# Research Papers — 2026-01 · bundle page 125 of 189

> Full-text markdown of papers #6201–6250 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/125
[← Previous bundle](https://aipod.dev/papers/archive/2026-01/page/124.md)
[Next bundle →](https://aipod.dev/papers/archive/2026-01/page/126.md)

### How vehicles change lanes after encountering crashes: Empirical analysis and modeling

Authors: Kequan Chen, Yuxuan Wang, Pan Liu, Victor L. Knoop, David Z. W. Wang, Yu Han
Published: 2026-01-13
Categories: cs.AI
arXiv: https://arxiv.org/abs/2601.08125v1

When a traffic crash occurs, following vehicles need to change lanes to bypass the obstruction. We define these maneuvers as post crash lane changes. In such scenarios, vehicles in the target lane may refuse to yield even after the lane change has already begun, increasing the complexity and crash risk of post crash LCs. However, the behavioral characteristics and motion patterns of post crash LCs remain unknown. To address this gap, we construct a post crash LC dataset by extracting vehicle trajectories from drone videos captured after crashes. Our empirical analysis reveals that, compared to mandatory LCs (MLCs) and discretionary LCs (DLCs), post crash LCs exhibit longer durations, lower insertion speeds, and higher crash risks. Notably, 79.4% of post crash LCs involve at least one instance of non yielding behavior from the new follower, compared to 21.7% for DLCs and 28.6% for MLCs. Building on these findings, we develop a novel trajectory prediction framework for post crash LCs. At its core is a graph based attention module that explicitly models yielding behavior as an auxiliary interaction aware task. This module is designed to guide both a conditional variational autoencoder and a Transformer based decoder to predict the lane changer's trajectory. By incorporating the interaction aware module, our model outperforms existing baselines in trajectory prediction performance by more than 10% in both average displacement error and final displacement error across different prediction horizons. Moreover, our model provides more reliable crash risk analysis by reducing false crash rates and improving conflict prediction accuracy. Finally, we validate the model's transferability using additional post crash LC datasets collected from different sites.

---

### Generalization Analysis and Method for Domain Generalization for a Family of Recurrent Neural Networks

Authors: Atefeh Termehchi, Ekram Hossain, Isaac Woungang
Published: 2026-01-13
Categories: cs.LG
arXiv: https://arxiv.org/abs/2601.08122v1

Deep learning (DL) has driven broad advances across scientific and engineering domains. Despite its success, DL models often exhibit limited interpretability and generalization, which can undermine trust, especially in safety-critical deployments. As a result, there is growing interest in (i) analyzing interpretability and generalization and (ii) developing models that perform robustly under data distributions different from those seen during training (i.e. domain generalization). However, the theoretical analysis of DL remains incomplete. For example, many generalization analyses assume independent samples, which is violated in sequential data with temporal correlations. Motivated by these limitations, this paper proposes a method to analyze interpretability and out-of-domain (OOD) generalization for a family of recurrent neural networks (RNNs). Specifically, the evolution of a trained RNN's states is modeled as an unknown, discrete-time, nonlinear closed-loop feedback system. Using Koopman operator theory, these nonlinear dynamics are approximated with a linear operator, enabling interpretability. Spectral analysis is then used to quantify the worst-case impact of domain shifts on the generalization error. Building on this analysis, a domain generalization method is proposed that reduces the OOD generalization error and improves the robustness to distribution shifts. Finally, the proposed analysis and domain generalization approach are validated on practical temporal pattern-learning tasks.

---

### Intra-tree Column Subsampling Hinders XGBoost Learning of Ratio-like Interactions

Authors: Mykola Pinchuk
Published: 2026-01-13
Categories: cs.LG
arXiv: https://arxiv.org/abs/2601.08121v1

Many applied problems contain signal that becomes clear only after combining multiple raw measurements. Ratios and rates are common examples. In gradient boosted trees, this combination is not an explicit operation: the model must synthesize it through coordinated splits on the component features. We study whether intra-tree column subsampling in XGBoost makes that synthesis harder. We use two synthetic data generating processes with cancellation-style structure. In both, two primitive features share a strong nuisance factor, while the target depends on a smaller differential factor. A log ratio cancels the nuisance and isolates the signal. We vary colsample_bylevel and colsample_bynode over s in {0.4, 0.6, 0.8, 0.9}, emphasizing mild subsampling (s >= 0.8). A control feature set includes the engineered ratio, removing the need for synthesis. Across both processes, intra-tree column subsampling reduces test PR-AUC in the primitives-only setting. In the main process the relative decrease reaches 54 percent when both parameters are set to 0.4. The effect largely disappears when the engineered ratio is present. A path-based co-usage metric drops in the same cells where performance deteriorates. Practically, if ratio-like structure is plausible, either avoid intra-tree subsampling or include the intended ratio features.

---

### Structure Detection for Contextual Reinforcement Learning

Authors: Tianyue Zhou, Jung-Hoon Cho, Cathy Wu
Published: 2026-01-13
Categories: cs.LG
arXiv: https://arxiv.org/abs/2601.08120v1

Contextual Reinforcement Learning (CRL) tackles the problem of solving a set of related Contextual Markov Decision Processes (CMDPs) that vary across different context variables. Traditional approaches--independent training and multi-task learning--struggle with either excessive computational costs or negative transfer. A recently proposed multi-policy approach, Model-Based Transfer Learning (MBTL), has demonstrated effectiveness by strategically selecting a few tasks to train and zero-shot transfer. However, CMDPs encompass a wide range of problems, exhibiting structural properties that vary from problem to problem. As such, different task selection strategies are suitable for different CMDPs. In this work, we introduce Structure Detection MBTL (SD-MBTL), a generic framework that dynamically identifies the underlying generalization structure of CMDP and selects an appropriate MBTL algorithm. For instance, we observe Mountain structure in which generalization performance degrades from the training performance of the target task as the context difference increases. We thus propose M/GP-MBTL, which detects the structure and adaptively switches between a Gaussian Process-based approach and a clustering-based approach. Extensive experiments on synthetic data and CRL benchmarks--covering continuous control, traffic control, and agricultural management--show that M/GP-MBTL surpasses the strongest prior method by 12.49% on the aggregated metric. These results highlight the promise of online structure detection for guiding source task selection in complex CRL environments.

---

### MirrorBench: A Benchmark to Evaluate Conversational User-Proxy Agents for Human-Likeness

Authors: Ashutosh Hathidara, Julien Yu, Vaishali Senthil, Sebastian Schreiber, Anil Babu Ankisettipalli
Published: 2026-01-13
Categories: cs.AI, cs.LG
arXiv: https://arxiv.org/abs/2601.08118v3

Large language models (LLMs) are increasingly used as human simulators, both for evaluating conversational systems and for generating fine-tuning data. However, naive "act-as-a-user" prompting often yields verbose, unrealistic utterances, motivating principled evaluation of *user proxy agents*. We present **MirrorBench**, a reproducible and extensible benchmarking framework that evaluates user proxies solely on their ability to produce human-like user utterances across diverse conversational regimes, explicitly decoupled from downstream task success. **MirrorBench** combines three lexical-diversity metrics (**MATTR**, **Yule's~$K$**, and **HD-D**) with three LLM-judge-based metrics (**GTEval**, **Pairwise Indistinguishability**, and **Rubric-and-Reason**), and contextualizes judge scores using Human-Human and Proxy-Proxy calibration controls. Across four public datasets, **MirrorBench** yields variance-aware comparisons and reveals systematic gaps between user proxies and real human users. The framework is open sourced at https://github.com/SAP/mirrorbench and includes a command-line interface for running and managing user-proxy benchmarking experiments.

---

### Learning a Stochastic Differential Equation Model of Tropical Cyclone Intensification from Reanalysis and Observational Data

Authors: Kenneth Gee, Sai Ravela
Published: 2026-01-13
Categories: cs.LG, math.DS, physics.ao-ph, stat.AP
arXiv: https://arxiv.org/abs/2601.08116v3

Tropical cyclones are among the most consequential weather hazards, yet estimates of their risk are limited by the relatively short historical record. To extend these records, researchers often generate large ensembles of synthetic storms using simplified models of cyclone intensification. Developing such models, however, has traditionally required substantial theoretical effort. Here we explore whether equation-discovery methods, a class of data-driven techniques designed to infer governing equations, can accelerate the process of developing simplified intensification models. Using observational storm data (IBTrACS) together with environmental conditions from reanalysis (ERA5), we learn a compact stochastic differential equation describing tropical cyclone intensity evolution. We focus on TCs because their dynamics are well studied and a hierarchy of reduced-order models exist, enabling direct comparison of the learned model to physically-derived counterparts. We find that the learned model simulates synthetic TCs whose intensification statistics and hazard estimates are consistent with observations and competitive with a leading physics-based TC intensification model. Our model also reproduces known nonlinear dynamical behavior of tropical cyclones, including as a saddle node bifurcation as inner core ventilation is increased. This result shows that equation-discovery approaches, when applied directly to storm intensity, can recover not only realistic statistics but also physically meaningful dynamical structure. These findings highlight the potential for data-driven methods to complement existing theory and reduced-order models in the study of extreme weather.

