id
large_stringlengths 9
16
| title
large_stringlengths 1
382
| abstract
large_stringlengths 3
6.09k
| publish_date
date32 | update_date
date32 | categories
large listlengths 1
13
| authors
large_stringlengths 3
62.8k
|
|---|---|---|---|---|---|---|
2601.03976
|
On-Device Deep Reinforcement Learning for Decentralized Task Offloading Performance trade-offs in the training process
|
Allowing less capable devices to offload computational tasks to more powerful devices or servers enables the development of new applications that may not run correctly on the device itself. Deciding where and why to run each of those applications is a complex task. Therefore, different approaches have been adopted to make offloading decisions. In this work, we propose a decentralized Deep Reinforcement Learning (DRL) agent to address the selection of computing locations. Unlike most existing work, we analyze it in a real testbed composed of various edge devices running the agent to determine where to execute each task. These devices are connected to a Multi-Access Edge Computing (MEC) server and a Cloud server through 5G communications. We evaluate not only the agent's performance in meeting task requirements but also the implications of running this type of agent locally, assessing the trade-offs of training locally versus remotely in terms of latency and energy consumption.
| 2026-01-07
| 2026-01-08
|
[
"cs.ET",
"cs.SY",
"eess.SY"
] |
Gorka Nieto, Idoia de la Iglesia, Cristina Perfecto, Unai Lopez-Novoa
|
2503.23314
|
SPIO: Ensemble and Selective Strategies via LLM-Based Multi-Agent Planning in Automated Data Science
|
Large Language Models (LLMs) have enabled dynamic reasoning in automated data analytics, yet recent multi-agent systems remain limited by rigid, single-path workflows that restrict strategic exploration and often lead to suboptimal outcomes. To overcome these limitations, we propose SPIO (Sequential Plan Integration and Optimization), a framework that replaces rigid workflows with adaptive, multi-path planning across four core modules: data preprocessing, feature engineering, model selection, and hyperparameter tuning. In each module, specialized agents generate diverse candidate strategies, which are cascaded and refined by an optimization agent. SPIO offers two operating modes: SPIO-S for selecting a single optimal pipeline, and SPIO-E for ensembling top-k pipelines to maximize robustness. Extensive evaluations on Kaggle and OpenML benchmarks show that SPIO consistently outperforms state-of-the-art baselines, achieving an average performance gain of 5.6%. By explicitly exploring and integrating multiple solution paths, SPIO delivers a more flexible, accurate, and reliable foundation for automated data science.
| 2026-01-07
| 2026-01-08
|
[
"cs.AI",
"cs.CL",
"cs.LG",
"cs.MA"
] |
Wonduk Seo, Juhyeon Lee, Yanjun Shao, Qingshan Zhou, Seunghyun Lee, Yi Bu
|
2504.05738
|
LLM-assisted Mutation for Whitebox API Testing
|
Cloud applications heavily rely on APIs to communicate with each other and exchange data. To ensure the reliability of cloud applications, cloud providers widely adopt API testing techniques. Unfortunately, existing API testing approaches are insufficient to reach strict conditions, a problem known as fitness plateaus, due to the lack of gradient provided by coverage metrics. To address this issue, we propose MioHint, a novel white-box API testing approach that leverages the code comprehension capabilities of Large Language Model (LLM) to boost API testing. The key challenge of LLM-based API testing lies in system-level testing, which emphasizes the dependencies between requests and targets across functions and files, thereby making the entire codebase the object of analysis. However, feeding the entire codebase to an LLM is impractical due to its limited context length and short memory. MioHint addresses this challenge by synergizing static analysis with LLMs. We retrieve relevant code with data-dependency analysis at the statement level, including def-use analysis for variables used in the target and function expansion for subfunctions called by the target.
To evaluate the effectiveness of our method, we conducted experiments across 16 real-world REST API services. The findings reveal that MioHint achieves an average increase of 4.95% absolute in line coverage compared to the baseline, EvoMaster, alongside a remarkable factor of 67x improvement in mutation accuracy. Furthermore, our method successfully covers over 57% of hard-to-cover targets while in baseline the coverage is less than 10%.
| 2026-01-07
| 2026-01-08
|
[
"cs.SE"
] |
Jia Li, Jiacheng Shen, Yuxin Su, Michael R. Lyu
|
2501.09280
|
Effect of accretion on scalar superradiant instability
|
Superradiance can lead to the formation of a black hole (BH) condensate system. We thoroughly investigate the accretion effect on the evolution of this system, and the gravitational wave signals it emits in the presence of multiple superradiance modes. Assuming the multiplication of the BH mass and scalar mass as a small number, we obtain the analytical approximations of all important quantities, which can be directly applied to phenomenological studies. In addition, we confirm that accretion could significantly enhance the gravitational wave (GW) emission and reduce its duration, and show that the GW beat signature is similarly modified.
| 2026-01-07
| 2026-01-08
|
[
"gr-qc",
"hep-ph"
] |
Yin-Da Guo, Shou-Shan Bao, Tianjun Li, Hong Zhang
|
2601.03588
|
AR Object Layout Method Using Miniature Room Generated from Depth Data
|
In augmented reality (AR), users can place virtual objects anywhere in a real-world room, called AR layout. Although several object manipulation techniques have been proposed in AR, it is difficult to use them for AR layout owing to the difficulty in freely changing the position and size of virtual objects. In this study, we make the World-in-Miniature (WIM) technique available in AR to support AR layout. The WIM technique is a manipulation technique that uses miniatures, which has been proposed as a manipulation technique for virtual reality (VR). Our system uses the AR device's depth sensors to acquire a mesh of the room in real-time to create and update a miniature of a room in real-time. In our system, users can use miniature objects to move virtual objects to arbitrary positions and scale them to arbitrary sizes. In addition, because the miniature object can be manipulated instead of the real-scale object, we assumed that our system will shorten the placement time and reduce the workload of the user. In our previous study, we created a prototype and investigated the properties of manipulating miniature objects in AR. In this study, we conducted an experiment to evaluate how our system can support AR layout. To conduct a task close to the actual use, we used various objects and made the participants design an AR layout of their own will. The results showed that our system significantly reduced workload in physical and temporal demand. Although, there was no significant difference in the total manipulation time.
