improves downstream ligand-docking success by 25%. Beyond proteins, CalPro applies to structured regression tasks in which priors encode local reliability, validated on non-biological benchmarks. --- ### Safeguarding LLM Fine-tuning via Push-Pull Distributional Alignment Authors: Haozhong Wang, Zhuo Li, Yibo Yang, He Zhao, Hongyuan Zha, Dandan Guo Published: 2026-01-12 Categories: cs.LG, cs.AI arXiv: https://arxiv.org/abs/2601.07200v1 The inherent safety alignment of Large Language Models (LLMs) is prone to erosion during fine-tuning, even when using seemingly innocuous datasets. While existing defenses attempt to mitigate this via data selection, they typically rely on heuristic, instance-level assessments that neglect the global geometry of the data distribution and fail to explicitly repel harmful patterns. To address this, we introduce Safety Optimal Transport (SOT), a novel framework that reframes safe fine-tuning from an instance-level filtering challenge to a distribution-level alignment task grounded in Optimal Transport (OT). At its core is a dual-reference ``push-pull'' weight-learning mechanism: SOT optimizes sample importance by actively pulling the downstream distribution towards a trusted safe anchor while simultaneously pushing it away from a general harmful reference. This establishes a robust geometric safety boundary that effectively purifies the training data. Extensive experiments across diverse model families and domains demonstrate that SOT significantly enhances model safety while maintaining competitive downstream performance, achieving a superior safety-utility trade-off compared to baselines. --- ### Forward versus Backward: Comparing Reasoning Objectives in Direct Preference Optimization Authors: Murtaza Nikzad, Raghuram Ramanujan Published: 2026-01-12 Categories: cs.LG, cs.AI arXiv: https://arxiv.org/abs/2601.07199v1 Large language models exhibit impressive reasoning capabilities yet frequently generate plausible but incorrect solutions, a phenomenon commonly termed hallucination. This paper investigates the effect of training objective composition on reasoning reliability through Direct Preference Optimization. Two complementary training signals are examined: forward chain-of-thought generation, which trains the model to produce correct reasoning traces, and backward verification, which trains the model to verify and acknowledge errors in candidate solutions. Experiments on GSM8K reveal a fundamental trade-off between these objectives. Forward-only DPO training achieves the highest accuracy improvement, increasing from 83.1% to 86.6% (+3.5 percentage points), while backward-only training yields minimal accuracy gains but substantially reduces the false positive rate from 13.4% to 4.3%. Notably, both training variants reduce acknowledgement rate compared to the baseline, suggesting that preference optimization increases model confidence in its outputs. These findings indicate that forward and backward reasoning objectives provide distinct and complementary learning signals: forward training improves problem-solving capability, while backward training improves verification calibration. The complete training and evaluation pipeline, implemented efficiently through Low-Rank Adaptation, is released to facilitate further research. --- ### Beyond Variance: Knowledge-Aware LLM Compression via Fisher-Aligned Subspace Diagnostics Authors: Ibne Farabi Shihab, Sanjeda Akter, Anuj Sharma Published: 2026-01-12 Categories: cs.LG, cs.AI arXiv: https://arxiv.org/abs/2601.07197v1 Post-training activation compression is essential for deploying Large Language Models (LLMs) on resource-constrained hardware. However, standard methods like Singular Value Decomposition (SVD) are gradient-blind: they preserve high-variance dimensions regardless of their impact on factual knowledge preservation. We introduce Fisher-Aligned Subspace Compression (FASC), a knowledge-aware compression framework that selects subspaces by directly modeling activation-gradient coupling, minimizing a second-order surrogate of the loss function. FASC leverages the Fisher Information Matrix to identify dimensions critical for factual knowledge, which often reside in low-variance but high-gradient-sensitivity subspaces. We propose the Dependence Violation Score (\r{ho}) as a general-purpose diagnostic metric that quantifies activation-gradient coupling, revealing where factual knowledge is stored within transformer architectures. Extensive experiments on Mistral-7B and Llama-3-8B