---

### CSQL: Mapping Documents into Causal Databases

Authors: Sridhar Mahadevan
Published: 2026-01-13
Categories: cs.DB, cs.AI, cs.IR, cs.LG
arXiv: https://arxiv.org/abs/2601.08109v1

We describe a novel system, CSQL, which automatically converts a collection of unstructured text documents into an SQL-queryable causal database (CDB). A CDB differs from a traditional DB: it is designed to answer "why'' questions via causal interventions and structured causal queries. CSQL builds on our earlier system, DEMOCRITUS, which converts documents into thousands of local causal models derived from causal discourse. Unlike RAG-based systems or knowledge-graph based approaches, CSQL supports causal analysis over document collections rather than purely associative retrieval. For example, given an article on the origins of human bipedal walking, CSQL enables queries such as: "What are the strongest causal influences on bipedalism?'' or "Which variables act as causal hubs with the largest downstream influence?'' Beyond single-document case studies, we show that CSQL can also ingest RAG/IE-compiled causal corpora at scale by compiling the Testing Causal Claims (TCC) dataset of economics papers into a causal database containing 265,656 claim instances spanning 45,319 papers, 44 years, and 1,575 reported method strings, thereby enabling corpus-level causal queries and longitudinal analyses in CSQL. Viewed abstractly, CSQL functions as a compiler from unstructured documents into a causal database equipped with a principled algebra of queries, and can be applied broadly across many domains ranging from business, humanities, and science.

---

### Debiasing Large Language Models via Adaptive Causal Prompting with Sketch-of-Thought

Authors: Bowen Li, Ziqi Xu, Jing Ren, Renqiang Luo, Xikun Zhang, Xiuzhen Zhang, Yongli Ren, Feng Xia
Published: 2026-01-13
Categories: cs.CL, cs.AI
arXiv: https://arxiv.org/abs/2601.08108v1

Despite notable advancements in prompting methods for Large Language Models (LLMs), such as Chain-of-Thought (CoT), existing strategies still suffer from excessive token usage and limited generalisability across diverse reasoning tasks. To address these limitations, we propose an Adaptive Causal Prompting with Sketch-of-Thought (ACPS) framework, which leverages structural causal models to infer the causal effect of a query on its answer and adaptively select an appropriate intervention (i.e., standard front-door and conditional front-door adjustments). This design enables generalisable causal reasoning across heterogeneous tasks without task-specific retraining. By replacing verbose CoT with concise Sketch-of-Thought, ACPS enables efficient reasoning that significantly reduces token usage and inference cost. Extensive experiments on multiple reasoning benchmarks and LLMs demonstrate that ACPS consistently outperforms existing prompting baselines in terms of accuracy, robustness, and computational efficiency.

---

### STO-RL: Offline RL under Sparse Rewards via LLM-Guided Subgoal Temporal Order

Authors: Chengyang Gu, Yuxin Pan, Hui Xiong, Yize Chen
Published: 2026-01-13
Categories: cs.LG, cs.AI, eess.SY
arXiv: https://arxiv.org/abs/2601.08107v1

Offline reinforcement learning (RL) enables policy learning from pre-collected datasets, avoiding costly and risky online interactions, but it often struggles with long-horizon tasks involving sparse rewards. Existing goal-conditioned and hierarchical offline RL methods decompose such tasks and generate intermediate rewards to mitigate limitations of traditional offline RL, but usually overlook temporal dependencies among subgoals and rely on imprecise reward shaping, leading to suboptimal policies. To address these issues, we propose STO-RL (Offline RL using LLM-Guided Subgoal Temporal Order), an offline RL framework that leverages large language models (LLMs) to generate temporally ordered subgoal sequences and corresponding state-to-subgoal-stage mappings. Using this temporal structure, STO-RL applies potential-based reward shaping to transform sparse terminal rewards into dense, temporally consistent signals, promoting subgoal progress while avoiding suboptimal solutions. The resulting augmented dataset with shaped rewards enables efficient offline training of high-performing policies. Evaluations on four discrete and continuous sparse-reward benchmarks demonstrate that STO-RL consistently outperforms state-of-the-art offline goal-conditioned and hierarchical RL baselines, achieving faster convergence, higher success rates, and shorter trajectories. Ablation studies further confirm STO-RL's robustness to imperfect or noisy LLM-generated subgoal sequences, demonstrating that LLM-guided subgoal temporal structures combined with theoretically grounded reward shaping provide a practical and scalable solution for long-horizon offline RL.

---

### Query Suggestion for Retrieval-Augmented Generation via Dynamic In-Context Learning

Authors: Fabian Spaeh, Tianyi Chen, Chen-Hao Chiang, Bin Shen
Published: 2026-01-13
Categories: cs.CL
arXiv: https://arxiv.org/abs/2601.08105v1

Retrieval-augmented generation with tool-calling agents (agentic RAG) has become increasingly powerful in understanding, processing, and responding to user queries. However, the scope of the grounding knowledge is limited and asking questions that exceed this scope may lead to issues like hallucination. While guardrail frameworks aim to block out-of-scope questions (Rodriguez et al., 2024), no research has investigated the question of suggesting answerable queries in order to complete the user interaction. In this paper, we initiate the study of query suggestion for agentic RAG. We consider the setting where user questions are not answerable, and the suggested queries should be similar to aid the user interaction. Such scenarios are frequent for tool-calling LLMs as communicating the restrictions of the tools or the underlying datasets to the user is difficult, and adding query suggestions enhances the interaction with the RAG agent. As opposed to traditional settings for query recommendations such as in search engines, ensuring that the suggested queries are answerable is a major challenge due to the RAG's multi-step workflow that demands a nuanced understanding of the RAG as a whole, which the executing LLM lacks. As such, we introduce robust dynamic few-shot learning which retrieves examples from relevant workflows. We show that our system can be self-learned, for instance on prior user queries, and is therefore easily applicable in practice. We evaluate our approach on three benchmark datasets based on two unlabeled question datasets collected from real-world user queries. Experiments on real-world datasets confirm that our method produces more relevant and answerable suggestions, outperforming few-shot and retrieval-only baselines, and thus enable safer, more effective user interaction with agentic RAG.

---

### High-Fidelity Modeling of Stochastic Chemical Dynamics on Complex Manifolds: A Multi-Scale SIREN-PINN Framework for the Curvature-Perturbed Ginzburg-Landau Equation

Authors: Julian Evan Chrisnanto, Salsabila Rahma Alia, Nurfauzi Fadillah, Yulison Herry Chrisnanto
Published: 2026-01-13
Categories: nlin.CD, cs.AI
arXiv: https://arxiv.org/abs/2601.08104v2

The accurate identification and control of spatiotemporal chaos in reaction-diffusion systems remains a grand challenge in chemical engineering, particularly when the underlying catalytic surface possesses complex, unknown topography. In the \textit{Defect Turbulence} regime, system dynamics are governed by topological phase singularities (spiral waves) whose motion couples to manifold curvature via geometric pinning. Conventional Physics-Informed Neural Networks (PINNs) using ReLU or Tanh activations suffer from fundamental \textit{spectral bias}, failing to resolve high-frequency gradients and causing amplitude collapse or phase drift. We propose a Multi-Scale SIREN-PINN architecture leveraging periodic sinusoidal activations with frequency-diverse initialization, embedding the appropriate inductive bias for wave-like physics directly into the network structure. This enables simultaneous resolution of macroscopic wave envelopes and microscopic defect cores. Validated on the complex Ginzburg-Landau equation evolving on latent Riemannian manifolds, our architecture achieves relative state prediction error $ε_{L_2} \approx 1.92 \times 10^{-2}$, outperforming standard baselines by an order of magnitude while preserving topological invariants ($|ΔN_{defects}| < 1$). We solve the ill-posed \textit{inverse pinning problem}, reconstructing hidden Gaussian curvature fields solely from partial observations of chaotic wave dynamics (Pearson correlation $ρ= 0.965$). Training dynamics reveal a distinctive Spectral Phase Transition at epoch $\sim 2,100$, where cooperative minimization of physics and geometry losses drives the solver to Pareto-optimal solutions. This work establishes a new paradigm for Geometric Catalyst Design, offering a mesh-free, data-driven tool for identifying surface heterogeneity and engineering passive control strategies in turbulent chemical reactors.