| 2026-01-07
| 2026-01-08
|
[
"cs.HC"
] |
Keiichi Ihara, Ikkaku Kawaguchi
|
2601.00744
|
Strong anchoring boundary conditions in nematic liquid crystals: Higher-order corrections to the Oseen-Frank limit and a revised small-domain theory
|
Strong anchoring boundary conditions are conventionally modelled by imposing Dirichlet conditions on the order parameter in Landau--de Gennes theory, neglecting the finite surface energy of realistic anchoring. This work revisits the strong anchoring limit for nematic liquid crystals in confined two-dimensional domains. By explicitly retaining a Rapini-Papoular surface energy and adopting a scaling where the extrapolation length $l_{ex}$ is comparable to the coherence length $ξ$, we analyse both the small-domain ($ε= h/ξ\to 0$; $h$ is the domain size) and Oseen-Frank $(ε\to \infty$) asymptotic regimes. In the small-domain limit, the leading-order equilibrium solution is given by the average of the boundary data, which can vanish in symmetrically frustrated geometries, leading to isotropic melting. In the large-domain limit, matched asymptotic expansions reveal that surface anchoring introduces an $O(1/ε)$ correction to the director field near boundaries, in contrast to the $O(1/ε^2)$ correction predicted by Dirichlet conditions. The analysis captures the detailed structure of interior and boundary defects, showing that mixed (Robin-type) boundary conditions yield smoother defect cores and more physical predictions than rigid Dirichlet conditions. Numerical solutions for square and circular wells with tangential anchoring illustrate the differences between the two boundary condition treatments, particularly in defect morphology. The results demonstrate that a consistent treatment of anchoring energetics is essential for accurate modelling of nematic equilibria in micro- and nano-scale confined geometries.
| 2026-01-07
| 2026-01-08
|
[
"cond-mat.soft"
] |
Prabakaran Rajamanickam
|
2601.04185
|
ImLoc: Revisiting Visual Localization with Image-based Representation
|
Existing visual localization methods are typically either 2D image-based, which are easy to build and maintain but limited in effective geometric reasoning, or 3D structure-based, which achieve high accuracy but require a centralized reconstruction and are difficult to update. In this work, we revisit visual localization with a 2D image-based representation and propose to augment each image with estimated depth maps to capture the geometric structure. Supported by the effective use of dense matchers, this representation is not only easy to build and maintain, but achieves highest accuracy in challenging conditions. With compact compression and a GPU-accelerated LO-RANSAC implementation, the whole pipeline is efficient in both storage and computation and allows for a flexible trade-off between accuracy and highest memory efficiency. Our method achieves a new state-of-the-art accuracy on various standard benchmarks and outperforms existing memory-efficient methods at comparable map sizes. Code will be available at https://github.com/cvg/Hierarchical-Localization.
| 2026-01-07
| 2026-01-08
|
[
"cs.CV"
] |
Xudong Jiang, Fangjinhua Wang, Silvano Galliani, Christoph Vogel, Marc Pollefeys
|
2601.03545
|
muT2-NMR: Micro-Scale Correlation Relaxometry for in-situ High-Pressure Nuclear Magnetic Resonance
|
Over the last decade, frequency-domain in-situ high-pressure nuclear magnetic resonance (NMR) spectroscopy in diamond anvil cells (DACs) has been employed as a structural and electronic probe of condensed matter systems at pressures well into the megabar range. However, extensive spin interactions and sample heterogeneities under pressure often lead to significant spectral overlap, inhibiting independent observation of chemically similar spin sub-species in the same sample. In this work, we introduce a time-domain relaxometry framework specifically suited for DAC experiments, named muT2-NMR. Experimental flexibility and operational robustness are benchmarked on three hydrogen-rich molecular solids at pressures up to 72 GPa. We demonstrate that muT2-NMR can resolve individual molecular subunits in relaxation space, paving the way for novel high-pressure, high-resolution NMR applications in molecular solids.
| 2026-01-07
| 2026-01-08
|
[
"physics.ins-det",
"cond-mat.mtrl-sci"
] |
Thomas Meier, Meng Yang, Yishan Zhou, Yunhua Fu, Rui Zhang, Ziliang Wang, Tianyao Zheng, Rajesh Jana, Takeshi Nakagawa
|
2601.03676
|
Towards Compositional Generalization of LLMs via Skill Taxonomy Guided Data Synthesis
|
Large Language Models (LLMs) and agent-based systems often struggle with compositional generalization due to a data bottleneck in which complex skill combinations follow a long-tailed, power-law distribution, limiting both instruction-following performance and generalization in agent-centric tasks. To address this challenge, we propose STEPS, a Skill Taxonomy guided Entropy-based Post-training data Synthesis framework for generating compositionally challenging data. STEPS explicitly targets compositional generalization by uncovering latent relationships among skills and organizing them into an interpretable, hierarchical skill taxonomy using structural information theory. Building on this taxonomy, we formulate data synthesis as a constrained information maximization problem, selecting skill combinations that maximize marginal structural information within the hierarchy while preserving semantic coherence. Experiments on challenging instruction-following benchmarks show that STEPS outperforms existing data synthesis baselines, while also yielding improved compositional generalization in downstream agent-based evaluations.
| 2026-01-07
| 2026-01-08
|
[
"cs.CL",
"cs.AI"
] |
Yifan Wei, Li Du, Xiaoyan Yu, Yang Feng, Angsheng Li
|
2403.18499
|
Mechanisms of THz Radiation Generation in Multi-Color Laser-Plasma Interactions: A Review Across Diverse Media
|
The exploration of Terahertz (THz) waves has captivated researchers across diverse scientific disciplines such as physics, spectroscopy, chemistry, biology, and engineering, driven by the myriad applications these waves offer. Within this expansive landscape, the development of efficient and reliable THz sources stands as a paramount objective. In the pursuit of this goal, a multitude of approaches have been undertaken, with a notable contender emerging in the form of laser-induced plasma. Harnessing the advancements in ultrafast pulses, laser-induced plasma has proven to be a promising tool for generating THz waves. Its appeal lies in the robust attributes of a high power threshold, intense THz signal, and an broadband THz spectrum. This paper delves into a comprehensive review of the physics and progress underlying THz generation from laser-induced plasmas, exploring scenarios where plasmas are induced in gases, liquids, and solids. The interactions between lasers and plasmas involve complex physical processes, resulting in a variety of laser plasma scenarios for THz generation. In this review, the focus is specifically placed on classifying THz generation based on different physical mechanisms and also examines the characteristics of the emitted THz waves. By categorizing the processes, a deeper understanding of the underlying principles can be attained.
| 2026-01-07
| 2026-01-08
|
[
"physics.plasm-ph",
"physics.app-ph"
] |
A. A. Molavi Choobini, S. S. Ghaffari-Oskooei, M. Shahmansouri, F. M. Aghamir
|
2601.03889
|
Spectral Manifold Regularization for Stable and Modular Routing in Deep MoE Architectures
|
Mixture of Experts (MoE) architectures enable efficient scaling of neural networks but suffer from expert collapse, where routing converges to a few dominant experts. This reduces model capacity and causes catastrophic interference during adaptation. We propose the Spectrally-Regularized Mixture of Experts (SR-MoE), which imposes geometric constraints on the routing manifold to enforce structural modularity. Our method uses dual regularization: spectral norm constraints bound routing function Lipschitz continuity, while stable rank penalties preserve high-dimensional feature diversity in expert selection. We evaluate SR-MoE across architectural scales and dataset complexities using modular one-shot adaptation tasks. Results show that traditional linear gating fails with increasing depth (accuracy drops up to 4.72% due to expert entanglement), while SR-MoE maintains structural integrity (mean interference -0.32%). Our spectral constraints facilitate positive knowledge transfer, enabling localized expert updates without global performance decay. SR-MoE provides a general solution for building high-capacity, modular networks capable of stable lifelong learning.