demonstrate that FASC preserves 6-8% more accuracy on knowledge-intensive benchmarks (MMLU, LAMA) compared to variance-based methods at 50% rank reduction, effectively enabling a 7B model to match the factual recall of a 13B uncompressed model. Our analysis reveals that \r{ho} serves as a fundamental signal of stored knowledge, with high-\r{ho} layers emerging only when models internalize factual associations during training. --- ### Relink: Constructing Query-Driven Evidence Graph On-the-Fly for GraphRAG Authors: Manzong Huang, Chenyang Bu, Yi He, Xingrui Zhuo, Xindong Wu Published: 2026-01-12 Categories: cs.CL, cs.AI arXiv: https://arxiv.org/abs/2601.07192v1 Graph-based Retrieval-Augmented Generation (GraphRAG) mitigates hallucinations in Large Language Models (LLMs) by grounding them in structured knowledge. However, current GraphRAG methods are constrained by a prevailing \textit{build-then-reason} paradigm, which relies on a static, pre-constructed Knowledge Graph (KG). This paradigm faces two critical challenges. First, the KG's inherent incompleteness often breaks reasoning paths. Second, the graph's low signal-to-noise ratio introduces distractor facts, presenting query-relevant but misleading knowledge that disrupts the reasoning process. To address these challenges, we argue for a \textit{reason-and-construct} paradigm and propose Relink, a framework that dynamically builds a query-specific evidence graph. To tackle incompleteness, \textbf{Relink} instantiates required facts from a latent relation pool derived from the original text corpus, repairing broken paths on the fly. To handle misleading or distractor facts, Relink employs a unified, query-aware evaluation strategy that jointly considers candidates from both the KG and latent relations, selecting those most useful for answering the query rather than relying on their pre-existence. This empowers Relink to actively discard distractor facts and construct the most faithful and precise evidence path for each query. Extensive experiments on five Open-Domain Question Answering benchmarks show that Relink achieves significant average improvements of 5.4\% in EM and 5.2\% in F1 over leading GraphRAG baselines, demonstrating the superiority of our proposed framework. --- ### On Lie Groups Preserving Subspaces of Degenerate Clifford Algebras Authors: E. R. Filimoshina, D. S. Shirokov Published: 2026-01-12 Categories: math.RA, cs.LG, math-ph arXiv: https://arxiv.org/abs/2601.07191v1 This paper introduces Lie groups in degenerate geometric (Clifford) algebras that preserve four fundamental subspaces determined by the grade involution and reversion under the adjoint and twisted adjoint representations. We prove that these Lie groups can be equivalently defined using norm functions of multivectors applied in the theory of spin groups. We also study the corresponding Lie algebras. Some of these Lie groups and algebras are closely related to Heisenberg Lie groups and algebras. The introduced groups are interesting for various applications in physics and computer science, in particular, for constructing equivariant neural networks. --- ### Active Context Compression: Autonomous Memory Management in LLM Agents Authors: Nikhil Verma Published: 2026-01-12 Categories: cs.AI arXiv: https://arxiv.org/abs/2601.07190v1 Large Language Model (LLM) agents struggle with long-horizon software engineering tasks due to "Context Bloat." As interaction history grows, computational costs explode, latency increases, and reasoning capabilities degrade due to distraction by irrelevant past errors. Existing solutions often rely on passive, external summarization mechanisms that the agent cannot control. This paper proposes Focus, an agent-centric architecture inspired by the biological exploration strategies of Physarum polycephalum (slime mold). The Focus Agent autonomously decides when to consolidate key learnings into a persistent "Knowledge" block and actively withdraws (prunes) the raw interaction history. Using an optimized scaffold matching industry best practices (persistent bash + string-replacement editor), we evaluated Focus on N=5 context-intensive instances from SWE-bench Lite using Claude Haiku 4.5. With aggressive prompting that encourages frequent compression, Focus achieves 22.7% token reduction (14.9M -> 11.5M tokens) while maintaining identical accuracy (3/5 = 60% for both agents). Focus performed 6.0 autonomous compressions per task on average, with token savings up to 57% on individual instances. We demonstrate