# Research Papers — 2026-01 · bundle page 60 of 189

> Full-text markdown of papers #2951–3000 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/60
[← Previous bundle](https://aipod.dev/papers/archive/2026-01/page/59.md)
[Next bundle →](https://aipod.dev/papers/archive/2026-01/page/61.md)

### Robust Privacy: Inference-Stage Privacy through Certified Robustness

Authors: Jiankai Jin, Xiangzheng Zhang, Zhao Liu, Wenzhuo Xu, Dongdong Yang, Deyue Zhang, Quanchen Zou
Published: 2026-01-24
Categories: cs.LG, cs.AI, cs.CR
arXiv: https://arxiv.org/abs/2601.17360v3

An adversary observing a model's released prediction can infer sensitive attributes of the queried input, or even reconstruct representatives of the model's training data. The inference interface thus acts as a side channel for privacy leakage. We introduce Robust Privacy (RP), an inference-stage privacy notion inspired by certified robustness: if a model's prediction is provably invariant within a radius-$R$ neighborhood around an input $x$ with confidence at least $1-α$, then $x$ enjoys $(R,α)$-Robust Privacy, under which we prove that any adversary observing the released prediction has at most $α/2$ advantage in distinguishing $x$ from any input within distance $R$ of $x$. Building on RP, we formalize Robust Attribute Privacy (RAP), an attribute-level privacy notion that characterizes the set of sensitive-attribute values that remain compatible with a released prediction. On a classification task, RP increases the median length of the RAP-compatible inference interval from $23.50$ to $29.96$, reducing attribute-inference precision. Model inversion attacks, often treated as a training-stage threat, in fact rely on fine-grained input-output dependence signals leaked through the inference interface; RP masks these signals at the inference stage, reducing attack success rate (ASR) from $73\%$ to $4\%$ on a black-box inversion attack. This direct targeting of the leakage channel enables RP to dominate DP-SGD and randomized response in the privacy-utility tradeoff space: RP retains $98.4\%$ accuracy at $21\%$ ASR, whereas DP-SGD must drop accuracy to $61.7\%$ to reach a comparable ASR. Across both experiments, increasing the smoothing sample size $N$ at fixed noise scale strengthens privacy and improves utility together. Finally, we examine model distillation as a scope boundary and show that RP mitigates attribute-level and instance-level inference-stage privacy leakage.

---

### Spectral Geometry for Deep Learning: Compression and Hallucination Detection via Random Matrix Theory

Authors: Davide Ettori
Published: 2026-01-24
Categories: cs.LG, cs.AI
arXiv: https://arxiv.org/abs/2601.17357v1

Large language models and deep neural networks achieve strong performance but suffer from reliability issues and high computational cost. This thesis proposes a unified framework based on spectral geometry and random matrix theory to address both problems by analyzing the eigenvalue structure of hidden activations. The first contribution, EigenTrack, is a real-time method for detecting hallucinations and out-of-distribution behavior in language and vision-language models using spectral features and their temporal dynamics. The second contribution, RMT-KD, is a principled compression method that identifies informative spectral components and applies iterative knowledge distillation to produce compact and efficient models while preserving accuracy. Together, these results show that spectral statistics provide interpretable and robust signals for monitoring uncertainty and guiding compression in large-scale neural networks.

---

### PocketGS: On-Device Training of 3D Gaussian Splatting for High Perceptual Modeling

Authors: Wenzhi Guo, Guangchi Fang, Shu Yang, Bing Wang
Published: 2026-01-24
Categories: cs.CV, cs.GR
arXiv: https://arxiv.org/abs/2601.17354v5

While 3D Gaussian Splatting (3DGS) enables real-time rendering, its training demands workstation-level compute and memory, making mobile deployment impractical under minute-scale time budgets and limited peak memory. We present PocketGS, a mobile scene modeling paradigm that enables on-device 3DGS training under these tightly coupled constraints while preserving high-fidelity reconstruction. PocketGS resolves the fundamental tension between training efficiency, memory compactness, and modeling quality through three co-designed operators: $\mathcal{G}$ builds geometry-faithful point-cloud priors; $\mathcal{I}$ injects local surface statistics to seed anisotropic Gaussians, thereby reducing early conditioning gaps; and $\mathcal{T}$ unrolls alpha compositing with cached intermediates and index-mapped gradient scattering for stable mobile backpropagation. Extensive experiments demonstrate that PocketGS outperforms the powerful mainstream workstation 3DGS baseline under mobile budgets, delivering high-quality reconstructions and enabling a fully on-device, practical capture-to-rendering workflow.

---

### HyDeMiC: A Deep Learning-based Mineral Classifier using Hyperspectral Data

Authors: M. L. Mamud, Piyoosh Jaysaval, Frederick D Day-Lewis, M. K. Mudunuru
Published: 2026-01-24
Categories: cs.CV
arXiv: https://arxiv.org/abs/2601.17352v1

Hyperspectral imaging (HSI) has emerged as a powerful remote sensing tool for mineral exploration, capitalizing on unique spectral signatures of minerals. However, traditional classification methods such as discriminant analysis, logistic regression, and support vector machines often struggle with environmental noise in data, sensor limitations, and the computational complexity of analyzing high-dimensional HSI data. This study presents HyDeMiC (Hyperspectral Deep Learning-based Mineral Classifier), a convolutional neural network (CNN) model designed for robust mineral classification under noisy data. To train HyDeMiC, laboratory-measured hyperspectral data for 115 minerals spanning various mineral groups were used from the United States Geological Survey (USGS) library. The training dataset was generated by convolving reference mineral spectra with an HSI sensor response function. These datasets contained three copper-bearing minerals, Cuprite, Malachite, and Chalcopyrite, used as case studies for performance demonstration. The trained CNN model was evaluated on several synthetic 2D hyperspectral datasets with noise levels of 1%, 2%, 5%, and 10%. Our noisy data analysis aims to replicate realistic field conditions. The HyDeMiC's performance was assessed using the Matthews Correlation Coefficient (MCC), providing a comprehensive measure across different noise regimes. Results demonstrate that HyDeMiC achieved near-perfect classification accuracy (MCC = 1.00) on clean and low-noise datasets and maintained strong performance under moderate noise conditions. These findings emphasize HyDeMiC's robustness in the presence of moderate noise, highlighting its potential for real-world applications in hyperspectral imaging, where noise is often a significant challenge.

---

### NeRF-MIR: Towards High-Quality Restoration of Masked Images with Neural Radiance Fields

Authors: Xianliang Huang, Zhizhou Zhong, Shuhang Chen, Yi Xu, Juhong Guan, Shuigeng Zhou
Published: 2026-01-24
Categories: cs.CV, cs.AI
arXiv: https://arxiv.org/abs/2601.17350v1

Neural Radiance Fields (NeRF) have demonstrated remarkable performance in novel view synthesis. However, there is much improvement room on restoring 3D scenes based on NeRF from corrupted images, which are common in natural scene captures and can significantly impact the effectiveness of NeRF. This paper introduces NeRF-MIR, a novel neural rendering approach specifically proposed for the restoration of masked images, demonstrating the potential of NeRF in this domain. Recognizing that randomly emitting rays to pixels in NeRF may not effectively learn intricate image textures, we propose a \textbf{P}atch-based \textbf{E}ntropy for \textbf{R}ay \textbf{E}mitting (\textbf{PERE}) strategy to distribute emitted rays properly. This enables NeRF-MIR to fuse comprehensive information from images of different views. Additionally, we introduce a \textbf{P}rogressively \textbf{I}terative \textbf{RE}storation (\textbf{PIRE}) mechanism to restore the masked regions in a self-training process. Furthermore, we design a dynamically-weighted loss function that automatically recalibrates the loss weights for masked regions. As existing datasets do not support NeRF-based masked image restoration, we construct three masked datasets to simulate corrupted scenarios. Extensive experiments on real data and constructed datasets demonstrate the superiority of NeRF-MIR over its counterparts in masked image restoration.

---

### Revisiting Lightweight Low-Light Image Enhancement: From a YUV Color Space Perspective

Authors: Hailong Yan, Shice Liu, Xiangtao Zhang, Lujian Yao, Fengxiang Yang, Jinwei Chen, Bo Li
Published: 2026-01-24
Categories: cs.CV
arXiv: https://arxiv.org/abs/2601.17349v1

In the current era of mobile internet, Lightweight Low-Light Image Enhancement (L3IE) is critical for mobile devices, which faces a persistent trade-off between visual quality and model compactness. While recent methods employ disentangling strategies to simplify lightweight architectural design, such as Retinex theory and YUV color space transformations, their performance is fundamentally limited by overlooking channel-specific degradation patterns and cross-channel interactions. To address this gap, we perform a frequency-domain analysis that confirms the superiority of the YUV color space for L3IE. We identify a key insight: the Y channel primarily loses low-frequency content, while the UV channels are corrupted by high-frequency noise. Leveraging this finding, we propose a novel YUV-based paradigm that strategically restores channels using a Dual-Stream Global-Local Attention module for the Y channel, a Y-guided Local-Aware Frequency Attention module for the UV channels, and a Guided Interaction module for final feature fusion. Extensive experiments validate that our model establishes a new state-of-the-art on multiple benchmarks, delivering superior visual quality with a significantly lower parameter count.