| 2026-01-07
| 2026-01-08
|
[
"cs.LG",
"cs.AI"
] |
Ibrahim Delibasoglu
|
2601.03836
|
Logic Programming with Extensible Types
|
Logic programming languages present clear advantages in terms of declarativeness and conciseness. However, the ideas of logic programming have been met with resistance in other programming communities, and have not generally been adopted by other paradigms and languages. This paper proposes a novel way to incorporate logic programming in an existing codebase in a typed functional programming language. Our approach integrates with the host language without sacrificing static typing, and leverages strengths of typed functional programming such as polymorphism and higher-order. We do so by combining three ideas. First, we use the extensible types technique to allow values of the host language to contain logic variables. Second, we implement a unification algorithm that works for any data structure that supports certain operations.Third, we introduce a domain-specific language to define and query predicates. We demonstrate our proposal via a series of examples, and provide aids to make the notation convenient for users, showing that the proposed approach is not just technically possible but also practical. Our ideas have been implemented in the language Haskell with very good results.
| 2026-01-07
| 2026-01-08
|
[
"cs.PL",
"cs.LO"
] |
Ivan Perez, Angel Herranz
|
2601.03645
|
LLM-MC-Affect: LLM-Based Monte Carlo Modeling of Affective Trajectories and Latent Ambiguity for Interpersonal Dynamic Insight
|
Emotional coordination is a core property of human interaction that shapes how relational meaning is constructed in real time. While text-based affect inference has become increasingly feasible, prior approaches often treat sentiment as a deterministic point estimate for individual speakers, failing to capture the inherent subjectivity, latent ambiguity, and sequential coupling found in mutual exchanges. We introduce LLM-MC-Affect, a probabilistic framework that characterizes emotion not as a static label, but as a continuous latent probability distribution defined over an affective space. By leveraging stochastic LLM decoding and Monte Carlo estimation, the methodology approximates these distributions to derive high-fidelity sentiment trajectories that explicitly quantify both central affective tendencies and perceptual ambiguity. These trajectories enable a structured analysis of interpersonal coupling through sequential cross-correlation and slope-based indicators, identifying leading or lagging influences between interlocutors. To validate the interpretive capacity of this approach, we utilize teacher-student instructional dialogues as a representative case study, where our quantitative indicators successfully distill high-level interaction insights such as effective scaffolding. This work establishes a scalable and deployable pathway for understanding interpersonal dynamics, offering a generalizable solution that extends beyond education to broader social and behavioral research.
| 2026-01-07
| 2026-01-08
|
[
"cs.CL",
"cs.CY"
] |
Yu-Zheng Lin, Bono Po-Jen Shih, John Paul Martin Encinas, Elizabeth Victoria Abraham Achom, Karan Himanshu Patel, Jesus Horacio Pacheco, Sicong Shao, Jyotikrishna Dass, Soheil Salehi, Pratik Satam
|
2512.04615
|
Ground state energy and phase transitions of Long-range XXZ using VQE
|
The variational quantum eigen solver (VQE), has been widely used to find the ground state energy of different Hamiltonians with no analytical solutions and are classically difficult to compute. In our work, we have used VQE to identify the phase transition boundary for an infinite order phase transition. We use long-range XXZ (LRXXZ) chain for our study. In order to probe infinite order phase transition, we propose to utilise the ground state energy obtained from VQE. The idea rests on the argument that VQE requires an ansatz circuit; therefore, the accuracy of the VQE will rely on this ansatz circuit. We have designed this circuit such that the estimated ground state energy is sensitive to the phase it is evaluated in. It is achieved by applying the constraint that the net spin remains constant throughout the optimisation process. Consequently, the ansatz works in a certain phase where it gives relatively small random error, as it should, when compared to the error in ground state energy calculations of the other phases, where the ansatz fails. By identifying these changes in the behaviour of the error in ground state energy using VQE, we were able to determine the phase boundaries. Using exact diagonalisation, we also compare the behaviour of the energy gradient and energy gap across both the phase transition boundaries for this model. Further, by increasing the depth of the optimisation circuit, we also accurately evaluate the ground energy of the LRXXZ chain for the value of coupling constant, J equal to -1
| 2026-01-07
| 2026-01-08
|
[
"quant-ph",
"cond-mat.str-el"
] |
Mrinal Dev, Shraddha Sharma
|
2601.04143
|
Local structure of etale algebras
|
The goal of this note is to provide a constructive version of the proof of local structure of etale algebras.
| 2026-01-07
| 2026-01-08
|
[
"math.AC"
] |
Thierry Coquand
|
2601.03743
|
O-Researcher: An Open Ended Deep Research Model via Multi-Agent Distillation and Agentic RL
|
The performance gap between closed-source and open-source large language models (LLMs) is largely attributed to disparities in access to high-quality training data. To bridge this gap, we introduce a novel framework for the automated synthesis of sophisticated, research-grade instructional data. Our approach centers on a multi-agent workflow where collaborative AI agents simulate complex tool-integrated reasoning to generate diverse and high-fidelity data end-to-end. Leveraging this synthesized data, we develop a two-stage training strategy that integrates supervised fine-tuning with a novel reinforcement learning method, designed to maximize model alignment and capability. Extensive experiments demonstrate that our framework empowers open-source models across multiple scales, enabling them to achieve new state-of-the-art performance on the major deep research benchmark. This work provides a scalable and effective pathway for advancing open-source LLMs without relying on proprietary data or models.
| 2026-01-07
| 2026-01-08
|
[
"cs.CL",
"cs.AI"
] |
Yi Yao, He Zhu, Piaohong Wang, Jincheng Ren, Xinlong Yang, Qianben Chen, Xiaowan Li, Dingfeng Shi, Jiaxian Li, Qiexiang Wang, Sinuo Wang, Xinpeng Liu, Jiaqi Wu, Minghao Liu, Wangchunshu Zhou
|
2601.03839
|
Logic Tensor Network-Enhanced Generative Adversarial Network
|
In this paper, we introduce Logic Tensor Network-Enhanced Generative Adversarial Network (LTN-GAN), a novel framework that enhances Generative Adversarial Networks (GANs) by incorporating Logic Tensor Networks (LTNs) to enforce domain-specific logical constraints during the sample generation process. Although GANs have shown remarkable success in generating realistic data, they often lack mechanisms to incorporate prior knowledge or enforce logical consistency, limiting their applicability in domains requiring rule adherence. LTNs provide a principled way to integrate first-order logic with neural networks, enabling models to reason over and satisfy logical constraints. By combining the strengths of GANs for realistic data synthesis with LTNs for logical reasoning, we gain valuable insights into how logical constraints influence the generative process while improving both the diversity and logical consistency of the generated samples. We evaluate LTN-GAN across multiple datasets, including synthetic datasets (gaussian, grid, rings) and the MNIST dataset, demonstrating that our model significantly outperforms traditional GANs in terms of adherence to predefined logical constraints while maintaining the quality and diversity of generated samples. This work highlights the potential of neuro-symbolic approaches to enhance generative modeling in knowledge-intensive domains.