that capable models can autonomously self-regulate their context when given appropriate tools and prompting, opening pathways for cost-aware agentic systems without sacrificing task performance. --- ### Standardization of Post-Publication Code Verification by Journals is Possible with the Support of the Community Authors: Susana Lopez-Moreno, Eric Dolores-Cuenca, Sangil Kim Published: 2026-01-12 Categories: cs.LG arXiv: https://arxiv.org/abs/2601.07189v1 Reproducibility remains a challenge in machine learning research. While code and data availability requirements have become increasingly common, post-publication verification in journals is still limited and unformalized. This position paper argues that it is plausible for journals and conference proceedings to implement post-publication verification. We propose a modification to ACM pre-publication verification badges that allows independent researchers to submit post-publication code replications to the journal, leading to visible verification badges included in the article metadata. Each article may earn up to two badges, each linked to verified code in its corresponding public repository. We describe the motivation, related initiatives, a formal framework, the potential impact, possible limitations, and alternative views. --- ### Defenses Against Prompt Attacks Learn Surface Heuristics Authors: Shawn Li, Chenxiao Yu, Zhiyu Ni, Hao Li, Charith Peris, Chaowei Xiao, Yue Zhao Published: 2026-01-12 Categories: cs.CR, cs.AI arXiv: https://arxiv.org/abs/2601.07185v1 Large language models (LLMs) are increasingly deployed in security-sensitive applications, where they must follow system- or developer-specified instructions that define the intended task behavior, while completing benign user requests. When adversarial instructions appear in user queries or externally retrieved content, models may override intended logic. Recent defenses rely on supervised fine-tuning with benign and malicious labels. Although these methods achieve high attack rejection rates, we find that they rely on narrow correlations in defense data rather than harmful intent, leading to systematic rejection of safe inputs. We analyze three recurring shortcut behaviors induced by defense fine-tuning. \emph{Position bias} arises when benign content placed later in a prompt is rejected at much higher rates; across reasoning benchmarks, suffix-task rejection rises from below \textbf{10\%} to as high as \textbf{90\%}. \emph{Token trigger bias} occurs when strings common in attack data raise rejection probability even in benign contexts; inserting a single trigger token increases false refusals by up to \textbf{50\%}. \emph{Topic generalization bias} reflects poor generalization beyond the defense data distribution, with defended models suffering test-time accuracy drops of up to \textbf{40\%}. These findings suggest that current prompt-injection defenses frequently respond to attack-like surface patterns rather than the underlying intent. We introduce controlled diagnostic datasets and a systematic evaluation across two base models and multiple defense pipelines, highlighting limitations of supervised fine-tuning for reliable LLM security. --- ### PRPO: Aligning Process Reward with Outcome Reward in Policy Optimization Authors: Ruiyi Ding, Yongxuan Lv, Xianhui Meng, Jiahe Song, Chao Wang, Chen Jiang, Yuan Cheng Published: 2026-01-12 Categories: cs.LG, cs.AI arXiv: https://arxiv.org/abs/2601.07182v3 Policy optimization for large language models often suffers from sparse reward signals in multi-step reasoning tasks. Critic-free methods like GRPO assign a single normalized outcome reward to all tokens, providing limited guidance for intermediate reasoning . While Process Reward Models (PRMs) offer dense feedback, they risk premature collapse when used alone, as early low-reward tokens can drive policies toward truncated outputs. We introduce Process Relative Policy Optimization (PRPO), which combines outcome reliability with process-level guidance in a critic-free framework. PRPO segments reasoning sequences based on semantic clues, normalizes PRM scores into token-level advantages, and aligns their distribution with outcome advantages through location-parameter shift. On MATH500, PRPO improves Qwen2.5-Math-1.5B accuracy from 61.2% to 64.4% over GRPO using only eight rollouts and no value network, demonstrating efficient fine-grained credit assignment within critic-free optimization. Code is available at: https://github.com/SchumiDing/srpocode ---