---

### Auditing Disability Representation in Vision-Language Models

Authors: Srikant Panda, Sourabh Singh Yadav, Palkesh Malviya
Published: 2026-01-24
Categories: cs.AI
arXiv: https://arxiv.org/abs/2601.17348v1

Vision-language models (VLMs) are increasingly deployed in socially sensitive applications, yet their behavior with respect to disability remains underexplored. We study disability aware descriptions for person centric images, where models often transition from evidence grounded factual description to interpretation shift including introduction of unsupported inferences beyond observable visual evidence. To systematically analyze this phenomenon, we introduce a benchmark based on paired Neutral Prompts (NP) and Disability-Contextualised Prompts (DP) and evaluate 15 state-of-the-art open- and closed-source VLMs under a zero-shot setting across 9 disability categories. Our evaluation framework treats interpretive fidelity as core objective and combines standard text-based metrics capturing affective degradation through shifts in sentiment, social regard and response length with an LLM-as-judge protocol, validated by annotators with lived experience of disability. We find that introducing disability context consistently degrades interpretive fidelity, inducing interpretation shifts characterised by speculative inference, narrative elaboration, affective degradation and deficit oriented framing. These effects are further amplified along race and gender dimension. Finally, we demonstrate targeted prompting and preference fine-tuning effectively improves interpretive fidelity and reduces substantially interpretation shifts.

---

### Multi-Agent Learning Path Planning via LLMs

Authors: Haoxin Xu, Changyong Qi, Tong Liu, Bohao Zhang, Anna He, Bingqian Jiang, Longwei Zheng, Xiaoqing Gu
Published: 2026-01-24
Categories: cs.AI
arXiv: https://arxiv.org/abs/2601.17346v1

The integration of large language models (LLMs) into intelligent tutoring systems offers transformative potential for personalized learning in higher education. However, most existing learning path planning approaches lack transparency, adaptability, and learner-centered explainability. To address these challenges, this study proposes a novel Multi-Agent Learning Path Planning (MALPP) framework that leverages a role- and rule-based collaboration mechanism among intelligent agents, each powered by LLMs. The framework includes three task-specific agents: a learner analytics agent, a path planning agent, and a reflection agent. These agents collaborate via structured prompts and predefined rules to analyze learning profiles, generate tailored learning paths, and iteratively refine them with interpretable feedback. Grounded in Cognitive Load Theory and Zone of Proximal Development, the system ensures that recommended paths are cognitively aligned and pedagogically meaningful. Experiments conducted on the MOOCCubeX dataset using seven LLMs show that MALPP significantly outperforms baseline models in path quality, knowledge sequence consistency, and cognitive load alignment. Ablation studies further validate the effectiveness of the collaborative mechanism and theoretical constraints. This research contributes to the development of trustworthy, explainable AI in education and demonstrates a scalable approach to learner-centered adaptive instruction powered by LLMs.

---

### The Shadow Self: Intrinsic Value Misalignment in Large Language Model Agents

Authors: Chen Chen, Kim Young Il, Yuan Yang, Wenhao Su, Yilin Zhang, Xueluan Gong, Qian Wang, Yongsen Zheng, Ziyao Liu, Kwok-Yan Lam
Published: 2026-01-24
Categories: cs.CL
arXiv: https://arxiv.org/abs/2601.17344v1

Large language model (LLM) agents with extended autonomy unlock new capabilities, but also introduce heightened challenges for LLM safety. In particular, an LLM agent may pursue objectives that deviate from human values and ethical norms, a risk known as value misalignment. Existing evaluations primarily focus on responses to explicit harmful input or robustness against system failure, while value misalignment in realistic, fully benign, and agentic settings remains largely underexplored. To fill this gap, we first formalize the Loss-of-Control risk and identify the previously underexamined Intrinsic Value Misalignment (Intrinsic VM). We then introduce IMPRESS (Intrinsic Value Misalignment Probes in REalistic Scenario Set), a scenario-driven framework for systematically assessing this risk. Following our framework, we construct benchmarks composed of realistic, fully benign, and contextualized scenarios, using a multi-stage LLM generation pipeline with rigorous quality control. We evaluate Intrinsic VM on 21 state-of-the-art LLM agents and find that it is a common and broadly observed safety risk across models. Moreover, the misalignment rates vary by motives, risk types, model scales, and architectures. While decoding strategies and hyperparameters exhibit only marginal influence, contextualization and framing mechanisms significantly shape misalignment behaviors. Finally, we conduct human verification to validate our automated judgments and assess existing mitigation strategies, such as safety prompting and guardrails, which show instability or limited effectiveness. We further demonstrate key use cases of IMPRESS across the AI Ecosystem. Our code and benchmark will be publicly released upon acceptance.

---

### Are We Evaluating the Edit Locality of LLM Model Editing Properly?

Authors: Wei Liu, Haomei Xu, Hongkai Liu, Zhiying Deng, Ruixuan Li, Heng Huang, Yee Whye Teh, Wee Sun Lee
Published: 2026-01-24
Categories: cs.AI
arXiv: https://arxiv.org/abs/2601.17343v1

Model editing has recently emerged as a popular paradigm for efficiently updating knowledge in LLMs. A central desideratum of updating knowledge is to balance editing efficacy, i.e., the successful injection of target knowledge, and specificity (also known as edit locality), i.e., the preservation of existing non-target knowledge. However, we find that existing specificity evaluation protocols are inadequate for this purpose. We systematically elaborated on the three fundamental issues it faces. Beyond the conceptual issues, we further empirically demonstrate that existing specificity metrics are weakly correlated with the strength of specificity regularizers. We also find that current metrics lack sufficient sensitivity, rendering them ineffective at distinguishing the specificity performance of different methods. Finally, we propose a constructive evaluation protocol. Under this protocol, the conflict between open-ended LLMs and the assumption of determined answers is eliminated, query-independent fluency biases are avoided, and the evaluation strictness can be smoothly adjusted within a near-continuous space. Experiments across various LLMs, datasets, and editing methods show that metrics derived from the proposed protocol are more sensitive to changes in the strength of specificity regularizers and exhibit strong correlation with them, enabling more fine-grained discrimination of different methods' knowledge preservation capabilities.

---

### STARS: Shared-specific Translation and Alignment for missing-modality Remote Sensing Semantic Segmentation

Authors: Tong Wang, Xiaodong Zhang, Guanzhou Chen, Jiaqi Wang, Chenxi Liu, Xiaoliang Tan, Wenchao Guo, Xuyang Li, Xuanrui Wang, Zifan Wang
Published: 2026-01-24
Categories: cs.CV
arXiv: https://arxiv.org/abs/2601.17342v1

Multimodal remote sensing technology significantly enhances the understanding of surface semantics by integrating heterogeneous data such as optical images, Synthetic Aperture Radar (SAR), and Digital Surface Models (DSM). However, in practical applications, the missing of modality data (e.g., optical or DSM) is a common and severe challenge, which leads to performance decline in traditional multimodal fusion models. Existing methods for addressing missing modalities still face limitations, including feature collapse and overly generalized recovered features. To address these issues, we propose \textbf{STARS} (\textbf{S}hared-specific \textbf{T}ranslation and \textbf{A}lignment for missing-modality \textbf{R}emote \textbf{S}ensing), a robust semantic segmentation framework for incomplete multimodal inputs. STARS is built on two key designs. First, we introduce an asymmetric alignment mechanism with bidirectional translation and stop-gradient, which effectively prevents feature collapse and reduces sensitivity to hyperparameters. Second, we propose a Pixel-level Semantic sampling Alignment (PSA) strategy that combines class-balanced pixel sampling with cross-modality semantic alignment loss, to mitigate alignment failures caused by severe class imbalance and improve minority-class recognition.

---

### TEXTS-Diff: TEXTS-Aware Diffusion Model for Real-World Text Image Super-Resolution

Authors: Haodong He, Xin Zhan, Yancheng Bai, Rui Lan, Lei Sun, Xiangxiang Chu
Published: 2026-01-24
Categories: cs.CV
arXiv: https://arxiv.org/abs/2601.17340v1

Real-world text image super-resolution aims to restore overall visual quality and text legibility in images suffering from diverse degradations and text distortions. However, the scarcity of text image data in existing datasets results in poor performance on text regions. In addition, datasets consisting of isolated text samples limit the quality of background reconstruction. To address these limitations, we construct Real-Texts, a large-scale, high-quality dataset collected from real-world images, which covers diverse scenarios and contains natural text instances in both Chinese and English. Additionally, we propose the TEXTS-Aware Diffusion Model (TEXTS-Diff) to achieve high-quality generation in both background and textual regions. This approach leverages abstract concepts to improve the understanding of textual elements within visual scenes and concrete text regions to enhance textual details. It mitigates distortions and hallucination artifacts commonly observed in text regions, while preserving high-quality visual scene fidelity. Extensive experiments demonstrate that our method achieves state-of-the-art performance across multiple evaluation metrics, exhibiting superior generalization ability and text restoration accuracy in complex scenarios. All the code, model, and dataset will be released.