---

### Towards A Unified PAC-Bayesian Framework for Norm-based Generalization Bounds

Authors: Xinping Yi, Gaojie Jin, Xiaowei Huang, Shi Jin
Published: 2026-01-13
Categories: stat.ML, cs.IT, cs.LG
arXiv: https://arxiv.org/abs/2601.08100v1

Understanding the generalization behavior of deep neural networks remains a fundamental challenge in modern statistical learning theory. Among existing approaches, PAC-Bayesian norm-based bounds have demonstrated particular promise due to their data-dependent nature and their ability to capture algorithmic and geometric properties of learned models. However, most existing results rely on isotropic Gaussian posteriors, heavy use of spectral-norm concentration for weight perturbations, and largely architecture-agnostic analyses, which together limit both the tightness and practical relevance of the resulting bounds. To address these limitations, in this work, we propose a unified framework for PAC-Bayesian norm-based generalization by reformulating the derivation of generalization bounds as a stochastic optimization problem over anisotropic Gaussian posteriors. The key to our approach is a sensitivity matrix that quantifies the network outputs with respect to structured weight perturbations, enabling the explicit incorporation of heterogeneous parameter sensitivities and architectural structures. By imposing different structural assumptions on this sensitivity matrix, we derive a family of generalization bounds that recover several existing PAC-Bayesian results as special cases, while yielding bounds that are comparable to or tighter than state-of-the-art approaches. Such a unified framework provides a principled and flexible way for geometry-/structure-aware and interpretable generalization analysis in deep learning.

---

### AdaJudge: Adaptive Multi-Perspective Judging for Reward Modeling

Authors: Yongliang Miao, Yangyang Liang, Mengnan Du
Published: 2026-01-13
Categories: cs.CL, cs.LG
arXiv: https://arxiv.org/abs/2601.08097v2

Reward modeling is essential for aligning large language models with human preferences, yet predominant architectures rely on a static pooling strategy to condense sequences into scalar scores. This paradigm, however, suffers from two key limitations: a static inductive bias that misaligns with task-dependent preference signals, and a representational mismatch, as the backbone's optimization for generation leaves its representations ill-suited to fine-grained discrimination. To address this, we propose AdaJudge, a unified framework that jointly adapts representation and aggregation. AdaJudge first improves backbone representations into a discrimination-oriented space via gated refinement blocks. It then replaces the static readout with an adaptive multi-view pooling module, which dynamically routes and combines evidence. Extensive experiments on RM-Bench and JudgeBench show that AdaJudge outperforms strong off-the-shelf reward models and traditional pooling baselines.

---

### From Prompts to Deployment: Auto-Curated Domain-Specific Dataset Generation via Diffusion Models

Authors: Dongsik Yoon, Jongeun Kim
Published: 2026-01-13
Categories: cs.CV
arXiv: https://arxiv.org/abs/2601.08095v1

In this paper, we present an automated pipeline for generating domain-specific synthetic datasets with diffusion models, addressing the distribution shift between pre-trained models and real-world deployment environments. Our three-stage framework first synthesizes target objects within domain-specific backgrounds through controlled inpainting. The generated outputs are then validated via a multi-modal assessment that integrates object detection, aesthetic scoring, and vision-language alignment. Finally, a user-preference classifier is employed to capture subjective selection criteria. This pipeline enables the efficient construction of high-quality, deployable datasets while reducing reliance on extensive real-world data collection.

---

### Local-Global Feature Fusion for Subject-Independent EEG Emotion Recognition

Authors: Zheng Zhou, Isabella McEvoy, Camilo E. Valderrama
Published: 2026-01-13
Categories: cs.LG, cs.AI
arXiv: https://arxiv.org/abs/2601.08094v1

Subject-independent EEG emotion recognition is challenged by pronounced inter-subject variability and the difficulty of learning robust representations from short, noisy recordings. To address this, we propose a fusion framework that integrates (i) local, channel-wise descriptors and (ii) global, trial-level descriptors, improving cross-subject generalization on the SEED-VII dataset. Local representations are formed per channel by concatenating differential entropy with graph-theoretic features, while global representations summarize time-domain, spectral, and complexity characteristics at the trial level. These representations are fused in a dual-branch transformer with attention-based fusion and domain-adversarial regularization, with samples filtered by an intensity threshold. Experiments under a leave-one-subject-out protocol demonstrate that the proposed method consistently outperforms single-view and classical baselines, achieving approximately 40% mean accuracy in 7-class subject-independent emotion recognition. The code has been released at https://github.com/Danielz-z/LGF-EEG-Emotion.

---

### Q-realign: Piggybacking Realignment on Quantization for Safe and Efficient LLM Deployment

Authors: Qitao Tan, Xiaoying Song, Ningxi Cheng, Ninghao Liu, Xiaoming Zhai, Lingzi Hong, Yanzhi Wang, Zhen Xiang, Geng Yuan
Published: 2026-01-13
Categories: cs.LG, cs.AI
arXiv: https://arxiv.org/abs/2601.08089v1

Public large language models (LLMs) are typically safety-aligned during pretraining, yet task-specific fine-tuning required for deployment often erodes this alignment and introduces safety risks. Existing defenses either embed safety recovery into fine-tuning or rely on fine-tuning-derived priors for post-hoc correction, leaving safety recovery tightly coupled with training and incurring high computational overhead and a complex workflow. To address these challenges, we propose \texttt{Q-realign}, a post-hoc defense method based on post-training quantization, guided by an analysis of representational structure. By reframing quantization as a dual-objective procedure for compression and safety, \texttt{Q-realign} decouples safety alignment from fine-tuning and naturally piggybacks into modern deployment pipelines. Experiments across multiple models and datasets demonstrate that our method substantially reduces unsafe behaviors while preserving task performance, with significant reductions in memory usage and GPU hours. Notably, our approach can recover the safety alignment of a fine-tuned 7B LLM on a single RTX 4090 within 40 minutes. Overall, our work provides a practical, turnkey solution for safety-aware deployment.

---

### CRAFT: Fine-Grained Cost-Aware Expert Replication For Efficient Mixture-of-Experts Serving

Authors: Adrian Zhao, Zhenkun Cai, Zhenyu Song, Lingfan Yu, Haozheng Fan, Jun Wu, Yida Wang, Nandita Vijaykumar
Published: 2026-01-12
Categories: cs.DC, cs.LG
arXiv: https://arxiv.org/abs/2603.28768v2

Mixture-of-Experts (MoE) has recently emerged as the mainstream architecture for efficiently scaling large language models while maintaining near-constant computational cost. Expert parallelism distributes parameters by partitioning experts across devices, but this introduces token-level load imbalance during inference. Expert replication is a widely adopted load-balancing technique in serving frameworks that alleviates load imbalance in large-scale deployments by replicating experts with high loads. In this work, we demonstrate that existing replication schemes often over-replicate, with many replicas providing marginal improvement. Replicas consume substantial GPU memory, which may lead to resource contention and throughput degradation. We present CRAFT, an efficient expert replication framework that maximizes load balance under a given memory budget by performing fine-grained, per-layer replication based on the estimated replication benefit. CRAFT can be seamlessly integrated into existing serving frameworks without any additional training or model changes. Our evaluation shows that CRAFT increases end-to-end serving throughput by $1.14\times$ on average (up to $1.2\times$) over existing replication techniques in large-scale deployments with models ranging from hundreds of billions to a trillion parameters.

---

### Bridge-RAG: An Abstract Bridge Tree Based Retrieval Augmented Generation Algorithm

Authors: Zihang Li, Wenjun Liu, Yikun Zong, Jiawen Tao, Siying Dai, Songcheng Ren, Zirui Liu, Yuhang Wang, Yanbing Jiang, Tong Yang
Published: 2026-01-12
Categories: cs.IR, cs.AI, cs.CL
arXiv: https://arxiv.org/abs/2603.26668v2

As an important paradigm for enhancing the generation quality of Large Language Models (LLMs), retrieval-augmented generation (RAG) faces the two challenges regarding retrieval accuracy and computational efficiency. This paper presents a novel RAG framework called Bridge-RAG. To overcome the accuracy challenge, we introduce the concept of abstract to bridge query entities and document chunks, providing robust semantic understanding. We organize the abstracts into a tree structure and design a multi-level retrieval strategy to ensure the inclusion of sufficient contextual information. While this hierarchical organization substantially improves answer quality, traversing the tree to locate the abstracts that contain a query entity inevitably introduces additional retrieval overhead. To restore retrieval efficiency, we further integrate the Cuckoo Filter in CFT-RAG, which provides O(1) entity lookup and naturally fits the entity-to-abstract pathway of our framework. Extensive experiments show that Bridge-RAG achieves consistent accuracy improvements across all metrics and up to $1.9\times$ faster retrieval compared to structured RAG baselines.