| 2026-01-07
| 2026-01-08
|
[
"cs.LG",
"cs.AI",
"cs.LO"
] |
Nijesh Upreti, Vaishak Belle
|
2510.04514
|
ChartAgent: A Multimodal Agent for Visually Grounded Reasoning in Complex Chart Question Answering
|
Recent multimodal LLMs have shown promise in chart-based visual question answering, but their performance declines sharply on unannotated charts-those requiring precise visual interpretation rather than relying on textual shortcuts. To address this, we introduce ChartAgent, a novel agentic framework that explicitly performs visual reasoning directly within the chart's spatial domain. Unlike textual chain-of-thought reasoning, ChartAgent iteratively decomposes queries into visual subtasks and actively manipulates and interacts with chart images through specialized actions such as drawing annotations, cropping regions (e.g., segmenting pie slices, isolating bars), and localizing axes, using a library of chart-specific vision tools to fulfill each subtask. This iterative reasoning process closely mirrors human cognitive strategies for chart comprehension. ChartAgent achieves state-of-the-art accuracy on the ChartBench and ChartX benchmarks, surpassing prior methods by up to 16.07% absolute gain overall and 17.31% on unannotated, numerically intensive queries. Furthermore, our analyses show that ChartAgent is (a) effective across diverse chart types, (b) achieves the highest scores across varying visual and reasoning complexity levels, and (c) serves as a plug-and-play framework that boosts performance across diverse underlying LLMs. Our work is among the first to demonstrate visually grounded reasoning for chart understanding using tool-augmented multimodal agents.
| 2026-01-07
| 2026-01-08
|
[
"cs.AI",
"cs.CE",
"cs.CL",
"cs.CV",
"stat.ME"
] |
Rachneet Kaur, Nishan Srishankar, Zhen Zeng, Sumitra Ganesh, Manuela Veloso
|
2506.14435
|
MoTE: Mixture of Ternary Experts for Memory-efficient Large Multimodal Models
|
Large multimodal Mixture-of-Experts (MoEs) effectively scale the model size to boost performance while maintaining fixed active parameters. However, previous works primarily utilized full-precision experts during sparse up-cycling. Despite they show superior performance on end tasks, the large amount of experts introduces higher memory footprint, which poses significant challenges for the deployment on edge devices. In this work, we propose MoTE, a scalable and memory-efficient approach to train Mixture-of-Ternary-Experts models from dense checkpoint. Instead of training fewer high-precision experts, we propose to train more low-precision experts during up-cycling. Specifically, we use the pre-trained FFN as a shared expert and train ternary routed experts with parameters in {-1, 0, 1}. Extensive experiments show that our approach has promising scaling trend along model size. MoTE achieves comparable performance to full-precision baseline MoE-LLaVA while offering lower memory footprint. Furthermore, our approach is compatible with post-training quantization methods and the advantage further amplifies when memory-constraint goes lower. Given the same amount of expert memory footprint of 3.4GB and combined with post-training quantization, MoTE outperforms MoE-LLaVA by a gain of 4.3% average accuracy on end tasks, demonstrating its effectiveness and potential for memory-constrained devices.
| 2026-01-07
| 2026-01-08
|
[
"cs.CV",
"cs.LG"
] |
Hongyu Wang, Jiayu Xu, Ruiping Wang, Yan Feng, Yitao Zhai, Peng Pei, Xunliang Cai, Xilin Chen
|
2601.03721
|
Liouville theorems and gradient estimates of a nonlinear elliptic equation for the V-Laplacian
|
In this paper we establish gradient estimates for positive solutions to the nonlinear elliptic equation $$Î_{V}u^{m}+μ(x)u+p(x)u^α=0 , \quad m>1$$on any smooth metric measure space whose $k$-Bakry-Ãmery curvature is bounded from below by $-(k-1)K$ with $K \geq 0$. Additionally, we obtain related Liouville theorems and Harnack inequalities. We partially extend conclusions of Wang, when $V=0$, $μ=0$ the equation becomes $Îu^{m}+p(x)u^α=0$. And $V=f$, $μ=c, p=0 $, the equation becomes $Î_{f}u^{m}+cu=0 $.
| 2026-01-07
| 2026-01-08
|
[
"math.AP",
"math.DG"
] |
Yike Jia
|
2601.03922
|
Integration and Resource Estimation of Cryoelectronics for Superconducting Fault-Tolerant Quantum Computers
|
Scaling superconducting quantum computers to the fault-tolerant regime calls for a commensurate scaling of the classical control and readout stack. Today's systems largely rely on room-temperature, rack-based instrumentation connected to dilution-refrigerator cryostats through many coaxial cables. Looking ahead, superconducting fault-tolerant quantum computers (FTQCs) will likely adopt a heterogeneous quantum-classical architecture that places selected electronics at cryogenic stages -- for example, cryo-CMOS at 4~K and superconducting digital logic at 4~K and/or mK stages -- to curb wiring and thermal-load overheads. This review distills key requirements, surveys representative room-temperature and cryogenic approaches, and provides a transparent first-order accounting framework for cryoelectronics. Using an RSA-2048-scale benchmark as a concrete reference point, we illustrate how scaling targets motivate constraints on multiplexing and stage-wise cryogenic power, and discuss implications for functional partitioning across room-temperature electronics, cryo-CMOS, and superconducting logic.
| 2026-01-07
| 2026-01-08
|
[
"quant-ph",
"cond-mat.mes-hall",
"cond-mat.supr-con",
"physics.app-ph"
] |
Shiro Kawabata
|
2601.03482
|
Personalization of Large Foundation Models for Health Interventions
|
Large foundation models (LFMs) transform healthcare AI in prevention, diagnostics, and treatment. However, whether LFMs can provide truly personalized treatment recommendations remains an open question. Recent research has revealed multiple challenges for personalization, including the fundamental generalizability paradox: models achieving high accuracy in one clinical study perform at chance level in others, demonstrating that personalization and external validity exist in tension. This exemplifies broader contradictions in AI-driven healthcare: the privacy-performance paradox, scale-specificity paradox, and the automation-empathy paradox. As another challenge, the degree of causal understanding required for personalized recommendations, as opposed to mere predictive capacities of LFMs, remains an open question. N-of-1 trials -- crossover self-experiments and the gold standard for individual causal inference in personalized medicine -- resolve these tensions by providing within-person causal evidence while preserving privacy through local experimentation. Despite their impressive capabilities, this paper argues that LFMs cannot replace N-of-1 trials. We argue that LFMs and N-of-1 trials are complementary: LFMs excel at rapid hypothesis generation from population patterns using multimodal data, while N-of-1 trials excel at causal validation for a given individual. We propose a hybrid framework that combines the strengths of both to enable personalization and navigate the identified paradoxes: LFMs generate ranked intervention candidates with uncertainty estimates, which trigger subsequent N-of-1 trials. Clarifying the boundary between prediction and causation and explicitly addressing the paradoxical tensions are essential for responsible AI integration in personalized medicine.