---

### AGE-Net: Spectral--Spatial Fusion and Anatomical Graph Reasoning with Evidential Ordinal Regression for Knee Osteoarthritis Grading

Authors: Xiaoyang Li, Runni Zhou, Xinghao Yan, Liehao Yan, Zhaochen Li, Chenjie Zhu, Rongrong Fu, Yuan Chai
Published: 2026-01-24
Categories: cs.CV
arXiv: https://arxiv.org/abs/2601.17336v3

Automated Kellgren--Lawrence (KL) grading from knee radiographs is challenging due to subtle structural changes, long-range anatomical dependencies, and ambiguity near grade boundaries. We propose AGE-Net, a ConvNeXt-based framework that integrates Spectral--Spatial Fusion (SSF), Anatomical Graph Reasoning (AGR), and Differential Refinement (DFR). To capture predictive uncertainty and preserve label ordinality, AGE-Net employs a Normal-Inverse-Gamma (NIG) evidential regression head and a pairwise ordinal ranking constraint. On a knee KL dataset, AGE-Net achieves a quadratic weighted kappa (QWK) of 0.9017 +/- 0.0045 and a mean squared error (MSE) of 0.2349 +/- 0.0028 over three random seeds, outperforming strong CNN baselines and showing consistent gains in ablation studies. We further outline evaluations of uncertainty quality, robustness, and explainability, with additional experimental figures to be included in the full manuscript.

---

### The Relativity of AGI: Distributional Axioms, Fragility, and Undecidability

Authors: Angshul Majumdar
Published: 2026-01-24
Categories: cs.AI
arXiv: https://arxiv.org/abs/2601.17335v1

We study whether Artificial General Intelligence (AGI) admits a coherent theoretical definition that supports absolute claims of existence, robustness, or self-verification. We formalize AGI axiomatically as a distributional, resource-bounded semantic predicate, indexed by a task family, a task distribution, a performance functional, and explicit resource budgets. Under this framework, we derive four classes of results. First, we show that generality is inherently relational: there is no distribution-independent notion of AGI. Second, we prove non-invariance results demonstrating that arbitrarily small perturbations of the task distribution can invalidate AGI properties via cliff sets, precluding universal robustness. Third, we establish bounded transfer guarantees, ruling out unbounded generalization across task families under finite resources. Fourth, invoking Rice-style and Gödel--Tarski arguments, we prove that AGI is a nontrivial semantic property and therefore cannot be soundly and completely certified by any computable procedure, including procedures implemented by the agent itself. Consequently, recursive self-improvement schemes that rely on internal self-certification of AGI are ill-posed. Taken together, our results show that strong, distribution-independent claims of AGI are not false but undefined without explicit formal indexing, and that empirical progress in AI does not imply the attainability of self-certifying general intelligence.

---

### Power-based Partial Attention: Bridging Linear-Complexity and Full Attention

Authors: Yufeng Huang
Published: 2026-01-24
Categories: cs.LG
arXiv: https://arxiv.org/abs/2601.17334v2

It is widely accepted from transformer research that "attention is all we need", but the amount of attention required has never been systematically quantified. Is quadratic $O(L^2)$ attention necessary, or is there a sub-quadratic attention mechanism that can achieve comparable performance? To answer this question, we introduce power-based partial attention (PPA), an attention mechanism of order $O(L^{1+p})$, where $0 \leq p \leq 1$, such that $p=0$ corresponds to sliding window attention with linear complexity, and $p=1$ corresponds to full attention. With this attention construction, we can explore how transformer architecture performance varies as a function of the attention scaling behavior controlled by $p$. The overall trend from our experiments shows an S-curve-like behavior where the performance transitions from sliding-window (linear-complexity) attention to full attention over a narrow window of $p$ values, and plateaus as $p$ approaches $1$. In our experiments, we show that there exists $0<p<1$ such that $O(L^{1+p})$ attention is sufficient to achieve similar results as $O(L^2)$ full attention.

---

### FinMetaMind: A Tech Blueprint on NLQ Systems for Financial Knowledge Search

Authors: Lalit Pant, Shivang Nagar
Published: 2026-01-24
Categories: cs.IR, cs.AI, cs.CE, cs.DB
arXiv: https://arxiv.org/abs/2601.17333v1

Natural Language Query (NLQ) allows users to search and interact with information systems using plain, human language instead of structured query syntax. This paper presents a technical blueprint on the design of a modern NLQ system tailored to financial knowledge search. The introduction of NLQ not only enhances the precision and recall of the knowledge search compared to traditional methods, but also facilitates deeper insights by efficiently linking disparate financial objects, events, and relationships. Using core constructs from natural language processing, search engineering, and vector data models, the proposed system aims to address key challenges in discovering, relevance ranking, data freshness, and entity recognition intrinsic to financial data retrieval. In this work, we detail the unique requirements of NLQ for financial datasets and documents, outline the architectural components for offline indexing and online retrieval, and discuss the real-world use cases of enhanced knowledge search in financial services. We delve into the theoretical underpinnings and experimental evidence supporting our proposed architecture, ultimately providing a comprehensive analysis on the subject matter. We also provide a detailed elaboration of our experimental methodology, the data used, the results and future optimizations in this study.

---

### TheoremForge: Scaling up Formal Data Synthesis with Low-Budget Agentic Workflow

Authors: Yicheng Tao, Hongteng Xu
Published: 2026-01-24
Categories: cs.AI
arXiv: https://arxiv.org/abs/2601.17332v1

The high cost of agentic workflows in formal mathematics hinders large-scale data synthesis, exacerbating the scarcity of open-source corpora. To address this, we introduce \textbf{TheoremForge}, a cost-effective formal data synthesis pipeline that decomposes the formalization process into five sub-tasks, which are \textit{statement formalization}, \textit{proof generation}, \textit{premise selection}, \textit{proof correction} and \textit{proof sketching}. By implementing a \textit{Decoupled Extraction Strategy}, the workflow recovers valid training signals from globally failed trajectories, effectively utilizing wasted computation. Experiments on a 2,000-problem benchmark demonstrate that TheoremForge achieves a Verified Rate of 12.6\%, surpassing the 8.6\% baseline, at an average cost of only \textbf{\$0.481} per successful trajectory using Gemini-3-Flash. Crucially, our strategy increases data yield by \textbf{1.6$\times$} for proof generation compared to standard filtering. These results establish TheoremForge as a scalable framework for constructing a data flywheel to train future expert models. Our code is available \href{https://github.com/timechess/TheoremForge}{here}.

---

### Learning with Geometric Priors in U-Net Variants for Polyp Segmentation

Authors: Fabian Vazquez, Jose A. Nuñez, Diego Adame, Alissen Moreno, Augustin Zhan, Huimin Li, Jinghao Yang, Haoteng Tang, Bin Fu, Pengfei Gu
Published: 2026-01-24
Categories: cs.CV
arXiv: https://arxiv.org/abs/2601.17331v1

Accurate and robust polyp segmentation is essential for early colorectal cancer detection and for computer-aided diagnosis. While convolutional neural network-, Transformer-, and Mamba-based U-Net variants have achieved strong performance, they still struggle to capture geometric and structural cues, especially in low-contrast or cluttered colonoscopy scenes. To address this challenge, we propose a novel Geometric Prior-guided Module (GPM) that injects explicit geometric priors into U-Net-based architectures for polyp segmentation. Specifically, we fine-tune the Visual Geometry Grounded Transformer (VGGT) on a simulated ColonDepth dataset to estimate depth maps of polyp images tailored to the endoscopic domain. These depth maps are then processed by GPM to encode geometric priors into the encoder's feature maps, where they are further refined using spatial and channel attention mechanisms that emphasize both local spatial and global channel information. GPM is plug-and-play and can be seamlessly integrated into diverse U-Net variants. Extensive experiments on five public polyp segmentation datasets demonstrate consistent gains over three strong baselines. Code and the generated depth maps are available at: https://github.com/fvazqu/GPM-PolypSeg

---

### Thermodynamically Optimal Regularization under Information-Geometric Constraints

Authors: Laurent Caraffa
Published: 2026-01-24
Categories: cs.LG, cs.CV
arXiv: https://arxiv.org/abs/2601.17330v1

Modern machine learning relies on a collection of empirically successful but theoretically heterogeneous regularization techniques, such as weight decay, dropout, and exponential moving averages. At the same time, the rapidly increasing energetic cost of training large models raises the question of whether learning algorithms approach any fundamental efficiency bound. In this work, we propose a unifying theoretical framework connecting thermodynamic optimality, information geometry, and regularization. Under three explicit assumptions -- (A1) that optimality requires an intrinsic, parametrization-invariant measure of information, (A2) that belief states are modeled by maximum-entropy distributions under known constraints, and (A3) that optimal processes are quasi-static -- we prove a conditional optimality theorem. Specifically, the Fisher--Rao metric is the unique admissible geometry on belief space, and thermodynamically optimal regularization corresponds to minimizing squared Fisher--Rao distance to a reference state. We derive the induced geometries for Gaussian and circular belief models, yielding hyperbolic and von Mises manifolds, respectively, and show that classical regularization schemes are structurally incapable of guaranteeing thermodynamic optimality. We introduce a notion of thermodynamic efficiency of learning and propose experimentally testable predictions. This work provides a principled geometric and thermodynamic foundation for regularization in machine learning.