---

### Trust, Safety, and Accuracy: Assessing LLMs for Routine Maternity Advice

Authors: V Sai Divya, A Bhanusree, Rimjhim, K Venkata Krishna Rao
Published: 2026-01-12
Categories: cs.CL, cs.CY
arXiv: https://arxiv.org/abs/2603.16872v1

Access to reliable maternal healthcare information is a major challenge in rural India due to limited medical resources and infrastructure. With over 830 million internet users and nearly half of rural women online, digital tools offer new opportunities for health education. This study evaluates large language models (LLMs) like ChatGPT-4o, Perplexity AI, and GeminiAI to provide reliable and understandable pregnancy-related information. Seventeen pregnancy-focused questions were posed to each model and compared with responses from maternal health professionals. Evaluations used semantic similarity, noun overlap, and readability metrics to measure content quality. Results show Perplexity closely matched expert semantics, while ChatGPT-4o produced clearer, more understandable text with better medical terminology. As internet access grows in rural areas, LLMs could serve as scalable aids for maternal health education. The study highlights the need for AI tools that balance accuracy and clarity to improve healthcare communication in underserved regions.

---

### Analytical Expression for Spherically Symmetric Photoacoustic Sources: A Unified General Solution (Theoretical Analysis and Derivation)

Authors: Shuang Li, Yibing Wang, Yu Zhang, Changhui Li
Published: 2026-01-12
Categories: physics.optics, cs.CV
arXiv: https://arxiv.org/abs/2602.23375v1

Here we present a comprehensive derivation of the analytical expression for the spatiotemporal acoustic pressure generated by photoacoustic sources with spherically symmetric initial pressure distributions. Starting from the fundamental photoacoustic wave equation, we derive a unified analytical solution applicable to arbitrary spherically symmetric initial distributions. Specific expressions are provided for several common distributions including uniform spherical sources, Gaussian distributions, exponential distributions, and power-law distributions. Far-field approximations are also discussed. The derived expressions provide valuable tools for photoacoustic imaging system design and signal analysis. We provide codes for ultrafast forward simulation using the general analytical spherically symmetric model, the implementation is available in the GitHub repository: \href{https://github.com/JaegerCQ/SlingBAG_Ultra}.

---

### From "Help" to Helpful: A Hierarchical Assessment of LLMs in Mental e-Health Applications

Authors: Philipp Steigerwald, Jens Albrecht
Published: 2026-01-12
Categories: cs.HC, cs.AI, cs.CL, cs.CY
arXiv: https://arxiv.org/abs/2602.18443v2

Psychosocial online counselling frequently encounters generic subject lines that impede efficient case prioritisation. This study evaluates eleven large language models generating six-word subject lines for German counselling emails through hierarchical assessment - first categorising outputs, then ranking within categories to enable manageable evaluation. Nine assessors (counselling professionals and AI systems) enable analysis via Krippendorff's $α$, Spearman's $ρ$, Pearson's $r$ and Kendall's $τ$. Results reveal performance trade-offs between proprietary services and privacy-preserving open-source alternatives, with German fine-tuning consistently improving performance. The study addresses critical ethical considerations for mental health AI deployment including privacy, bias and accountability.

---

### EdgeNav-QE: QLoRA Quantization and Dynamic Early Exit for LAM-based Navigation on Edge Devices

Authors: Mengyun Liu, Shanshan Huang, Jianan Jiang
Published: 2026-01-12
Categories: cs.RO, cs.AI
arXiv: https://arxiv.org/abs/2602.15836v1

Large Action Models (LAMs) have shown immense potential in autonomous navigation by bridging high-level reasoning with low-level control. However, deploying these multi-billion parameter models on edge devices remains a significant challenge due to memory constraints and latency requirements. In this paper, we propose EdgeNav-QE, a novel framework that integrates Quantized Low-Rank Adaptation (QLoRA) with a dynamic early-exit (DEE) mechanism to optimize LAMs for real-time edge navigation. By quantizing the backbone to 4-bit precision and strategically placing early-exit branches, we enable the model to terminate inference early for simple navigation tasks while retaining full depth for complex decision-making. Experimental results on the Habitat-Sim environment with Matterport3D dataset using OpenVLA-7B backbone, demonstrate that EdgeNav-QE reduces inference latency by 82.7% and memory footprint by 66.7% compared to full-precision baselines, while maintaining 81.8% navigation success rate. Furthermore, it outperforms state-of-the-art static early-exit method by 17.9% in latency, demonstrating the superiority of content-aware adaptive computation for safety-critical applications.

---

### A Safety-Constrained Reinforcement Learning Framework for Reliable Wireless Autonomy

Authors: Abdikarim Mohamed Ibrahim, Rosdiadee Nordin
Published: 2026-01-12
Categories: cs.NI, cs.AI
arXiv: https://arxiv.org/abs/2602.13207v1

Artificial intelligence (AI) and reinforcement learning (RL) have shown significant promise in wireless systems, enabling dynamic spectrum allocation, traffic management, and large-scale Internet of Things (IoT) coordination. However, their deployment in mission-critical applications introduces the risk of unsafe emergent behaviors, such as UAV collisions, denial-of-service events, or instability in vehicular networks. Existing safety mechanisms are predominantly reactive, relying on anomaly detection or fallback controllers that intervene only after unsafe actions occur, which cannot guarantee reliability in ultra-reliable low-latency communication (URLLC) settings. In this work, we propose a proactive safety-constrained RL framework that integrates proof-carrying control (PCC) with empowerment-budgeted (EB) enforcement. Each agent action is verified through lightweight mathematical certificates to ensure compliance with interference constraints, while empowerment budgets regulate the frequency of safety overrides to balance safety and autonomy. We implement this framework on a wireless uplink scheduling task using Proximal Policy Optimization (PPO). Simulation results demonstrate that the proposed PCC+EB controller eliminates unsafe transmissions while preserving system throughput and predictable autonomy. Compared with unconstrained and reactive baselines, our method achieves provable safety guarantees with minimal performance degradation. These results highlight the potential of proactive safety constrained RL to enable trustworthy wireless autonomy in future 6G networks.

---

### Does Visual Rendering Bypass Tokenization? Investigating Script-Tokenizer Misalignment in Pixel-Based Language Models

Authors: Lucky Susanto, Musa Izzanardi Wijanarko, Khumaisa Nur'aini, Farid Adilazuarda, Alham Fikri Aji, Derry Tanti Wijaya
Published: 2026-01-12
Categories: cs.CL, cs.AI, cs.LG
arXiv: https://arxiv.org/abs/2602.06973v1

While pixel-based language modeling aims to bypass the sub-word tokenization bottleneck by rendering text as images, recent multimodal variants such as DualGPT reintroduce text tokenizers to improve autoregressive performance. We investigate a fundamental question, does visual rendering truly decouple a model from tokenization constraints? Focusing on four Indonesian low-resource local languages that have their own non-Latin scripts (i.e., Javanese, Balinese, Sundanese, and Lampungnese), we evaluate the impact of script-tokenizer alignment within the DualGPT architecture. Our results show that, despite visual rendering, reintegrating a text tokenizer into the architecture reintroduces the same issue that pixel-based language modeling aims to resolve, which is the tokenizer misalignment problem. Despite having lower OOV and fertility rates, we show that the Llama 2 tokenizer performs significantly worse than a custom tokenizer, with improvements of up to 30.15 chrF++. Our findings serve as a warning for future multimodal variants, as text tokenizers remain a significant barrier to equitable models.

---

### Recontextualizing Famous Quotes for Brand Slogan Generation

Authors: Ziao Yang, Zizhang Chen, Lei Zhang, Hongfu Liu
Published: 2026-01-12
Categories: cs.CL, cs.AI
arXiv: https://arxiv.org/abs/2602.06049v1

Slogans are concise and memorable catchphrases that play a crucial role in advertising by conveying brand identity and shaping public perception. However, advertising fatigue reduces the effectiveness of repeated slogans, creating a growing demand for novel, creative, and insightful slogan generation. While recent work leverages large language models (LLMs) for this task, existing approaches often produce stylistically redundant outputs that lack a clear brand persona and appear overtly machine-generated. We argue that effective slogans should balance novelty with familiarity and propose a new paradigm that recontextualizes persona-related famous quotes for slogan generation. Well-known quotes naturally align with slogan-length text, employ rich rhetorical devices, and offer depth and insight, making them a powerful resource for creative generation. Technically, we introduce a modular framework that decomposes slogan generation into interpretable subtasks, including quote matching, structural decomposition, vocabulary replacement, and remix generation. Extensive automatic and human evaluations demonstrate marginal improvements in diversity, novelty, emotional impact, and human preference over three state-of-the-art LLM baselines.

---

### PTCBENCH: Benchmarking Contextual Stability of Personality Traits in LLM Systems

Authors: Jiongchi Yu, Yuhan Ma, Xiaoyu Zhang, Junjie Wang, Qiang Hu, Chao Shen, Xiaofei Xie
Published: 2026-01-12
Categories: cs.CL, cs.AI
arXiv: https://arxiv.org/abs/2602.00016v1

With the increasing deployment of large language models (LLMs) in affective agents and AI systems, maintaining a consistent and authentic LLM personality becomes critical for user trust and engagement. However, existing work overlooks a fundamental psychological consensus that personality traits are dynamic and context-dependent. To bridge this gap, we introduce PTCBENCH, a systematic benchmark designed to quantify the consistency of LLM personalities under controlled situational contexts. PTCBENCH subjects models to 12 distinct external conditions spanning diverse location contexts and life events, and rigorously assesses the personality using the NEO Five-Factor Inventory. Our study on 39,240 personality trait records reveals that certain external scenarios (e.g., "Unemployment") can trigger significant personality changes of LLMs, and even alter their reasoning capabilities. Overall, PTCBENCH establishes an extensible framework for evaluating personality consistency in realistic, evolving environments, offering actionable insights for developing robust and psychologically aligned AI systems.