| 2026-01-07
| 2026-01-08
|
[
"cs.AI",
"cs.LG",
"stat.AP"
] |
Stefan Konigorski, Johannes E. Vedder, Babajide Alamu Owoyele, Ä°brahim Ãzkan
|
2601.03725
|
EDCO: Dynamic Curriculum Orchestration for Domain-specific Large Language Model Fine-tuning
|
Domain-specific large language models (LLMs), typically developed by fine-tuning a pre-trained general-purpose LLM on specialized datasets, represent a significant advancement in applied AI. A common strategy in LLM fine-tuning is curriculum learning, which pre-orders training samples based on metrics like difficulty to improve learning efficiency compared to a random sampling strategy. However, most existing methods for LLM fine-tuning rely on a static curriculum, designed prior to training, which lacks adaptability to the model's evolving needs during fine-tuning. To address this, we propose EDCO, a novel framework based on two key concepts: inference entropy and dynamic curriculum orchestration. Inspired by recent findings that maintaining high answer entropy benefits long-term reasoning gains, EDCO prioritizes samples with high inference entropy in a continuously adapted curriculum. EDCO integrates three core components: an efficient entropy estimator that uses prefix tokens to approximate full-sequence entropy, an entropy-based curriculum generator that selects data points with the highest inference entropy, and an LLM trainer that optimizes the model on the selected curriculum. Comprehensive experiments in communication, medicine and law domains, EDCO outperforms traditional curriculum strategies for fine-tuning Qwen3-4B and Llama3.2-3B models under supervised and reinforcement learning settings. Furthermore, the proposed efficient entropy estimation reduces computational time by 83.5% while maintaining high accuracy.
| 2026-01-07
| 2026-01-08
|
[
"cs.LG"
] |
Jing-Cheng Pang, Liu Sun, Chang Zhou, Xian Tang, Haichuan Ma, Kun Jiang, Jianlong Wang, Kai Zhang, Sijie Wu, Haoran Cai, Chenwei Wu, Xubin Li, Xin Chen
|
2510.03959
|
Operational early warning of thunderstorm-driven power outages from open data: a two-stage machine learning approach
|
Thunderstorm-driven power outages are difficult to predict because most storms do not cause damage, convective processes occur rapidly and chaotically, and the available public data are noisy and incomplete. Severe convective storms now account for a large and rising share of U.S. weather losses, yet thunderstorm-induced outages remain understudied. We develop a 48-hour early-warning model for summer thunderstorm-related outages in Michigan using only open-source outage (EAGLE-I) and weather (METAR) data. Relative to prior work, we (i) rely solely on public data, (ii) preserve convective extremes from a sparse station network via parameter-specific kriging and causal spatiotemporal features, and (iii) use a multi-level LSTM-based architecture evaluated on event-centric peak metrics. The pipeline builds rolling and k-NN inverse-distance aggregates to capture moisture advection, wind shifts, and pressure drops. A two-stage design uses a logistic gate followed by a long short-term memory (LSTM) regressor to filter routine periods and limit noise exposure. Evaluation focuses on state-level peaks of at least 50,000 customers without power, using hits, misses, false alarms, and peak-conditional MASE (cMASE) within 48-hour windows, with uncertainty quantified by block bootstrapping. On the test sample, the Two-Stage model detects more peaks with only one additional false alarm and reduces cMASE near peaks, providing event-focused early warnings without the utility-specific data.
| 2026-01-07
| 2026-01-08
|
[
"cs.LG"
] |
Iryna Stanishevska, Seth Guikema
|
2601.03738
|
A glimpse into the Ultrametric spectrum
|
The non-relativistic string spectrum is built from integer-spaced energy quanta in such a way that the high-temperature asymptotics, via the Hardy-Ramanujan formula for integer partitions, reduces to standard two-dimensional thermodynamics. Here we explore deformed realizations of this behavior motivated by $p$-adic string theory and Lorentzian versions thereof with a non-trivial spectrum. We study the microstate scaling that results on associating quantum harmonic oscillators to the normal modes of tree-graphs rather than string graphs and observe that Hardy-Ramanujan scaling is not realized. But by computing the eigenvalues of the derivative operator on the $p$-adic circle and by determining the eigenspectrum of the Neumann-to-Dirichlet operator, we uncover a spectrum of exponentially growing energies but with exponentially growing degeneracies balanced in such a way that Hardy-Ramanujan scaling is realized, but modulated with log-periodic fluctuations.
| 2026-01-07
| 2026-01-08
|
[
"hep-th"
] |
An Huang, Christian B. Jepsen
|
2601.02075
|
MDAgent2: Large Language Model for Code Generation and Knowledge Q&A in Molecular Dynamics
|
Molecular dynamics (MD) simulations are essential for understanding atomic-scale behaviors in materials science, yet writing LAMMPS scripts remains highly specialized and time-consuming tasks. Although LLMs show promise in code generation and domain-specific question answering, their performance in MD scenarios is limited by scarce domain data, the high deployment cost of state-of-the-art LLMs, and low code executability. Building upon our prior MDAgent, we present MDAgent2, the first end-to-end framework capable of performing both knowledge Q&A and code generation within the MD domain. We construct a domain-specific data-construction pipeline that yields three high-quality datasets spanning MD knowledge, question answering, and code generation. Based on these datasets, we adopt a three stage post-training strategy--continued pre-training (CPT), supervised fine-tuning (SFT), and reinforcement learning (RL)--to train two domain-adapted models, MD-Instruct and MD-Code. Furthermore, we introduce MD-GRPO, a closed-loop RL method that leverages simulation outcomes as reward signals and recycles low-reward trajectories for continual refinement. We further build MDAgent2-RUNTIME, a deployable multi-agent system that integrates code generation, execution, evaluation, and self-correction. Together with MD-EvalBench proposed in this work, the first benchmark for LAMMPS code generation and question answering, our models and system achieve performance surpassing several strong baselines.This work systematically demonstrates the adaptability and generalization capability of large language models in industrial simulation tasks, laying a methodological foundation for automatic code generation in AI for Science and industrial-scale simulations. URL: https://github.com/FredericVAN/PKU_MDAgent2
| 2026-01-07
| 2026-01-08
|
[
"cs.CE",
"cs.LG"
] |
Zhuofan Shi, Hubao A, Yufei Shao, Dongliang Huang, Hongxu An, Chunxiao Xin, Haiyang Shen, Zhenyu Wang, Yunshan Na, Gang Huang, Xiang Jing
|
2601.03552
|
From Risk Perception to Behavior Large Language Models-Based Simulation of Pandemic Prevention Behaviors
|
Individual prevention behaviors are a primary line of defense during the early stages of novel infectious disease outbreaks, yet their adoption is heterogeneous and difficult to forecast-especially when empirical data are scarce and epidemic-policy contexts evolve rapidly. To address this gap, we develop an LLM-based prevention-behavior simulation framework that couples (i) a static module for behavior-intensity prediction under a specified external context and (ii) a dynamic module that updates residents' perceived risk over time and propagates these updates into behavior evolution. The model is implemented via structured prompt engineering in a first-person perspective and is evaluated against two rounds of survey data from Beijing residents (R1: December 2020; R2: August 2021) under progressively realistic data-availability settings: zero-shot, few-shot, and cross-context transfer. Using Kolmogorov-Smirnov tests to compare simulated and observed behavior distributions (p > 0.001 as the validity criterion), the framework demonstrates robust performance and improves with limited reference examples; reported predictive accuracy increases from 72.7% (zero-shot) to 81.8% (few-shot), and remains high at 77.8% under transfer to novel contexts. We further apply the framework to simulate behavior changes during China's December 2022 policy relaxation and to stress-test behavioral responses across 120 systematically varied epidemic conditions (R0, CFR, and control-measure tiers). Results indicate broad behavioral loosening under relaxation but a distinctive counter-trend increase in drain-related disinfection, highlighting how low-cost, low-friction behaviors may persist or intensify even when external constraints recede-raising a potential environmental tradeoff.