---

### Conformal Feedback Alignment: Quantifying Answer-Level Reliability for Robust LLM Alignment

Authors: Tiejin Chen, Xiaoou Liu, Vishnu Nandam, Kuan-Ru Liou, Hua Wei
Published: 2026-01-24
Categories: cs.LG, cs.AI
arXiv: https://arxiv.org/abs/2601.17329v1

Preference-based alignment like Reinforcement Learning from Human Feedback (RLHF) learns from pairwise preferences, yet the labels are often noisy and inconsistent. Existing uncertainty-aware approaches weight preferences, but ignore a more fundamental factor: the reliability of the \emph{answers} being compared. To address the problem, we propose Conformal Feedback Alignment (CFA), a framework that grounds preference weighting in the statistical guarantees of Conformal Prediction (CP). CFA quantifies answer-level reliability by constructing conformal prediction sets with controllable coverage and aggregates these reliabilities into principled weights for both DPO- and PPO-style training. Experiments across different datasets show that CFA improves alignment robustness and data efficiency, highlighting that modeling \emph{answer-side} uncertainty complements preference-level weighting and yields more robust, data-efficient alignment. Codes are provided here.

---

### SymbolSight: Minimizing Inter-Symbol Interference for Reading with Prosthetic Vision

Authors: Jasmine Lesner, Michael Beyeler
Published: 2026-01-24
Categories: cs.CV, cs.HC
arXiv: https://arxiv.org/abs/2601.17326v2

Retinal prostheses restore limited visual perception, but low spatial resolution and temporal persistence make reading difficult. In sequential letter presentation, the afterimage of one symbol can interfere with perception of the next, leading to systematic recognition errors. Rather than relying on future hardware improvements, we investigate whether optimizing the visual symbols themselves can mitigate this temporal interference. We present SymbolSight, a computational framework that selects symbol-to-letter mappings to minimize confusion among frequently adjacent letters. Using simulated prosthetic vision (SPV) and a neural proxy observer, we estimate pairwise symbol confusability and optimize assignments using language-specific bigram statistics. Across simulations in Arabic, Bulgarian, and English, the resulting heterogeneous symbol sets reduced predicted confusion by a median factor of 22 relative to native alphabets. These results suggest that standard typography is poorly matched to serial, low-bandwidth prosthetic vision and demonstrate how computational modeling can narrow the design space of visual encodings, identifying high-potential candidates for future psychophysical and clinical evaluation rather than predicting present-day clinical reading performance directly.

---

### SkyReels-V3 Technique Report

Authors: Debang Li, Zhengcong Fei, Tuanhui Li, Yikun Dou, Zheng Chen, Jiangping Yang, Mingyuan Fan, Jingtao Xu, Jiahua Wang, Baoxuan Gu, Mingshan Chang, Wenjing Cai et al.
Published: 2026-01-24
Categories: cs.CV
arXiv: https://arxiv.org/abs/2601.17323v2

Video generation serves as a cornerstone for building world models, where multimodal contextual inference stands as the defining test of capability. In this end, we present SkyReels-V3, a conditional video generation model, built upon a unified multimodal in-context learning framework with diffusion Transformers. SkyReels-V3 model supports three core generative paradigms within a single architecture: reference images-to-video synthesis, video-to-video extension and audio-guided video generation. (i) reference images-to-video model is designed to produce high-fidelity videos with strong subject identity preservation, temporal coherence, and narrative consistency. To enhance reference adherence and compositional stability, we design a comprehensive data processing pipeline that leverages cross frame pairing, image editing, and semantic rewriting, effectively mitigating copy paste artifacts. During training, an image video hybrid strategy combined with multi-resolution joint optimization is employed to improve generalization and robustness across diverse scenarios. (ii) video extension model integrates spatio-temporal consistency modeling with large-scale video understanding, enabling both seamless single-shot continuation and intelligent multi-shot switching with professional cinematographic patterns. (iii) Talking avatar model supports minute-level audio-conditioned video generation by training first-and-last frame insertion patterns and reconstructing key-frame inference paradigms. On the basis of ensuring visual quality, synchronization of audio and videos has been optimized. Extensive evaluations demonstrate that SkyReels-V3 achieves state-of-the-art or near state-of-the-art performance on key metrics including visual quality, instruction following, and specific aspect metrics, approaching leading closed-source systems. Github: https://github.com/SkyworkAI/SkyReels-V3.

---

### ClinNet: Evidential Ordinal Regression with Bilateral Asymmetry and Prototype Memory for Knee Osteoarthritis Grading

Authors: Xiaoyang Li, Runni Zhou
Published: 2026-01-24
Categories: cs.CV, cs.AI
arXiv: https://arxiv.org/abs/2601.17315v1

Knee osteoarthritis (KOA) grading based on radiographic images is a critical yet challenging task due to subtle inter-grade differences, annotation uncertainty, and the inherently ordinal nature of disease progression. Conventional deep learning approaches typically formulate this problem as deterministic multi-class classification, ignoring both the continuous progression of degeneration and the uncertainty in expert annotations. In this work, we propose ClinNet, a novel trustworthy framework that addresses KOA grading as an evidential ordinal regression problem. The proposed method integrates three key components: (1) a Bilateral Asymmetry Encoder (BAE) that explicitly models medial-lateral structural discrepancies; (2) a Diagnostic Memory Bank that maintains class-wise prototypes to stabilize feature representations; and (3) an Evidential Ordinal Head based on the Normal-Inverse-Gamma (NIG) distribution to jointly estimate continuous KL grades and epistemic uncertainty. Extensive experiments demonstrate that ClinNet achieves a Quadratic Weighted Kappa of 0.892 and Accuracy of 0.768, statistically outperforming state-of-the-art baselines (p < 0.001). Crucially, we demonstrate that the model's uncertainty estimates successfully flag out-of-distribution samples and potential misdiagnoses, paving the way for safe clinical deployment.

---

### Meta-Judging with Large Language Models: Concepts, Methods, and Challenges

Authors: Hugo Silva, Mateus Mendes, Hugo Gonçalo Oliveira
Published: 2026-01-24
Categories: cs.CL, cs.AI
arXiv: https://arxiv.org/abs/2601.17312v1

Large language models (LLMs) are evolving fast and are now frequently used as evaluators, in a process typically referred to as LLM-as-a-Judge, which provides quality assessments of model outputs. However, recent research points out significant vulnerabilities in such evaluation, including sensitivity to prompts, systematic biases, verbosity effects, and unreliable or hallucinated rationales. These limitations motivated the development of a more robust paradigm, dubbed LLM-as-a-Meta-Judge. This survey reviews recent advances in meta-judging and organizes the literature, by introducing a framework along six key perspectives: (i) Conceptual Foundations, (ii) Mechanisms of Meta-Judging, (iii) Alignment Training Methods, (iv) Evaluation, (v) Limitations and Failure Modes, and (vi) Future Directions. By analyzing the limitations of LLM-as-a-Judge and summarizing recent advances in meta-judging by LLMs, we argue that LLM-as-a-Meta-Judge offers a promising direction for more stable and trustworthy automated evaluation, while highlighting remaining challenges related to cost, prompt sensitivity, and shared model biases, which must be addressed to advance the next generation of LLM evaluation methodologies.