---

### G-MemLLM: Gated Latent Memory Augmentation for Long-Context Reasoning in Large Language Models

Authors: Xun Xu
Published: 2026-01-12
Categories: cs.CL, cs.AI
arXiv: https://arxiv.org/abs/2602.00015v1

Large Language Models (LLMs) have demonstrated remarkable capabilities in natural language understanding, yet they remain constrained by the finite capacity of their context windows and the inherent difficulty of maintaining long-term factual consistency during multi-hop reasoning. While existing methods utilize context compression or recurrent tokens, they often suffer from ``context rot'' or the dilution of information over long horizons. In this paper, we propose \textbf{G-MemLLM}, a memory-augmented architecture that integrates a frozen LLM backbone with a trainable \textbf{Latent Memory Bank}. Our key innovation is a GRU-style gated update logic that allows the model to selectively update, preserve, or overwrite latent memory slots, preventing the vanishing gradients of knowledge common in recurrent systems. We evaluate G-MemLLM across scales, from GPT-2 (124M) to Llama 3.1 (8B), on the HotpotQA and Zero-Shot Relation Extraction (ZsRE) benchmarks. Our results demonstrate that G-MemLLM significantly enhances multi-hop reasoning and relational precision, achieving a 13.3\% accuracy boost on ZsRE for Llama 3.1-8B, and it also yields improvements across model scales, boosting Answer F1 by 8.56 points for GPT-2 and increasing Supporting Fact F1 by 6.89 points for Llama 3.1-8B on HotpotQA.

---

### Quantifying non deterministic drift in large language models

Authors: Claire Nicholson
Published: 2026-01-12
Categories: cs.CL, cs.AI
arXiv: https://arxiv.org/abs/2601.19934v1

Large language models (LLMs) are widely used for tasks ranging from summarisation to decision support. In practice, identical prompts do not always produce identical outputs, even when temperature and other decoding parameters are fixed. In this work, we conduct repeated-run experiments to empirically quantify baseline behavioural drift, defined as output variability observed when the same prompt is issued multiple times under operator-free conditions. We evaluate two publicly accessible models, gpt-4o-mini and llama3.1-8b, across five prompt categories using exact repeats, perturbed inputs, and reuse modes at temperatures of 0.0 and 0.7. Drift is measured using unique output fractions, lexical similarity, and word count statistics, enabling direct comparison across models, prompting modes, and deployment types. The results show that nondeterminism persists even at temperature 0.0, with distinct variability patterns by model size, deployment, and prompt type. We situate these findings within existing work on concept drift, behavioural drift, and infrastructure-induced nondeterminism, discuss the limitations of lexical metrics, and highlight emerging semantic approaches. By establishing a systematic empirical baseline in the absence of stabilisation techniques, this study provides a reference point for evaluating future drift mitigation and control methods.

---

### NRR-Phi: A Typed External Text-to-State Interface and Update Contract for Inspectable Ambiguity-State Maintenance

Authors: Kei Saito
Published: 2026-01-12
Categories: cs.CL, cs.AI, cs.LG
arXiv: https://arxiv.org/abs/2601.19933v7

Ambiguity-bearing inputs reach downstream systems through interfaces that favor a single resolved response before later context arrives. Even when alternatives are externalized, their representation and relative activation depend on the update rule. We address this state-maintenance problem within Non-Resolution Reasoning (NRR) by specifying a typed external text-to-state interface and explicit state-update contract. A mapping (phi: T -> S) constructs typed (v,c,w,m) records; declared operators make record carry-forward inspectable, while a record-weight entropy criterion separately tests normalized-weight concentration. The contract covers dampening, calibration, deferred resolution, contradiction-preserving integration, and temporal persistence. Across 580 constructed states/pairs, an executable suite performs 2,740 operator-state measurements. Tested non-violating transitions and calibration/identity checks show 0% record-weight entropy violations; a uniform-subtraction comparison violates the criterion at 1.7%, 6.1%, and 17.8% as subtraction increases. A separate 68-input construction audit finds multiple positive-weight typed records (mean record-weight entropy H = 1.087 bits) under the reported rule-based and archived LLM-assisted procedures, including a Japanese marker-set instantiation. The entropy criterion does not by itself certify record identity or cardinality, semantic adequacy, or end-to-end behavioral improvement. The repository provides deterministic reruns, archived prompt/output artifacts, transcript audit, and a fixed 18-set LLM-case sanity rerun. Phi therefore turns retained-state maintenance into a typed, executable, and falsifiable interface: record carry-forward is inspectable in declared operators, and normalized-weight concentration is separately testable before downstream commitment.

---

### "Newspaper Eat" Means "Not Tasty": A Taxonomy and Benchmark for Coded Language in Real-World Chinese Online Reviews

Authors: Ruyuan Wan, Changye Li, Ting-Hao 'Kenneth' Huang
Published: 2026-01-12
Categories: cs.CL, cs.HC
arXiv: https://arxiv.org/abs/2601.19932v2

Coded language is an important part of human communication. It refers to cases where users intentionally encode meaning so that the surface text differs from the intended meaning and must be decoded to be understood. Current language models handle coded language poorly. Progress has been limited by the lack of real-world datasets and clear taxonomies. This paper introduces CodedLang, a dataset of 7,744 Chinese Google Maps reviews, including 900 reviews with span-level annotations of coded language. We developed a seven-class taxonomy that captures common encoding strategies, including phonetic, orthographic, and cross-lingual substitutions. We benchmarked language models on coded language detection, classification, and review rating prediction. Results show that even strong models can fail to identify or understand coded language. Because many coded expressions rely on pronunciation-based strategies, we further conducted a phonetic analysis of coded and decoded forms. Our code and dataset are publicly available. Together, our results highlight coded language as an important and underexplored challenge for real-world NLP systems.

---

### CascadeMind at SemEval-2026 Task 4: A Hybrid Neuro-Symbolic Cascade for Narrative Similarity

Authors: Sebastien Kawada, Dylan Holyoak
Published: 2026-01-12
Categories: cs.CL
arXiv: https://arxiv.org/abs/2601.19931v3

Across self-consistency samples from an LLM, vote agreement tracks instance difficulty: on SemEval-2026 Task 4 (Narrative Story Similarity), supermajority cases (>= 7/8 votes) resolve at 85 percent accuracy, split votes at 67 percent, and perfect ties at 61 percent, a monotone gradient that holds across the development set. We exploit this in CascadeMind, which routes eight Gemini 2.5 Flash votes by consensus, escalates split votes to additional sampling rounds, and falls through to a symbolic ensemble of theory-inspired narrative signals only on perfect ties (5 percent of cases). The system reached 72.75 percent on Track A test, placing 10th of 44 teams. Ablations show that the symbolic component contributes negligibly end-to-end and that nearly all gains come from confidence-aware routing. The takeaway is methodological: for narrative similarity, calibrating when to spend more compute on a hard instance matters more than adding auxiliary representations to reason about it.

---

### A Dataset of Dengue Hospitalizations in Brazil (1999 to 2021) with Weekly Disaggregation from Monthly Counts

Authors: Lucas M. Morello, Matheus Lima Castro, Pedro Cesar M. G. Camargo, Liliane Moreira Nery, Darllan Collins da Cunha e Silva, Leopoldo Lusquino Filho
Published: 2026-01-12
Categories: cs.LG
arXiv: https://arxiv.org/abs/2601.16994v1

This data paper describes and publicly releases this dataset (v1.0.0), published on Zenodo under DOI 10.5281/zenodo.18189192. Motivated by the need to increase the temporal granularity of originally monthly data to enable more effective training of AI models for epidemiological forecasting, the dataset harmonizes municipal-level dengue hospitalization time series across Brazil and disaggregates them to weekly resolution (epidemiological weeks) through an interpolation protocol with a correction step that preserves monthly totals. The statistical and temporal validity of this disaggregation was assessed using a high-resolution reference dataset from the state of Sao Paulo (2024), which simultaneously provides monthly and epidemiological-week counts, enabling a direct comparison of three strategies: linear interpolation, jittering, and cubic spline. Results indicated that cubic spline interpolation achieved the highest adherence to the reference data, and this strategy was therefore adopted to generate weekly series for the 1999 to 2021 period. In addition to hospitalization time series, the dataset includes a comprehensive set of explanatory variables commonly used in epidemiological and environmental modeling, such as demographic density, CH4, CO2, and NO2 emissions, poverty and urbanization indices, maximum temperature, mean monthly precipitation, minimum relative humidity, and municipal latitude and longitude, following the same temporal disaggregation scheme to ensure multivariate compatibility. The paper documents the datasets provenance, structure, formats, licenses, limitations, and quality metrics (MAE, RMSE, R2, KL, JSD, DTW, and the KS test), and provides usage recommendations for multivariate time-series analysis, environmental health studies, and the development of machine learning and deep learning models for outbreak forecasting.