| 2026-01-07
| 2026-01-08
|
[
"cs.SI"
] |
Lujia Bo, Mingxuan Chen, Youduo Chen, Xiaofan Gui, Jiang Bian, Chunyan Wang, Yi Liu
|
2601.03493
|
Submodular Evaluation Subset Selection in Automatic Prompt Optimization
|
Automatic prompt optimization reduces manual prompt engineering, but relies on task performance measured on a small, often randomly sampled evaluation subset as its main source of feedback signal. Despite this, how to select that evaluation subset is usually treated as an implementation detail. We study evaluation subset selection for prompt optimization from a principled perspective and propose SESS, a submodular evaluation subset selection method. We frame selection as maximizing an objective set function and show that, under mild conditions, it is monotone and submodular, enabling greedy selection with theoretical guarantees. Across GSM8K, MATH, and GPQA-Diamond, submodularly selected evaluation subsets can yield better optimized prompts than random or heuristic baselines.
| 2026-01-07
| 2026-01-08
|
[
"cs.CL",
"cs.AI"
] |
Jinming Nian, Zhiyuan Peng, Hongwei Shang, Dae Hoon Park, Yi Fang
|
2601.03510
|
G2P: Gaussian-to-Point Attribute Alignment for Boundary-Aware 3D Semantic Segmentation
|
Semantic segmentation on point clouds is critical for 3D scene understanding. However, sparse and irregular point distributions provide limited appearance evidence, making geometry-only features insufficient to distinguish objects with similar shapes but distinct appearances (e.g., color, texture, material). We propose Gaussian-to-Point (G2P), which transfers appearance-aware attributes from 3D Gaussian Splatting to point clouds for more discriminative and appearance-consistent segmentation. Our G2P address the misalignment between optimized Gaussians and original point geometry by establishing point-wise correspondences. By leveraging Gaussian opacity attributes, we resolve the geometric ambiguity that limits existing models. Additionally, Gaussian scale attributes enable precise boundary localization in complex 3D scenes. Extensive experiments demonstrate that our approach achieves superior performance on standard benchmarks and shows significant improvements on geometrically challenging classes, all without any 2D or language supervision.
| 2026-01-07
| 2026-01-08
|
[
"cs.CV"
] |
Hojun Song, Chae-yeong Song, Jeong-hun Hong, Chaewon Moon, Dong-hwi Kim, Gahyeon Kim, Soo Ye Kim, Yiyi Liao, Jaehyup Lee, Sang-hyo Park
|
2601.02953
|
Spectral and Phase Structure of a QCD-Inspired Unitary Matrix Model with Fisher-Hartwig Singularities
|
We investigate the large $N$ limit of a complex action unitary matrix model with Fisher-Hartwig singularities, motivated by QCD-inspired models with complexified potentials. We show that the model exhibits multiple ungapped phases and a single gapped phase. The phases are characterized by Fisher-Hartwig singularities in the complex plane. We show that the phase transitions are third order, with transitions between ungapped phases forbidden. We also briefly discuss the implications for the QCD phase diagram at the end.
| 2026-01-07
| 2026-01-08
|
[
"hep-th"
] |
Anuj Malik, Anees Ahmed
|
2511.06095
|
Characterizing all $K_4$-free well-edge-dominated graphs of girth 3
|
Given a graph $G$, a set $F$ of edges is an edge dominating set if all edges in $G$ are either in $F$ or adjacent to an edge in $F$. $G$ is said to be well-edge-dominated if every minimal edge dominating set is also minimum. In 2022, it was proven that there are precisely three nonbipartite, well-edge-dominated graphs with girth at least four. Then in 2025, a characterization of all well-edge-dominated graphs containing exactly one triangle was found. In this paper, we characterize all well-edge-dominated graphs that contain a triangle and yet are $K_4$-free.
| 2026-01-07
| 2026-01-08
|
[
"math.CO"
] |
Sarah E. Anderson, Kirsti Kuenzel
|
2601.03925
|
Polarization rotation through differential transmission in refractive CMB telescopes identified using a hybrid physical optics method
|
We identify a polarization rotation systematic in the far field beams of refractive cosmic microwave background (CMB) telescopes caused by differential transmission in anti-reflection (AR) coatings of optical elements. This systematic was identified following the development of a hybrid physical optics method that incorporates full-wave electromagnetic simulations of AR coatings to model the full polarization response of refractive systems. Applying this method to a two-lens CMB telescope with non-ideal AR coating, we show that polarization-dependent transmission can produce a rotation of the far-field polarization angle that varies across the focal plane with a typical amplitude of 0.05-0.5 degrees. If ignored in analysis, this effect can produce temperature to polarization leakage and Stokes Q/U mixing.
| 2026-01-07
| 2026-01-08
|
[
"astro-ph.IM",
"astro-ph.CO",
"physics.optics"
] |
Xiaodong Ren, Rustam Balafendiev, Jon E. Gudmundsson
|
2601.03643
|
On $k$-connectivity oracles in $k$-connected graphs
|
A $k$-connectivity oracle for a graph $G=(V,E)$ is a data structure that given $s,t \in V$ determines whether there are at least $k+1$ internally disjoint $st$-paths in $G$. For undirected graphs, Pettie, Saranurak & Yin [STOC 2022, pp. 151-161] proved that any $k$-connectivity oracle requires $Ω(kn)$ bits of space. They asked whether $Ω(kn)$ bits are still necessary if $G$ is $k$-connected. We will show by a very simple proof that this is so even if $G$ is $k$-connected, answering this open question.