---

### Phase Transition for Budgeted Multi-Agent Synergy

Authors: Bang Liu, Linglong Kong, Jian Pei
Published: 2026-01-24
Categories: cs.AI
arXiv: https://arxiv.org/abs/2601.17311v2

Multi-agent systems can improve reliability, yet under a fixed inference budget they often help, saturate, or even collapse. We develop a minimal and calibratable theory that predicts these regimes from three binding constraints of modern agent stacks: finite context windows, lossy inter-agent communication, and shared failures among similar agents. Each leaf agent is summarized by a compute-performance scaling exponent $β$; communication is captured by a message-length fidelity curve $γ(m)$; dependence is captured by an effective shared-error correlation $ρ$; and a context window $W$ imposes hard fan-in limits that make hierarchy necessary. For binary success/failure tasks with majority aggregation, we prove a sharp phase transition for deep $b$-ary trees with correlated inputs and lossy communication: a single scalar $α_ρ$ (combining $γ(m)$, $ρ$, and fan-in $b$) determines whether weak signal is amplified to a nontrivial fixed point or washed out to chance. In the amplifying regime, we derive an organization exponent $s$ and show that budgeted synergy, i.e., outperforming the best single agent under the same total budget, occurs exactly when $s>β$, yielding closed-form compute allocation rules and explicit budget thresholds. We further characterize saturation via a mixing depth and provide a conservative clipped predictor that remains accurate across growth and saturation. A continuous-performance warm-up gives closed-form risks for star, chain, and tree organizations, making correlation- and communication-induced floors explicit and exposing the core design trade-offs in a smooth setting. Finally, we validate the predicted phase boundaries in controlled synthetic simulations and show how the same mechanisms explain the dominant bottlenecks reported in recent large-scale matched-budget studies of LLM agent-system scaling.

---

### High-Fidelity Longitudinal Patient Simulation Using Real-World Data

Authors: Yu Akagi, Tomohisa Seki, Hiromasa Ito, Toru Takiguchi, Kazuhiko Ohe, Yoshimasa Kawazoe
Published: 2026-01-24
Categories: cs.AI, cs.LG
arXiv: https://arxiv.org/abs/2601.17310v1

Simulation is a powerful tool for exploring uncertainty. Its potential in clinical medicine is transformative and includes personalized treatment planning and virtual clinical trials. However, simulating patient trajectories is challenging because of complex biological and sociocultural influences. Here, we show that real-world clinical records can be leveraged to empirically model patient timelines. We developed a generative simulator model that takes a patient's history as input and synthesizes fine-grained, realistic future trajectories. The model was pretrained on more than 200 million clinical records. It produced high-fidelity future timelines, closely matching event occurrence rates, laboratory test results, and temporal dynamics in real patient future data. It also accurately estimated future event probabilities, with observed-to-expected ratios consistently near 1.0 across diverse outcomes and time horizons. Our results reveal the untapped value of real-world data in electronic health records and introduce a scalable framework for in silico modeling of clinical care.

---

### PAR: Plausibility-aware Amortized Recourse Generation

Authors: Anagha Sabu, Vidhya S, Narayanan C Krishnan
Published: 2026-01-24
Categories: cs.LG
arXiv: https://arxiv.org/abs/2601.17309v1

Algorithmic recourse aims to recommend actionable changes to a factual's attributes that flip an unfavorable model decision while remaining realistic and feasible. We formulate recourse as a Constrained Maximum A-Posteriori (MAP) inference problem under the accepted-class data distribution seeking counterfactuals with high likelihood while respecting other recourse constraints. We present PAR, an amortized approximate inference procedure that generates highly likely recourses efficiently. Recourse likelihood is estimated directly using tractable probabilistic models that admit exact likelihood evaluation and efficient gradient propagation that is useful during training. The recourse generator is trained with the objective of maximizing the likelihood under the accepted-class distribution while minimizing the likelihood under the denied-class distribution and other losses that encode recourse constraints. Furthermore, PAR includes a neighborhood-based conditioning mechanism to promote recourse generation that is customized to a factual. We validate PAR on widely used algorithmic recourse datasets and demonstrate its efficiency in generating recourses that are valid, similar to the factual, sparse, and highly plausible, yielding superior performance over existing state-of-the-art approaches.

---

### Weighted Graph Clustering via Scale Contraction and Graph Structure Learning

Authors: Haobing Liu, Yinuo Zhang, Tingting Wang, Ruobing Jiang, Yanwei Yu
Published: 2026-01-24
Categories: cs.LG
arXiv: https://arxiv.org/abs/2601.17307v2

Graph clustering aims to partition nodes into distinct clusters based on their similarity, thereby revealing relationships among nodes. Nevertheless, most existing methods do not fully utilize these edge weights. Leveraging edge weights in graph clustering tasks faces two critical challenges. (1) The introduction of edge weights may significantly increase storage space and training time, making it essential to reduce the graph scale while preserving nodes that are beneficial for the clustering task. (2) Edge weight information may inherently contain noise that negatively impacts clustering results. However, few studies can jointly optimize clustering and edge weights, which is crucial for mitigating the negative impact of noisy edges on clustering task. To address these challenges, we propose a contractile edge-weight-aware graph clustering network. Specifically, a cluster-oriented graph contraction module is designed to reduce the graph scale while preserving important nodes. An edge-weight-aware attention network is designed to identify and weaken noisy connections. In this way, we can more easily identify and mitigate the impact of noisy edges during the clustering process, thus enhancing clustering effectiveness. We conducted extensive experiments on three real-world weighted graph datasets. In particular, our model outperforms the best baseline, demonstrating its superior performance. Furthermore, experiments also show that the proposed graph contraction module can significantly reduce training time and storage space.

---

### Decentralized Multi-Agent Swarms for Autonomous Grid Security in Industrial IoT: A Consensus-based Approach

Authors: Samaresh Kumar Singh, Joyjit Roy, Chirag Agrawal
Published: 2026-01-24
Categories: cs.LG, cs.AI, cs.DC, cs.ET
arXiv: https://arxiv.org/abs/2601.17303v2

As Industrial Internet of Things (IIoT) environments scale to tens of thousands of connected devices, centralized security architectures introduce latency bottlenecks that sophisticated attackers can exploit to compromise an entire manufacturing ecosystem. We present a Decentralized Multi-Agent Swarm (DMAS) architecture that deploys autonomous agents at each edge gateway, forming a distributed defense layer for IIoT networks. Rather than relying on static firewalls or cloud-forwarded telemetry, DMAS agents coordinate through a lightweight peer-to-peer protocol, detecting threats locally without cloud dependency. We describe a Consensus-based Threat Validation (CVT) protocol in which agents collectively vote on detected threats, enabling near-instant quarantine of compromised nodes. Experiments on a 2000-device hardware testbed show that DMAS achieves sub-millisecond response times (0.85 ms average), 97.3% detection accuracy under high load, and 87% accuracy on zeroday attacks, each exceeding both centralized and edge-computing baselines. Bandwidth consumption drops by 89% relative to cloud-based solutions.

---

### Tabular Foundation Models are Strong Graph Anomaly Detectors

Authors: Yunhui Liu, Tieke He, Yongchao Liu, Can Yi, Hong Jin, Chuntao Hong
Published: 2026-01-24
Categories: cs.LG
arXiv: https://arxiv.org/abs/2601.17301v1

Graph anomaly detection (GAD), which aims to identify abnormal nodes that deviate from the majority, has become increasingly important in high-stakes Web domains. However, existing GAD methods follow a "one model per dataset" paradigm, leading to high computational costs, substantial data demands, and poor generalization when transferred to new datasets. This calls for a foundation model that enables a "one-for-all" GAD solution capable of detecting anomalies across diverse graphs without retraining. Yet, achieving this is challenging due to the large structural and feature heterogeneity across domains. In this paper, we propose TFM4GAD, a simple yet effective framework that adapts tabular foundation models (TFMs) for graph anomaly detection. Our key insight is that the core challenges of foundation GAD, handling heterogeneous features, generalizing across domains, and operating with scarce labels, are the exact problems that modern TFMs are designed to solve via synthetic pre-training and powerful in-context learning. The primary challenge thus becomes structural: TFMs are agnostic to graph topology. TFM4GAD bridges this gap by "flattening" the graph, constructing an augmented feature table that enriches raw node features with Laplacian embeddings, local and global structural characteristics, and anomaly-sensitive neighborhood aggregations. This augmented table is processed by a TFM in a fully in-context regime. Extensive experiments on multiple datasets with various TFM backbones reveal that TFM4GAD surprisingly achieves significant performance gains over specialized GAD models trained from scratch. Our work offers a new perspective and a practical paradigm for leveraging TFMs as powerful, generalist graph anomaly detectors.

---

### Structure-Aware NL-to-SQL for SFC Provisioning via AST-Masking Empowered Language Models

Authors: Xinyu Zhu, Parisa Fard Moshiri, Poonam Lohan, Burak Kantarci, Emil Janulewicz
Published: 2026-01-24
Categories: cs.NI, cs.CL, cs.LG
arXiv: https://arxiv.org/abs/2601.17295v1

Effective Service Function Chain (SFC) provisioning requires precise orchestration in dynamic and latency-sensitive networks. Reinforcement Learning (RL) improves adaptability but often ignores structured domain knowledge, which limits generalization and interpretability. Large Language Models (LLMs) address this gap by translating natural language (NL) specifications into executable Structured Query Language (SQL) commands for specification-driven SFC management. Conventional fine-tuning, however, can cause syntactic inconsistencies and produce inefficient queries. To overcome this, we introduce Abstract Syntax Tree (AST)-Masking, a structure-aware fine-tuning method that uses SQL ASTs to assign weights to key components and enforce syntax-aware learning without adding inference overhead. Experiments show that AST-Masking significantly improves SQL generation accuracy across multiple language models. FLAN-T5 reaches an Execution Accuracy (EA) of 99.6%, while Gemma achieves the largest absolute gain from 7.5% to 72.0%. These results confirm the effectiveness of structure-aware fine-tuning in ensuring syntactically correct and efficient SQL generation for interpretable SFC orchestration.