---

### BibAgent: An Agentic Framework for Traceable Miscitation Detection in Scientific Literature

Authors: Peiran Li, Fangzhou Lin, Shuo Xing, Xiang Zheng, Xi Hong, Siyuan Yang, Jiashuo Sun, Zhengzhong Tu, Chaoqun Ni
Published: 2026-01-12
Categories: cs.DL, cs.AI
arXiv: https://arxiv.org/abs/2601.16993v2

Citations are the bedrock of scientific authority, yet their integrity is compromised by widespread miscitations: ranging from nuanced distortions to fabricated references. Systematic citation verification is currently unfeasible; manual review cannot scale to modern publishing volumes, while existing automated tools are restricted by abstract-only analysis or small-scale, domain-specific datasets in part due to the "paywall barrier" of full-text access. We introduce BibAgent, a scalable, end-to-end agentic framework for automated citation verification. BibAgent integrates retrieval, reasoning, and adaptive evidence aggregation, applying distinct strategies for accessible and paywalled sources. For paywalled references, it leverages a novel Evidence Committee mechanism that infers citation validity via downstream citation consensus. To support systematic evaluation, we contribute a 5-category Miscitation Taxonomy and MisciteBench, a massive cross-disciplinary benchmark comprising 6,350 miscitation samples spanning 254 fields. Our results demonstrate that BibAgent outperforms state-of-the-art Large Language Model (LLM) baselines in citation verification accuracy and interpretability, providing scalable, transparent detection of citation misalignments across the scientific literature.

---

### Limits of n-gram Style Control for LLMs via Logit-Space Injection

Authors: Sami-ul Ahmed
Published: 2026-01-12
Categories: cs.CL
arXiv: https://arxiv.org/abs/2601.16224v1

Large language models (LLMs) are typically personalized via prompt engineering or parameter-efficient fine-tuning such as LoRA. However, writing style can be difficult to distill into a single prompt, and LoRA fine-tuning requires computationally intensive training and infrastructure. We investigate a possible lightweight alternative: steering a frozen LLM with n-gram style priors injected in logit space at decoding time. We train an n-gram model on stylistically distinct corpora -- including Don Quixote, CNN/DailyMail news headlines, and arXiv abstracts -- constructing an interpolated 1-to-3-gram prior over next-token probabilities. During generation we modify the LLM's logits by adding a weighted sum of style log-probabilities from each n-gram order that matches the current context, scaled by a control parameter lambda in (0, 1). We sweep lambda and style corpora and report style perplexity under the n-gram model, base-model perplexity as a proxy for fluency, Jensen-Shannon (JS) divergence between the original and steered token distributions, and token-overlap statistics. On TinyLlama-1.1B we identify a single narrow regime (for the Don Quixote corpus at lambda=0.1) where style perplexity improves by 24.7% and base-model perplexity improves by 51.4% relative to the frozen model. Outside this regime, and for multi-author corpora such as CNN/DailyMail and arXiv abstracts, even small nonzero lambda values generally result in worse style and fluency, and larger lambda values lead to collapse with extreme perplexities and incoherent text. Logit-space injection of n-gram style priors provides lightweight, tunable style control, but it is fragile: it operates effectively only within a narrow range of low lambda values and is consistently outperformed by prompting and LoRA.

---

### Gated Sparse Attention: Combining Computational Efficiency with Training Stability for Long-Context Language Models

Authors: Alfred Shen, Aaron Shen
Published: 2026-01-12
Categories: cs.AI
arXiv: https://arxiv.org/abs/2601.15305v1

The computational burden of attention in long-context language models has motivated two largely independent lines of work: sparse attention mechanisms that reduce complexity by attending to selected tokens, and gated attention variants that improve training sta-bility while mitigating the attention sink phenomenon. We observe that these approaches address complementary weaknesses and propose Gated Sparse Attention (GSA), an architecture that realizes the benefits of both. GSA incorporates a gated lightning indexer with sigmoid activations that produce bounded, interpretable selection scores, an adaptive sparsity controller that modulates the number of attended tokens based on local uncertainty, and dual gating at the value and output stages. We establish theoretical foundations for the approach, including complexity analysis, expressiveness results, and convergence guarantees. In experiments with 1.7B parameter models trained on 400B tokens, GSA matches the efficiency of sparse-only baselines (12-16x speedup at 128K context) while achieving the quality gains associated with gated attention: perplexity improves from 6.03 to 5.70, RULER scores at 128K context nearly double, and attention to the first token, a proxy for attention sinks, drops from 47% to under 4%. Training stability improves markedly, with loss spikes reduced by 98%.

---

### Concatenated Matrix SVD: Compression Bounds, Incremental Approximation, and Error-Constrained Clustering

Authors: Maksym Shamrai
Published: 2026-01-12
Categories: math.NA, cs.LG
arXiv: https://arxiv.org/abs/2601.11626v2

Large collections of matrices arise throughout modern machine learning, signal processing, and scientific computing, where they are commonly compressed by concatenation followed by truncated singular value decomposition (SVD). This strategy enables parameter sharing and efficient reconstruction and has been widely adopted across domains ranging from multi-view learning and signal processing to neural network compression. However, it leaves a fundamental question unanswered: which matrices can be safely concatenated and compressed together under explicit reconstruction error constraints? Existing approaches rely on heuristic or architecture-specific grouping and provide no principled guarantees on the resulting SVD approximation error. In the present work, we introduce a theory-driven framework for compression-aware clustering of matrices under SVD compression constraints. Our analysis establishes new spectral bounds for horizontally concatenated matrices, deriving global upper bounds on the optimal rank-$r$ SVD reconstruction error from lower bounds on singular value growth. The first bound follows from Weyl-type monotonicity under blockwise extensions, while the second leverages singular values of incremental residuals to yield tighter, per-block guarantees. We further develop an efficient approximate estimator based on incremental truncated SVD that tracks dominant singular values without forming the full concatenated matrix. Therefore, we propose three clustering algorithms that merge matrices only when their predicted joint SVD compression error remains below a user-specified threshold. The algorithms span a trade-off between speed, provable accuracy, and scalability, enabling compression-aware clustering with explicit error control.

---

### Reasoning Stabilization Point: A Training-Time Signal for Stable Evidence and Shortcut Reliance

Authors: Sahil Rajesh Dhayalkar
Published: 2026-01-12
Categories: cs.AI, cs.LG
arXiv: https://arxiv.org/abs/2601.11625v1

Fine-tuning pretrained language models can improve task performance while subtly altering the evidence a model relies on. We propose a training-time interpretability view that tracks token-level attributions across finetuning epochs. We define explanation driftas the epoch-to-epoch change in normalized token attributions on a fixed probe set, and introduce the Reasoning Stabilization Point(RSP), the earliest epoch after which drift remains consistently low. RSP is computed from within-run drift dynamics and requires no tuning on out-of-distribution data. Across multiple lightweight transformer classifiers and benchmark classification tasks, drift typically collapses into a low, stable regime early in training, while validation accuracy continues to change only marginally. In a controlled shortcut setting with label-correlated trigger tokens, attribution dynamics expose increasing reliance on the shortcut even when validation accuracy remains competitive. Overall, explanation drift provides a simple, low-cost diagnostic for monitoring how decision evidence evolves during fine-tuning and for selecting checkpoints in a stable-evidence regime.

---

### AnyECG: Evolved ECG Foundation Model for Holistic Health Profiling

Authors: Jun Li, Hongling Zhu, Yujie Xiao, Qinghao Zhao, Yalei Ke, Gongzheng Tang, Guangkun Nie, Deyun Zhang, Jin Li, Canqing Yu, Shenda Hong
Published: 2026-01-12
Categories: eess.SP, cs.AI, cs.LG
arXiv: https://arxiv.org/abs/2601.10748v1

Background: Artificial intelligence enabled electrocardiography (AI-ECG) has demonstrated the ability to detect diverse pathologies, but most existing models focus on single disease identification, neglecting comorbidities and future risk prediction. Although ECGFounder expanded cardiac disease coverage, a holistic health profiling model remains needed. Methods: We constructed a large multicenter dataset comprising 13.3 million ECGs from 2.98 million patients. Using transfer learning, ECGFounder was fine-tuned to develop AnyECG, a foundation model for holistic health profiling. Performance was evaluated using external validation cohorts and a 10-year longitudinal cohort for current diagnosis, future risk prediction, and comorbidity identification. Results: AnyECG demonstrated systemic predictive capability across 1172 conditions, achieving an AUROC greater than 0.7 for 306 diseases. The model revealed novel disease associations, robust comorbidity patterns, and future disease risks. Representative examples included high diagnostic performance for hyperparathyroidism (AUROC 0.941), type 2 diabetes (0.803), Crohn disease (0.817), lymphoid leukemia (0.856), and chronic obstructive pulmonary disease (0.773). Conclusion: The AnyECG foundation model provides substantial evidence that AI-ECG can serve as a systemic tool for concurrent disease detection and long-term risk prediction.