| 2026-01-07
| 2026-01-08
|
[
"cs.DS"
] |
Zeev Nutov
|
2601.04184
|
Transforming Video Subjective Testing with Training, Engagement, and Real-Time Feedback
|
Subjective video quality assessment is crucial for optimizing streaming and compression, yet traditional protocols face limitations in capturing nuanced perceptual differences and ensuring reliable user input. We propose an integrated framework that enhances rater training, enforces attention through real-time scoring, and streamlines pairwise comparisons to recover quality scores with fewer comparisons. Participants first undergo an automated training quiz to learn key video quality indicators (e.g., compression artifacts) and verify their readiness. During the test, a real-time attention scoring mechanism, using "golden" video pairs, monitors and reinforces rater focus by applying penalties for lapses. An efficient chain-based pairwise comparison procedure is then employed, yielding quality scores in Just-Objectionable-Differences (JOD) units. Experiments comparing three groups (no training, training without feedback, and training with feedback) with 80 participants demonstrate that training-quiz significantly improves data quality in terms of golden unit accuracy and reduces tie rate, while real-time feedback further improves data quality and yields the most monotonic quality ratings. The new training, quiz, testing with feedback, 3-phase approach can significantly reduce the non-monotonic cases on the high quality part of the R-Q curve where normal viewer typically prefer the slightly compressed less-grainy content and help train a better objective video quality metric.
| 2026-01-07
| 2026-01-08
|
[
"cs.MM"
] |
Kumar Rahul, Sriram Sethuraman, Andrew Segall, Yixu Chen
|
2601.03662
|
How Does the Thinking Step Influence Model Safety? An Entropy-based Safety Reminder for LRMs
|
Large Reasoning Models (LRMs) achieve remarkable success through explicit thinking steps, yet the thinking steps introduce a novel risk by potentially amplifying unsafe behaviors. Despite this vulnerability, conventional defense mechanisms remain ineffective as they overlook the unique reasoning dynamics of LRMs. In this work, we find that the emergence of safe-reminding phrases within thinking steps plays a pivotal role in ensuring LRM safety. Motivated by this finding, we propose SafeRemind, a decoding-time defense method that dynamically injects safe-reminding phrases into thinking steps. By leveraging entropy triggers to intervene at decision-locking points, SafeRemind redirects potentially harmful trajectories toward safer outcomes without requiring any parameter updates. Extensive evaluations across five LRMs and six benchmarks demonstrate that SafeRemind substantially enhances safety, achieving improvements of up to 45.5%p while preserving core reasoning utility.
| 2026-01-07
| 2026-01-08
|
[
"cs.AI"
] |
Su-Hyeon Kim, Hyundong Jin, Yejin Lee, Yo-Sub Han
|
2504.08532
|
Search for charged-lepton flavour violation in top quark interactions with an up-type quark, a muon, and a $Ï$ lepton in proton-proton collisions at $\sqrt{s}$ = 13 TeV
|
A search for charged-lepton flavour violation (CLFV) in top quark (t) production and decay is presented. The search uses proton-proton collision data corresponding to 138 fb$^{-1}$ collected with the CMS experiment at $\sqrt{s}$ = 13 TeV. The signal consists of the production of a single top quark via a CLFV interaction or top quark pair production followed by a CLFV decay. The analysis selects events containing a hadronically decaying $Ï$ lepton and a muon of opposite electric charge, as well as at least three jets, one of which is identified as originating from the fragmentation of a bottom quark. Machine learning classification techniques are used to distinguish signal from standard model background events. The results of this search are consistent with the standard model expectations. The upper limits at 95% confidence level on the branching fraction $\mathcal{B}$ for CLFV top quark decays to a muon, a $Ï$ lepton, and an up or a charm quark are set at $\mathcal{B}$(t $\to$ $μÏ$u) $\lt$ (0.04, 0.08, and 0.12) $\times$ 10$^{-6}$, and $\mathcal{B}$(t $\to$ $μÏ$c) $\lt$ (0.81, 1.71, and 2.05) $\times$ 10$^{-6}$ for scalar, vector, and tensor-like operators, respectively.
| 2026-01-07
| 2026-01-08
|
[
"hep-ex"
] |
CMS Collaboration
|
2601.03472
|
Kinetic theory of dilute granular gases with hard-core and inverse power-law potentials under uniform shear flow: comparison with simplified model
|
We develop a kinetic-theory framework to investigate the steady rheology of a dilute gas interacting via a repulsive potential under uniform shear flow. Starting from the Boltzmann equation with a restitution coefficient that depends on the impact velocity and potential strength, we derive evolution equations for the stress tensor based on Grad's moment expansion. The resulting expressions for the collisional rates and transport coefficients are fitted with simple analytical functions that capture their temperature dependence over a wide range of shear rates. Comparison with direct simulation Monte Carlo (DSMC) results shows excellent quantitative agreement for the shear stress, temperature anisotropy, and steady viscosity. We also analyze the velocity distribution functions, revealing that the system remains nearly Maxwellian even under strong shear.
| 2026-01-07
| 2026-01-08
|
[
"cond-mat.soft",
"cond-mat.stat-mech"
] |
Yuria Kobayashi, Makoto R. Kikuchi, Shunsuke Iizuka, Satoshi Takada
|
2410.21381
|
Unveiling the nature of SN 2022jli: The first double-peaked stripped-envelope supernova showing periodic undulations and dust emission at late times
|
We present optical and infrared observations from maximum light until around +800 days of supernova (SN) 2022jli, a peculiar stripped-envelope (SE) SN showing two maxima, each one with a peak luminosity of about $3 \times 10^{42}$ erg s$^{-1}$, separated by 50 days. The second maximum is followed by unprecedented periodic undulations with a period of $P \sim 12.5$ days. The spectra and the photometric evolution of the first maximum are consistent with the behaviour of a standard SE SN with an ejecta mass of $\sim 1.5$ $M_{\odot}$ and a radioactive $^{56}$Ni mass of $\sim 0.12$ $M_{\odot}$. The optical spectra after +400 days relative to the first maximum correspond to a standard SN Ic event, and at late times SN 2022jli exhibits a significant drop in the optical luminosity, implying that the physical phenomena that produced the secondary maximum have ceased to power the SN light curve. Among other potential scenarios, we discuss how the second maximum could be powered by a magnetar, while the light curve periodic undulations could be produced by accretion of material from a companion star onto the neutron star in a binary system. The near-infrared spectra shows clear first CO overtone emission from about +190 days after the first maximum, and it becomes undetected at +400 days. A significant near-infrared excess from hot dust emission is detected at +238 days, having been produced by either newly formed dust in the SN ejecta or a strong near-infrared dust echo. Depending on the assumptions of the dust composition, the estimated dust mass is $2-16 \times 10^{-4}$ $M_{\odot}$. The potential magnetar power of the second maximum can fit into a more general picture in which magnetars are the power source of SE super-luminous SNe, and could explain bumps, undulations, and late-time excess emission in SE SNe.