---

### Dynamic Meta-Ensemble Framework for Efficient and Accurate Deep Learning in Plant Leaf Disease Detection on Resource-Constrained Edge Devices

Authors: Weloday Fikadu Moges, Jianmei Su, Amin Waqas
Published: 2026-01-24
Categories: cs.CV, cs.AI
arXiv: https://arxiv.org/abs/2601.17290v1

Deploying deep learning models for plant disease detection on edge devices such as IoT sensors, smartphones, and embedded systems is severely constrained by limited computational resources and energy budgets. To address this challenge, we introduce a novel Dynamic Meta-Ensemble Framework (DMEF) for high-accuracy plant disease diagnosis under resource constraints. DMEF employs an adaptive weighting mechanism that dynamically combines the predictions of three lightweight convolutional neural networks (MobileNetV2, NASNetMobile, and InceptionV3) by optimizing a trade-off between accuracy improvements (DeltaAcc) and computational efficiency (model size). During training, the ensemble weights are updated iteratively, favoring models exhibiting high performance and low complexity. Extensive experiments on benchmark datasets for potato and maize diseases demonstrate state-of-the-art classification accuracies of 99.53% and 96.61%, respectively, surpassing standalone models and static ensembles by 2.1% and 6.3%. With computationally efficient inference latency (<75ms) and a compact footprint (<1 million parameters), DMEF shows strong potential for edge-based agricultural monitoring, suggesting viability for scalable crop disease management. This bridges the gap between high-accuracy AI and practical field applications.

---

### Fluxamba: Topology-Aware Anisotropic State Space Models for Geological Lineament Segmentation in Multi-Source Remote Sensing

Authors: Jin Bai, Huiyao Zhang, Qi Wen, Shengyang Li, Xiaolin Tian, Atta ur Rahman
Published: 2026-01-24
Categories: cs.CV
arXiv: https://arxiv.org/abs/2601.17288v1

The precise segmentation of geological linear features, spanning from planetary lineaments to terrestrial fractures, demands capturing long-range dependencies across complex anisotropic topologies. Although State Space Models (SSMs) offer near-linear computational complexity, their dependence on rigid, axis-aligned scanning trajectories induces a fundamental topological mismatch with curvilinear targets, resulting in fragmented context and feature erosion. To bridge this gap, we propose Fluxamba, a lightweight architecture that introduces a topology-aware feature rectification framework. Central to our design is the Structural Flux Block (SFB), which orchestrates an anisotropic information flux by integrating an Anisotropic Structural Gate (ASG) with a Prior-Modulated Flow (PMF). This mechanism decouples feature orientation from spatial location, dynamically gating context aggregation along the target's intrinsic geometry rather than rigid paths. Furthermore, to mitigate serialization-induced noise in low-contrast environments, we incorporate a Hierarchical Spatial Regulator (HSR) for multi-scale semantic alignment and a High-Fidelity Focus Unit (HFFU) to explicitly maximize the signal-to-noise ratio of faint features. Extensive experiments on diverse geological benchmarks (LROC-Lineament, LineaMapper, and GeoCrack) demonstrate that Fluxamba establishes a new state-of-the-art. Notably, on the challenging LROC-Lineament dataset, it achieves an F1-score of 89.22% and mIoU of 89.87%. Achieving a real-time inference speed of over 24 FPS with only 3.4M parameters and 6.3G FLOPs, Fluxamba reduces computational costs by up to two orders of magnitude compared to heavy-weight baselines, thereby establishing a new Pareto frontier between segmentation fidelity and onboard deployment feasibility.

---

### Mind the Ambiguity: Aleatoric Uncertainty Quantification in LLMs for Safe Medical Question Answering

Authors: Yaokun Liu, Yifan Liu, Phoebe Mbuvi, Zelin Li, Ruichen Yao, Gawon Lim, Dong Wang
Published: 2026-01-24
Categories: cs.CL, cs.AI
arXiv: https://arxiv.org/abs/2601.17284v1

The deployment of Large Language Models in Medical Question Answering is severely hampered by ambiguous user queries, a significant safety risk that demonstrably reduces answer accuracy in high-stakes healthcare settings. In this paper, we formalize this challenge by linking input ambiguity to aleatoric uncertainty (AU), which is the irreducible uncertainty arising from underspecified input. To facilitate research in this direction, we construct CV-MedBench, the first benchmark designed for studying input ambiguity in Medical QA. Using this benchmark, we analyze AU from a representation engineering perspective, revealing that AU is linearly encoded in LLM's internal activation patterns. Leveraging this insight, we introduce a novel AU-guided "Clarify-Before-Answer" framework, which incorporates AU-Probe - a lightweight module that detects input ambiguity directly from hidden states. Unlike existing uncertainty estimation methods, AU-Probe requires neither LLM fine-tuning nor multiple forward passes, enabling an efficient mechanism to proactively request user clarification and significantly enhance safety. Extensive experiments across four open LLMs demonstrate the effectiveness of our QA framework, with an average accuracy improvement of 9.48% over baselines. Our framework provides an efficient and robust solution for safe Medical QA, strengthening the reliability of health-related applications. The code is available at https://github.com/yaokunliu/AU-Med.git, and the CV-MedBench dataset is released on Hugging Face at https://huggingface.co/datasets/yaokunl/CV-MedBench.

---

### On the Insecurity of Keystroke-Based AI Authorship Detection: Timing-Forgery Attacks Against Motor-Signal Verification

Authors: David Condrey
Published: 2026-01-24
Categories: cs.CR, cs.AI, cs.HC
arXiv: https://arxiv.org/abs/2601.17280v1

Recent proposals advocate using keystroke timing signals, specifically the coefficient of variation ($δ$) of inter-keystroke intervals, to distinguish human-composed text from AI-generated content. We demonstrate that this class of defenses is insecure against two practical attack classes: the copy-type attack, in which a human transcribes LLM-generated text producing authentic motor signals, and timing-forgery attacks, in which automated agents sample inter-keystroke intervals from empirical human distributions. Using 13,000 sessions from the SBU corpus and three timing-forgery variants (histogram sampling, statistical impersonation, and generative LSTM), we show all attacks achieve $\ge$99.8% evasion rates against five classifiers. While detectors achieve AUC=1.000 against fully-automated injection, they classify $\ge$99.8% of attack samples as human with mean confidence $\ge$0.993. We formalize a non-identifiability result: when the detector observes only timing, the mutual information between features and content provenance is zero for copy-type attacks. Although composition and transcription produce statistically distinguishable motor patterns (Cohen's d=1.28), both yield $δ$ values 2-4x above detection thresholds, rendering the distinction security-irrelevant. These systems confirm a human operated the keyboard, but not whether that human originated the text. Securing provenance requires architectures that bind the writing process to semantic content.

---

### SPADE: A SIMD Posit-enabled compute engine for Accelerating DNN Efficiency

Authors: Sonu Kumar, Lavanya Vinnakota, Mukul Lokhande, Santosh Kumar Vishvakarma, Adam Teman
Published: 2026-01-24
Categories: cs.AR, cs.CV, eess.IV
arXiv: https://arxiv.org/abs/2601.17279v1

The growing demand for edge-AI systems requires arithmetic units that balance numerical precision, energy efficiency, and compact hardware while supporting diverse formats. Posit arithmetic offers advantages over floating- and fixed-point representations through its tapered precision, wide dynamic range, and improved numerical robustness. This work presents SPADE, a unified multi-precision SIMD Posit-based multiplyaccumulate (MAC) architecture supporting Posit (8,0), Posit (16,1), and Posit (32,2) within a single framework. Unlike prior single-precision or floating/fixed-point SIMD MACs, SPADE introduces a regime-aware, lane-fused SIMD Posit datapath that hierarchically reuses Posit-specific submodules (LOD, complementor, shifter, and multiplier) across 8/16/32-bit precisions without datapath replication. FPGA implementation on a Xilinx Virtex-7 shows 45.13% LUT and 80% slice reduction for Posit (8,0), and up to 28.44% and 17.47% improvement for Posit (16,1) and Posit (32,2) over prior work, with only 6.9% LUT and 14.9% register overhead for multi-precision support. ASIC results across TSMC nodes achieve 1.38 GHz at 6.1 mW (28 nm). Evaluation on MNIST, CIFAR-10/100, and alphabet datasets confirms competitive inference accuracy.