---

### Sensor Placement for Urban Traffic Interpolation: A Data-Driven Evaluation to Inform Policy

Authors: Silke K. Kaiser
Published: 2026-01-12
Categories: eess.SP, cs.LG
arXiv: https://arxiv.org/abs/2601.10747v1

Data on citywide street-segment traffic volumes are essential for urban planning and sustainable mobility management. Yet such data are available only for a limited subset of streets due to the high costs of sensor deployment and maintenance. Traffic volumes on the remaining network are therefore interpolated based on existing sensor measurements. However, current sensor locations are often determined by administrative priorities rather than by data-driven optimization, leading to biased coverage and reduced estimation performance. This study provides a large-scale, real-world benchmarking of easily implementable, data-driven strategies for optimizing the placement of permanent and temporary traffic sensors, using segment-level data from Berlin (Strava bicycle counts) and Manhattan (taxi counts). It compares spatial placement strategies based on network centrality, spatial coverage, feature coverage, and active learning. In addition, the study examines temporal deployment schemes for temporary sensors. The findings highlight that spatial placement strategies that emphasize even spatial coverage and employ active learning achieve the lowest prediction errors. With only 10 sensors, they reduce the mean absolute error by over 60% in Berlin and 70% in Manhattan compared to alternatives. Temporal deployment choices further improve performance: distributing measurements evenly across weekdays reduces error by an additional 7% in Berlin and 21% in Manhattan. Together, these spatial and temporal principles allow temporary deployments to closely approximate the performance of optimally placed permanent deployments. From a policy perspective, the results indicate that cities can substantially improve data usefulness by adopting data-driven sensor placement strategies, while retaining flexibility in choosing between temporary and permanent deployments.

---

### SAGE: Tool-Augmented LLM Task Solving Strategies in Scalable Multi-Agent Environments

Authors: Robert K. Strehlow, Tobias Küster, Oskar F. Kupke, Brandon Llanque Kurps, Fikret Sivrikaya, Sahin Albayrak
Published: 2026-01-12
Categories: cs.SE, cs.AI, cs.HC, cs.MA
arXiv: https://arxiv.org/abs/2601.09750v1

Large language models (LLMs) have proven to work well in question-answering scenarios, but real-world applications often require access to tools for live information or actuation. For this, LLMs can be extended with tools, which are often defined in advance, also allowing for some fine-tuning for specific use cases. However, rapidly evolving software landscapes and individual services require the constant development and integration of new tools. Domain- or company-specific tools can greatly elevate the usefulness of an LLM, but such custom tools can be problematic to integrate, or the LLM may fail to reliably understand and use them. For this, we need strategies to define new tools and integrate them into the LLM dynamically, as well as robust and scalable zero-shot prompting methods that can make use of those tools in an efficient manner. In this paper, we present SAGE, a specialized conversational AI interface, based on the OPACA framework for tool discovery and execution. The integration with OPACA makes it easy to add new tools or services for the LLM to use, while SAGE itself presents rich extensibility and modularity. This not only provides the ability to seamlessly switch between different models (e.g. GPT, LLAMA), but also to add and select prompting methods, involving various setups of differently prompted agents for selecting and executing tools and evaluating the results. We implemented a number of task-solving strategies, making use of agentic concepts and prompting methods in various degrees of complexity, and evaluated those against a comprehensive set of benchmark services. The results are promising and highlight the distinct strengths and weaknesses of different task-solving strategies. Both SAGE and the OPACA framework, as well as the different benchmark services and results, are available as Open Source/Open Data on GitHub.

---

### R-LAM: Reproducibility-Constrained Large Action Models for Scientific Workflow Automation

Authors: Suriya Sureshkumar
Published: 2026-01-12
Categories: cs.SE, cs.AI, cs.LG
arXiv: https://arxiv.org/abs/2601.09749v1

Large Action Models (LAMs) extend large language models by enabling autonomous decision-making and tool execution, making them promising for automating scientific workflows. However, scientific workflows impose strict requirements on reproducibility, auditability, and deterministic execution, which are not satisfied by generic LLM-based agents. Unconstrained action generation can lead to silent state changes, non-deterministic executions, and irreproducible experimental results, limiting the applicability of LAMs in scientific settings. In this paper, we propose R-LAM, a reproducibility-constrained framework for applying Large Action Models to scientific workflow automation. R-LAM introduces structured action schemas, deterministic execution policies, and explicit provenance tracking to ensure that every action and intermediate artifact is auditable and replayable. The framework supports failure-aware execution loops and controlled workflow forking, enabling iterative experimentation without compromising reproducibility. We implement R-LAM as a lightweight Python framework and release it as an open-source PyPI package to facilitate reproducible research. An experimental evaluation of representative scientific workflows demonstrates that R-LAM improves reproducibility success rates and execution reliability compared to unconstrained LLM-based agents, while retaining adaptive control over workflow execution.

---

### Bridging the Gap: Empowering Small Models in Reliable OpenACC-based Parallelization via GEPA-Optimized Prompting

Authors: Samyak Jhaveri, Cristina V. Lopes
Published: 2026-01-12
Categories: cs.SE, cs.AI, cs.DC
arXiv: https://arxiv.org/abs/2601.08884v1

OpenACC lowers the barrier to GPU offloading, but writing high-performing pragma remains complex, requiring deep domain expertise in memory hierarchies, data movement, and parallelization strategies. Large Language Models (LLMs) present a promising potential solution for automated parallel code generation, but naive prompting often results in syntactically incorrect directives, uncompilable code, or performance that fails to exceed CPU baselines. We present a systematic prompt optimization approach to enhance OpenACC pragma generation without the prohibitive computational costs associated with model post-training. Leveraging the GEPA (GEnetic-PAreto) framework, we iteratively evolve prompts through a reflective feedback loop. This process utilizes crossover and mutation of instructions, guided by expert-curated gold examples and structured feedback based on clause- and clause parameter-level mismatches between the gold and predicted pragma. In our evaluation on the PolyBench suite, we observe an increase in compilation success rates for programs annotated with OpenACC pragma generated using the optimized prompts compared to those annotated using the simpler initial prompt, particularly for the "nano"-scale models. Specifically, with optimized prompts, the compilation success rate for GPT-4.1 Nano surged from 66.7% to 93.3%, and for GPT-5 Nano improved from 86.7% to 100%, matching or surpassing the capabilities of their significantly larger, more expensive versions. Beyond compilation, the optimized prompts resulted in a 21% increase in the number of programs that achieve functional GPU speedups over CPU baselines. These results demonstrate that prompt optimization effectively unlocks the potential of smaller, cheaper LLMs in writing stable and effective GPU-offloading directives, establishing a cost-effective pathway to automated directive-based parallelization in HPC workflows.

---

### Compressing Vision Transformers in Geospatial Transfer Learning with Manifold-Constrained Optimization

Authors: Thomas Snyder, H. Lexie Yang, Stefan Schnake, Steffen Schotthöfer
Published: 2026-01-12
Categories: cs.CV, cs.AI
arXiv: https://arxiv.org/abs/2601.08882v1

Deploying geospatial foundation models on resource-constrained edge devices demands compact architectures that maintain high downstream performance. However, their large parameter counts and the accuracy loss often induced by compression limit practical adoption. In this work, we leverage manifold-constrained optimization framework DLRT to compress large vision transformer-based geospatial foundation models during transfer learning. By enforcing structured low-dimensional parameterizations aligned with downstream objectives, this approach achieves strong compression while preserving task-specific accuracy. We show that the method outperforms of-the-shelf low-rank methods as LoRA. Experiments on diverse geospatial benchmarks confirm substantial parameter reduction with minimal accuracy loss, enabling high-performing, on-device geospatial models.

---

### TAG-MoE: Task-Aware Gating for Unified Generative Mixture-of-Experts

Authors: Yu Xu, Hongbin Yan, Juan Cao, Yiji Cheng, Tiankai Hang, Runze He, Zijin Yin, Shiyi Zhang, Yuxin Zhang, Jintao Li, Chunyu Wang, Qinglin Lu et al.
Published: 2026-01-12
Categories: cs.CV, cs.AI
arXiv: https://arxiv.org/abs/2601.08881v2

Unified image generation and editing models suffer from severe task interference in dense diffusion transformers architectures, where a shared parameter space must compromise between conflicting objectives (e.g., local editing v.s. subject-driven generation). While the sparse Mixture-of-Experts (MoE) paradigm is a promising solution, its gating networks remain task-agnostic, operating based on local features, unaware of global task intent. This task-agnostic nature prevents meaningful specialization and fails to resolve the underlying task interference. In this paper, we propose a novel framework to inject semantic intent into MoE routing. We introduce a Hierarchical Task Semantic Annotation scheme to create structured task descriptors (e.g., scope, type, preservation). We then design Predictive Alignment Regularization to align internal routing decisions with the task's high-level semantics. This regularization evolves the gating network from a task-agnostic executor to a dispatch center. Our model effectively mitigates task interference, outperforming dense baselines in fidelity and quality, and our analysis shows that experts naturally develop clear and semantically correlated specializations.