| 2026-01-07
| 2026-01-08
|
[
"astro-ph.HE",
"astro-ph.CO"
] |
Régis Cartier, Carlos Contreras, Maximilian Stritzinger, Mario Hamuy, Pilar Ruiz-Lapuente, Jose L. Prieto, Joseph P. Anderson, Aleksandar Cikota, Matthias Gerlach
|
2601.04103
|
Modeling the Effect of C/O Ratio on Complex Carbon Chemistry in Cold Molecular Clouds
|
Elemental abundances, which are often depleted with respect to the solar values, are important input parameters for kinetic models of interstellar chemistry. In particular, the amount of carbon relative to oxygen is known to have a strong effect on modeled abundances of many species. While previous studies have focused on comparison of modeled and observed abundances to constrain the C/O ratio, the effects of this parameter on the underlying chemistry have not been well-studied. We investigated the role of the C/O ratio on dark cloud chemistry using the NAUTILUS code and machine learning techniques for molecular representation. We find that modeled abundances are quite sensitive to the C/O ratio, especially for carbon-rich species such as carbon chains and polycyclic aromatic hydrocarbons (PAHs). CO and simple ice-phase species are found to be major carbon reservoirs under both oxygen-poor and oxygen-rich conditions. The appearance of C3H4 isomers as significant carbon reservoirs, even under oxygen-rich conditions, indicates the efficiency of gas-phase C3 formation followed by adsorption and grain-surface hydrogenation. Our model is not able to reproduce the observed, gas-phase C/H ratio of TMC-1 CP at the time of best fit with any C/O ratio between 0.1 and 3, suggesting that the modeled freeze-out of carbon-bearing molecules may be too rapid. Future investigations are needed to understand the reactivity of major carbon reservoirs and their conversion to complex organic molecules.
| 2026-01-07
| 2026-01-08
|
[
"astro-ph.GA"
] |
Alex N. Byrne, Christopher N. Shingledecker, Edwin A. Bergin, Martin S. Holdren, Gabi Wenzel, Ci Xue, Troy Van Voorhis, Brett A. McGuire
|
2601.03577
|
Variational Inference, Entropy, and Orthogonality: A Unified Theory of Mixture-of-Experts
|
Mixture-of-Experts models enable large language models to scale efficiently, as they only activate a subset of experts for each input. Their core mechanisms, Top-k routing and auxiliary load balancing, remain heuristic, however, lacking a cohesive theoretical underpinning to support them. To this end, we build the first unified theoretical framework that rigorously derives these practices as optimal sparse posterior approximation and prior regularization from a Bayesian perspective, while simultaneously framing them as mechanisms to minimize routing ambiguity and maximize channel capacity from an information-theoretic perspective. We also pinpoint the inherent combinatorial hardness of routing, defining it as the NP-hard sparse subset selection problem. We rigorously prove the existence of a "Coherence Barrier"; when expert representations exhibit high mutual coherence, greedy routing strategies theoretically fail to recover the optimal expert subset. Importantly, we formally verify that imposing geometric orthogonality in the expert feature space is sufficient to narrow the divide between the NP-hard global optimum and polynomial-time greedy approximation. Our comparative analyses confirm orthogonality regularization as the optimal engineering relaxation for large-scale models. Our work offers essential theoretical support and technical assurance for a deeper understanding and novel designs of MoE.
| 2026-01-07
| 2026-01-08
|
[
"cs.LG"
] |
Ye Su, Yong Liu
|
2509.18944
|
Easy estimates of Lyapunov exponents for random products of matrices
|
The problems that we consider in this paper are as follows. Let $A_1, \ldots, A_k$ be square matrices (over reals). Let $W=w(A_1, \ldots, A_k)$ be a random product of $n$ matrices. What is the expected absolute value of the largest (in the absolute value) entry in such a random product? What is the (maximal) Lyapunov exponent for a random matrix product like that? We give a complete answer to the first question. For the second question, we offer a very simple and efficient method to produce an upper bound on the Lyapunov exponent.
| 2026-01-07
| 2026-01-08
|
[
"math.GR",
"math.DS"
] |
Nadya Nabahi, Vladimir Shpilrain
|
2601.00347
|
Evidence for the suppression of the hybrid skin-topological effect by fragile topology
|
Topological insulators are well-known for their topological edge states, which are protected by the non-trivial bulk topology and characterized by gapless Wannier bands, a phenomenon known as the bulk-boundary correspondence. However, fragile topology challenged this concept, the Wannier bands are gapped, but the edge states still exist with similar protection. Previous studies on fragile topology have primarily focused on the spectral flow under twisted boundary conditions, but the discussion on the physical interpretation of the Wannier gap is limited. In this study, we introduce a bilayer breathing honeycomb lattice with spiral interlayer couplings inside the unit cell. As we increase the interlayer coupling strength, the Wannier gap increases monotonically and the bandgap first increases then decreases. After introducing a gain-loss domain wall, the hybrid skin-topological effect (HSTE) emerges, and the topological edge states under the periodic boundary condition (PBC) change into corner states under the open boundary condition (OBC) associated with the significant spectral difference. HSTE is suppressed as the interlayer coupling strength increases, the spectral difference between the two boundary conditions has an overall decreasing trend, which more closely mirrors the evolution of the inverse of the Wannier gap. Moreover, some of the corner states transform into edge states. Our work first provides evidence for the relation between fragile topology and HSTE, shedding new insights into the underlying mechanism of Non-Hermitian skin effect (NHSE).
| 2026-01-07
| 2026-01-08
|
[
"cond-mat.mes-hall"
] |
Tianrui Liu
|
2410.19007
|
MOND as a transformation between non-inertial reference frames via Sciama's interpretation of Mach's Principle
|
Moderhai Milgrom's Modified Newtonian Dynamics (MOND) correction to Newtonian gravity or inertia is shown to be equivalent to a more fundamental formulation considering a non-inertial local reference frame and the fixed background of the observable universe, in the spirit of Mach's principle. Both Newton's gravitational constant $G\sim c^2/(M_u/R_u)$ and Milgrom's MOND acceleration scale constant $a_0\sim GM_u/R_u^2$ are replaced by two varying, measurable, and cosmological quantities determined by the causally connected mass and size of the universe. They arise from an inverse and an inverse squared distance scalar fields of matter density, respectively. This Machian interpretation of MOND is invariant under global rescalings of mass, length, and time across all regimes and is free from fundamental constants and free parameters, except for the speed of light. Machian MOND satisfies the fundamental consequences of Mach's principle not featured in Newton's and Einstein's theories: the decrease of inertia of a body when masses are removed from its neighborhood, and in the absence of a cosmic background, rotational motion is undefined up to the speed of light. Consequently, Machian MOND provides the necessary limiting behavior to which any phenomenological non-linear theory of modified inertia or gravity that incorporates Mach's principle, in agreement with galaxy rotation curves, should reduce as an effective approximation.
| 2026-01-07
| 2026-01-08
|
[
"physics.gen-ph"
] |
Manuel Uruena Palomo
|
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.