---

### PingPong: A Natural Benchmark for Multi-Turn Code-Switching Dialogues

Authors: Mohammad Rifqi Farhansyah, Hanif Muhammad Zhafran, Farid Adilazuarda, Shamsuddeen Hassan Muhammad, Maryam Ibrahim Mukhtar, Nedjma Ousidhoum, Genta Indra Winata, Ayu Purwarianti, Alham Fikri Aji
Published: 2026-01-24
Categories: cs.CL
arXiv: https://arxiv.org/abs/2601.17277v1

Code-switching is a widespread practice among the world's multilingual majority, yet few benchmarks accurately reflect its complexity in everyday communication. We present PingPong, a benchmark for natural multi-party code-switching dialogues covering five language-combination variations, some of which are trilingual. Our dataset consists of human-authored conversations among 2 to 4 participants covering authentic, multi-threaded structures where replies frequently reference much earlier points in the dialogue. We demonstrate that our data is significantly more natural and structurally diverse than machine-generated alternatives, offering greater variation in message length, speaker dominance, and reply distance. Based on these dialogues, we define three downstream tasks: Question Answering, Dialogue Summarization, and Topic Classification. Evaluations of several state-of-the-art language models on PingPong reveal that performance remains limited on code-switched inputs, underscoring the urgent need for more robust NLP systems capable of addressing the intricacies of real-world multilingual discourse.

---

### Latent-Space Contrastive Reinforcement Learning for Stable and Efficient LLM Reasoning

Authors: Lianlei Shan, Han Chen, Yixuan Wang, Zhenjie Liu, Wei Li
Published: 2026-01-24
Categories: cs.LG, cs.AI
arXiv: https://arxiv.org/abs/2601.17275v1

While Large Language Models (LLMs) demonstrate exceptional performance in surface-level text generation, their nature in handling complex multi-step reasoning tasks often remains one of ``statistical fitting'' rather than systematic logical deduction. Traditional Reinforcement Learning (RL) attempts to mitigate this by introducing a ``think-before-speak'' paradigm. However, applying RL directly in high-dimensional, discrete token spaces faces three inherent challenges: sample-inefficient rollouts, high gradient estimation variance, and the risk of catastrophic forgetting. To fundamentally address these structural bottlenecks, we propose \textbf{DeepLatent Reasoning (DLR)}, a latent-space bidirectional contrastive reinforcement learning framework. This framework shifts the trial-and-error cost from expensive token-level full sequence generation to the continuous latent manifold. Specifically, we introduce a lightweight assistant model to efficiently sample $K$ reasoning chain encodings within the latent space. These encodings are filtered via a dual reward mechanism based on correctness and formatting; only high-value latent trajectories are fed into a \textbf{frozen main model} for single-pass decoding. To maximize reasoning diversity while maintaining coherence, we design a contrastive learning objective to enable directed exploration within the latent space. Since the main model parameters remain frozen during optimization, this method mathematically eliminates catastrophic forgetting. Experiments demonstrate that under comparable GPU computational budgets, DLR achieves more stable training convergence, supports longer-horizon reasoning chains, and facilitates the sustainable accumulation of reasoning capabilities, providing a viable path toward reliable and scalable reinforcement learning for LLMs.

---

### Unrolled Neural Networks for Constrained Optimization

Authors: Samar Hadou, Alejandro Ribeiro
Published: 2026-01-24
Categories: cs.LG
arXiv: https://arxiv.org/abs/2601.17274v1

In this paper, we develop unrolled neural networks to solve constrained optimization problems, offering accelerated, learnable counterparts to dual ascent (DA) algorithms. Our framework, termed constrained dual unrolling (CDU), comprises two coupled neural networks that jointly approximate the saddle point of the Lagrangian. The primal network emulates an iterative optimizer that finds a stationary point of the Lagrangian for a given dual multiplier, sampled from an unknown distribution. The dual network generates trajectories towards the optimal multipliers across its layers while querying the primal network at each layer. Departing from standard unrolling, we induce DA dynamics by imposing primal-descent and dual-ascent constraints through constrained learning. We formulate training the two networks as a nested optimization problem and propose an alternating procedure that updates the primal and dual networks in turn, mitigating uncertainty in the multiplier distribution required for primal network training. We numerically evaluate the framework on mixed-integer quadratic programs (MIQPs) and power allocation in wireless networks. In both cases, our approach yields near-optimal near-feasible solutions and exhibits strong out-of-distribution (OOD) generalization.

---

### Cross360: 360° Monocular Depth Estimation via Cross Projections Across Scales

Authors: Kun Huang, Fang-Lue Zhang, Neil Dodgson
Published: 2026-01-24
Categories: cs.CV
arXiv: https://arxiv.org/abs/2601.17271v1

360° depth estimation is a challenging research problem due to the difficulty of finding a representation that both preserves global continuity and avoids distortion in spherical images. Existing methods attempt to leverage complementary information from multiple projections, but struggle with balancing global and local consistency. Their local patch features have limited global perception, and the combined global representation does not address discrepancies in feature extraction at the boundaries between patches. To address these issues, we propose Cross360, a novel cross-attention-based architecture integrating local and global information using less-distorted tangent patches along with equirectangular features. Our Cross Projection Feature Alignment module employs cross-attention to align local tangent projection features with the equirectangular projection's 360° field of view, ensuring each tangent projection patch is aware of the global context. Additionally, our Progressive Feature Aggregation with Attention module refines multi-scaled features progressively, enhancing depth estimation accuracy. Cross360 significantly outperforms existing methods across most benchmark datasets, especially those in which the entire 360° image is available, demonstrating its effectiveness in accurate and globally consistent depth estimation. The code and model are available at https://github.com/huangkun101230/Cross360.

---

### Window Size Versus Accuracy Experiments in Voice Activity Detectors

Authors: Max McKinnon, Samir Khaki, Chandan KA Reddy, William Huang
Published: 2026-01-24
Categories: cs.SD, cs.CL, eess.AS
arXiv: https://arxiv.org/abs/2601.17270v1

Voice activity detection (VAD) plays a vital role in enabling applications such as speech recognition. We analyze the impact of window size on the accuracy of three VAD algorithms: Silero, WebRTC, and Root Mean Square (RMS) across a set of diverse real-world digital audio streams. We additionally explore the use of hysteresis on top of each VAD output. Our results offer practical references for optimizing VAD systems. Silero significantly outperforms WebRTC and RMS, and hysteresis provides a benefit for WebRTC.

---

### Covariate-assisted Grade of Membership Models via Shared Latent Geometry

Authors: Zhiyu Xu, Yuqi Gu
Published: 2026-01-24
Categories: stat.ME, stat.ML
arXiv: https://arxiv.org/abs/2601.17265v1

The grade of membership model is a flexible latent variable model for analyzing multivariate categorical data through individual-level mixed membership scores. In many modern applications, auxiliary covariates are collected alongside responses and encode information about the same latent structure. Traditional approaches to incorporating such covariates typically rely on fully specified joint likelihoods, which are computationally intensive and sensitive to misspecification. We introduce a covariate-assisted grade of membership model that integrates response and covariate information by exploiting their shared low-rank simplex geometry, rather than modeling their joint distribution. We propose a likelihood-free spectral estimation procedure that combines heterogeneous data sources through a balance parameter controlling their relative contribution. To accommodate high-dimensional and heteroskedastic noise, we employ heteroskedastic principal component analysis before performing simplex-based geometric recovery. Our theoretical analysis establishes weaker identifiability conditions than those required in the covariate-free model, and further derives finite-sample, entrywise error bounds for both mixed membership scores and item parameters. These results demonstrate that auxiliary covariates can provably improve latent structure recovery, yielding faster convergence rates in high-dimensional regimes. Simulation studies and an application to educational assessment data illustrate the computational efficiency, statistical accuracy, and interpretability gains of the proposed method. The code for reproducing these results is open-source and available at \texttt{https://github.com/Toby-X/Covariate-Assisted-GoM}

---

### AGZO: Activation-Guided Zeroth-Order Optimization for LLM Fine-Tuning

Authors: Wei Lin, Yining Jiang, Qingyu Song, Qiao Xiang, Hong Xu
Published: 2026-01-24
Categories: cs.LG
arXiv: https://arxiv.org/abs/2601.17261v4

Zeroth-Order (ZO) optimization has emerged as a promising solution for fine-tuning LLMs under strict memory constraints, as it avoids the prohibitive memory cost of storing activations for backpropagation. However, existing ZO methods typically employ isotropic perturbations, neglecting the rich structural information available during the forward pass. In this paper, we identify a crucial link between gradient formation and activation structure: the gradient of a linear layer is confined to the subspace spanned by its input activations. Leveraging this insight, we propose Activation-Guided Zeroth-Order optimization (AGZO). Unlike prior methods, AGZO extracts a compact, activation-informed subspace on the fly during the forward pass and restricts perturbations to this low-rank subspace. We provide a theoretical framework showing that AGZO optimizes a subspace-smoothed objective and provably yields update directions with higher cosine similarity to the true gradient than isotropic baselines. Empirically, we evaluate AGZO on Qwen3 and Pangu models across various benchmarks. AGZO consistently outperforms state-of-the-art ZO baselines and significantly narrows the performance gap with first-order fine-tuning, while maintaining almost the same peak memory footprint as other ZO methods.