---

### The Semantic Lifecycle in Embodied AI: Acquisition, Representation and Storage via Foundation Models

Authors: Shuai Chen, Hao Chen, Yuanchen Bei, Tianyang Zhao, Zhibo Zhou, Feiran Huang
Published: 2026-01-12
Categories: cs.CV
arXiv: https://arxiv.org/abs/2601.08876v1

Semantic information in embodied AI is inherently multi-source and multi-stage, making it challenging to fully leverage for achieving stable perception-to-action loops in real-world environments. Early studies have combined manual engineering with deep neural networks, achieving notable progress in specific semantic-related embodied tasks. However, as embodied agents encounter increasingly complex environments and open-ended tasks, the demand for more generalizable and robust semantic processing capabilities has become imperative. Recent advances in foundation models (FMs) address this challenge through their cross-domain generalization abilities and rich semantic priors, reshaping the landscape of embodied AI research. In this survey, we propose the Semantic Lifecycle as a unified framework to characterize the evolution of semantic knowledge within embodied AI driven by foundation models. Departing from traditional paradigms that treat semantic processing as isolated modules or disjoint tasks, our framework offers a holistic perspective that captures the continuous flow and maintenance of semantic knowledge. Guided by this embodied semantic lifecycle, we further analyze and compare recent advances across three key stages: acquisition, representation, and storage. Finally, we summarize existing challenges and outline promising directions for future research.

---

### Learning Domain-Invariant Representations for Cross-Domain Image Registration via Scene-Appearance Disentanglement

Authors: Jiahao Qin, Yiwen Wang
Published: 2026-01-12
Categories: cs.CV, cs.AI, cs.LG
arXiv: https://arxiv.org/abs/2601.08875v2

Image registration under domain shift remains a fundamental challenge in computer vision and medical imaging: when source and target images exhibit systematic intensity differences, the brightness constancy assumption underlying conventional registration methods is violated, rendering correspondence estimation ill-posed. We propose SAR-Net, a unified framework that addresses this challenge through principled scene-appearance disentanglement. Our key insight is that observed images can be decomposed into domain-invariant scene representations and domain-specific appearance codes, enabling registration via re-rendering rather than direct intensity matching. We establish theoretical conditions under which this decomposition enables consistent cross-domain alignment (Proposition 1) and prove that our scene consistency loss provides a sufficient condition for geometric correspondence in the shared latent space (Proposition 2). Empirically, we validate SAR-Net on the ANHIR (Automatic Non-rigid Histological Image Registration) challenge benchmark, where multi-stain histopathology images exhibit coupled domain shift from different staining protocols and geometric distortion from tissue preparation. Our method achieves a median relative Target Registration Error (rTRE) of 0.25%, outperforming the state-of-the-art MEVIS method (0.27% rTRE) by 7.4%, with robustness of 99.1%. Code is available at https://github.com/D-ST-Sword/SAR-NET .

---

### The Illusion of Friendship: Why Generative AI Demands Unprecedented Ethical Vigilance

Authors: Md Zahidul Islam
Published: 2026-01-12
Categories: cs.CY, cs.AI
arXiv: https://arxiv.org/abs/2601.08874v1

GenAI systems are increasingly used for drafting, summarisation, and decision support, offering substantial gains in productivity and reduced cognitive load. However, the same natural language fluency that makes these systems useful can also blur the boundary between tool and companion. This boundary confusion may encourage some users to experience GenAI as empathic, benevolent, and relationally persistent. Emerging reports suggest that some users may form emotionally significant attachments to conversational agents, in some cases with harmful consequences, including dependency and impaired judgment. This paper develops a philosophical and ethical argument for why the resulting illusion of friendship is both understandable and can be ethically risky. Drawing on classical accounts of friendship, the paper explains why users may understandably interpret sustained supportive interaction as friend like. It then advances a counterargument that despite relational appearances, GenAI lacks moral agency: consciousness, intention, and accountability and therefore does not qualify as a true friend. To demystify the illusion, the paper presents a mechanism level explanation of how transformer based GenAI generates responses often producing emotionally resonant language without inner states or commitments. Finally, the paper proposes a safeguard framework for safe and responsible GenAI use to reduce possible anthropomorphic cues generated by the GenAI systems. The central contribution is to demystify the illusion of friendship and explain the computational background so that we can shift the emotional attachment with GenAI towards necessary human responsibility and thereby understand how institutions, designers, and users can preserve GenAI's benefits while mitigating over reliance and emotional misattribution.

---

### ForensicFormer: Hierarchical Multi-Scale Reasoning for Cross-Domain Image Forgery Detection

Authors: Hema Hariharan Samson
Published: 2026-01-12
Categories: cs.CV, cs.AI, cs.LG, cs.MM
arXiv: https://arxiv.org/abs/2601.08873v1

The proliferation of AI-generated imagery and sophisticated editing tools has rendered traditional forensic methods ineffective for cross-domain forgery detection. We present ForensicFormer, a hierarchical multi-scale framework that unifies low-level artifact detection, mid-level boundary analysis, and high-level semantic reasoning via cross-attention transformers. Unlike prior single-paradigm approaches, which achieve <75% accuracy on out-of-distribution datasets, our method maintains 86.8% average accuracy across seven diverse test sets, spanning traditional manipulations, GAN-generated images, and diffusion model outputs - a significant improvement over state-of-the-art universal detectors. We demonstrate superior robustness to JPEG compression (83% accuracy at Q=70 vs. 66% for baselines) and provide pixel-level forgery localization with a 0.76 F1-score. Extensive ablation studies validate that each hierarchical component contributes 4-10% accuracy improvement, and qualitative analysis reveals interpretable forensic features aligned with human expert reasoning. Our work bridges classical image forensics and modern deep learning, offering a practical solution for real-world deployment where manipulation techniques are unknown a priori.

---

### Semantic visually-guided acoustic highlighting with large vision-language models

Authors: Junhua Huang, Chao Huang, Chenliang Xu
Published: 2026-01-12
Categories: cs.SD, cs.AI, eess.AS
arXiv: https://arxiv.org/abs/2601.08871v1

Balancing dialogue, music, and sound effects with accompanying video is crucial for immersive storytelling, yet current audio mixing workflows remain largely manual and labor-intensive. While recent advancements have introduced the visually guided acoustic highlighting task, which implicitly rebalances audio sources using multimodal guidance, it remains unclear which visual aspects are most effective as conditioning signals.We address this gap through a systematic study of whether deep video understanding improves audio remixing. Using textual descriptions as a proxy for visual analysis, we prompt large vision-language models to extract six types of visual-semantic aspects, including object and character appearance, emotion, camera focus, tone, scene background, and inferred sound-related cues. Through extensive experiments, camera focus, tone, and scene background consistently yield the largest improvements in perceptual mix quality over state-of-the-art baselines. Our findings (i) identify which visual-semantic cues most strongly support coherent and visually aligned audio remixing, and (ii) outline a practical path toward automating cinema-grade sound design using lightweight guidance derived from large vision-language models.

---

### CtrlFuse: Mask-Prompt Guided Controllable Infrared and Visible Image Fusion

Authors: Yiming Sun, Yuan Ruan, Qinghua Hu, Pengfei Zhu
Published: 2026-01-12
Categories: cs.CV
arXiv: https://arxiv.org/abs/2601.08619v1

Infrared and visible image fusion generates all-weather perception-capable images by combining complementary modalities, enhancing environmental awareness for intelligent unmanned systems. Existing methods either focus on pixel-level fusion while overlooking downstream task adaptability or implicitly learn rigid semantics through cascaded detection/segmentation models, unable to interactively address diverse semantic target perception needs. We propose CtrlFuse, a controllable image fusion framework that enables interactive dynamic fusion guided by mask prompts. The model integrates a multi-modal feature extractor, a reference prompt encoder (RPE), and a prompt-semantic fusion module (PSFM). The RPE dynamically encodes task-specific semantic prompts by fine-tuning pre-trained segmentation models with input mask guidance, while the PSFM explicitly injects these semantics into fusion features. Through synergistic optimization of parallel segmentation and fusion branches, our method achieves mutual enhancement between task performance and fusion quality. Experiments demonstrate state-of-the-art results in both fusion controllability and segmentation accuracy, with the adapted task branch even outperforming the original segmentation model.

---