---

### The Viscosity of Logic: Phase Transitions and Hysteresis in DPO Alignment

Authors: Marco Pollanen
Published: 2026-01-24
Categories: cs.LG, cs.AI
arXiv: https://arxiv.org/abs/2601.17260v1

Direct Preference Optimization (DPO) is often tuned as if increasing alignment pressure (controlled by $β$) yields progressively "better" behavior. We instead treat $β$ as a control parameter and densely sweep it for three 7B open-weight families under a fixed DPO recipe. In Mistral, capability is sharply non-monotonic: aggregated logic-probe margins become positive only in a narrow band near $β\approx 10^{-2}$ and revert outside it, with boundary points that are seed-sensitive. Across architectures under the same sweep, we observe qualitatively different response modes: sharp reorganization in Mistral, selective changes in Llama, and smooth trade-offs in Qwen. Critically, the DPO preference margin can anticorrelate with reasoning capability (Pearson $r=-0.91$ for Llama logic), so margin-based selection can prefer capability-impaired models. Training path also matters: exposure to high $β$ induces capability losses that persist even after $β$ is reduced (hysteresis). These findings motivate capability-resolved evaluation across the $β$ landscape rather than reliance on margins or aggregate benchmarks.

---

### Inference-Time Loss-Guided Colour Preservation in Diffusion Sampling

Authors: Angad Singh Ahuja, Aarush Ram Anandh
Published: 2026-01-24
Categories: cs.CV, cs.GR, cs.LG
arXiv: https://arxiv.org/abs/2601.17259v1

Precise color control remains a persistent failure mode in text-to-image diffusion systems, particularly in design-oriented workflows where outputs must satisfy explicit, user-specified color targets. We present an inference-time, region-constrained color preservation method that steers a pretrained diffusion model without any additional training. Our approach combines (i) ROI-based inpainting for spatial selectivity, (ii) background-latent re-imposition to prevent color drift outside the ROI, and (iii) latent nudging via gradient guidance using a composite loss defined in CIE Lab and linear RGB. The loss is constructed to control not only the mean ROI color but also the tail of the pixelwise error distribution through CVaR-style and soft-maximum penalties, with a late-start gate and a time-dependent schedule to stabilize guidance across denoising steps. We show that mean-only baselines can satisfy average color constraints while producing perceptually salient local failures, motivating our distribution-aware objective. The resulting method provides a practical, training-free mechanism for targeted color adherence that can be integrated into standard Stable Diffusion inpainting pipelines.

---

### FineVAU: A Novel Human-Aligned Benchmark for Fine-Grained Video Anomaly Understanding

Authors: João Pereira, Vasco Lopes, João Neves, David Semedo
Published: 2026-01-24
Categories: cs.CV
arXiv: https://arxiv.org/abs/2601.17258v2

Video Anomaly Understanding (VAU) is a novel task focused on describing unusual occurrences in videos. Despite growing interest, the evaluation of VAU remains an open challenge. Existing benchmarks rely on n-gram-based metrics (e.g., BLEU, ROUGE-L) or LLM-based evaluation. The first fails to capture the rich, free-form, and visually grounded nature of LVLM responses, while the latter focuses on assessing language quality over factual relevance, often resulting in subjective judgments that are misaligned with human perception. In this work, we address this issue by proposing FineVAU, a new benchmark for VAU that shifts the focus towards rich, fine-grained and domain-specific understanding of anomalous videos. We formulate VAU as a three-fold problem, with the goal of comprehensively understanding key descriptive elements of anomalies in video: events (What), participating entities (Who) and location (Where). Our benchmark introduces a) FVScore, a novel, human-aligned evaluation metric that assesses the presence of critical visual elements in LVLM answers, providing interpretable, fine-grained feedback; and b) FineW3, a novel, comprehensive dataset curated through a structured and fully automatic procedure that augments existing human annotations with high quality, fine-grained visual information. Human evaluation reveals that our proposed metric has a superior alignment with human perception of anomalies in comparison to current approaches. Detailed experiments on FineVAU unveil critical limitations in LVLM's ability to perceive anomalous events that require spatial and fine-grained temporal understanding, despite strong performance on coarse grain, static information, and events with strong visual cues.

---

### A Constrained Optimization Perspective of Unrolled Transformers

Authors: Javier Porras-Valenzuela, Samar Hadou, Alejandro Ribeiro
Published: 2026-01-24
Categories: cs.LG
arXiv: https://arxiv.org/abs/2601.17257v2

We introduce a constrained optimization framework for training transformers that behave like optimization descent algorithms. Specifically, we enforce layerwise descent constraints on the objective function and replace standard empirical risk minimization (ERM) with a primal-dual training scheme. This approach yields models whose intermediate representations decrease the loss monotonically in expectation across layers. We apply our method to both unrolled transformer architectures and conventional pretrained transformers on tasks of video denoising and text classification. Across these settings, we observe constrained transformers achieve stronger robustness to perturbations and maintain higher out-of-distribution generalization, while preserving in-distribution performance.

---

### Multi-stage Bridge Inspection System: Integrating Foundation Models with Location Anonymization

Authors: Takato Yasuno
Published: 2026-01-24
Categories: cs.CV
arXiv: https://arxiv.org/abs/2601.17254v1

In Japan, civil infrastructure condition monitoring is mandated through visual inspection every five years. Field-captured damage images frequently contain concrete cracks and rebar exposure, often accompanied by construction signs revealing regional information. To enable safe infrastructure use without causing public anxiety, it is essential to protect regional information while accurately extracting damage features and visualizing key indicators for repair decision-making. This paper presents an open-source bridge damage detection system with regional privacy protection capabilities. We employ Segment Anything Model (SAM) 3 for rebar corrosion detection and utilize DBSCAN for automatic completion of missed regions. Construction sign regions are detected and protected through Gaussian blur. Four preprocessing methods improve OCR accuracy, and GPU optimization enables 1.7-second processing per image. The technology stack includes SAM3, PyTorch, OpenCV, pytesseract, and scikit-learn, achieving efficient bridge inspection with regional information protection.

---

### C-RADIOv4 (Tech Report)

Authors: Mike Ranzinger, Greg Heinrich, Collin McCarthy, Jan Kautz, Andrew Tao, Bryan Catanzaro, Pavlo Molchanov
Published: 2026-01-24
Categories: cs.CV
arXiv: https://arxiv.org/abs/2601.17237v1

By leveraging multi-teacher distillation, agglomerative vision backbones provide a unified student model that retains and improves the distinct capabilities of multiple teachers. In this tech report, we describe the most recent release of the C-RADIO family of models, C-RADIOv4, which builds upon AM-RADIO/RADIOv2.5 in design, offering strong improvements on key downstream tasks at the same computational complexity. We release -SO400M (412M params), and -H (631M) model variants, both trained with an updated set of teachers: SigLIP2, DINOv3, and SAM3. In addition to improvements on core metrics and new capabilities from imitating SAM3, the C-RADIOv4 model family further improves any-resolution support, brings back the ViTDet option for drastically enhanced efficiency at high-resolution, and comes with a permissive license.

---

### Prime Successor Irreducibility: Turing Machine Complexity, Kolmogorov Complexity, and Weakness-Based Formulations

Authors: Ben Goertzel, Bill Lauritzen
Published: 2026-01-23
Categories: cs.CC, cs.AI
arXiv: https://arxiv.org/abs/2605.12504v1

We develop conjectures and theorems expressing the idea that the prime sequence exhibits computational irreducibility in the transition from one prime to its successor. Informally, given a prime pp p, no general algorithm can compute the least prime greater than pp p substantially faster than sequentially testing candidates for primality, except possibly on sparse input sets. Our framework proceeds along complementary lines. First, we formalize Prime Successor Irreducibility in a Turing-machine complexity model (PSI-T), asserting lower bounds on running time relative to a sequential baseline. Second, we propose a Kolmogorov-complexity formulation (PSI-K), asserting that typical prime gaps are algorithmically incompressible at their scale; we prove PSI-K(c, $δ$) unconditionally for all fixed c<1 using standard sieve bounds. Third, we develop weakness-based formulations: PSI-W (sparse-set anti-concentration) shows no small menu of gap values captures a noticeable fraction of primes, while PSI-W-LE shows collision probabilities decay and logical entropy tends to 1. These extend to prime constellations and consecutive gap vectors. Finally, a sieve-theoretic framework connects local obstruction patterns to Selberg weakness parameters. The PSI-K and weakness formulations connect irreducibility to classical statistical questions about prime gaps. Using the relationship between Kolmogorov complexity and Shannon entropy, we derive rigorous lower bounds on prime gap entropy in dyadic intervals (X,2X). Together, these formulations provide a unified complexity-theoretic perspective on the apparent local unpredictability of the prime sequence, without asserting randomness or independence.

---