arXiv:2607.13064v1 Announce Type: cross
Abstract: With the rapid development of low-altitude economy, the radiation environment safety of low-altitude aircraft such as drones and electric vertical take-off and landing aircraft has attracted increasing attention. Although the dense lower atmosphere traditionally serves as an effective shield against cosmic radiation, the shrinking feature sizes of modern integrated circuits greatly enhance their vulnerability to single-event effects (SEEs). This study quantitatively evaluates muon-induced SEE risks for low-altitude aircraft in various regions of China under both static cosmic-ray background and ground-level enhancement (GLE) events, aiming to provide critical guidance for the next-generation low-altitude aviation platforms.Using city-specific atmospheric models within the CORSIKA framework, we simulate atmospheric shower processes and obtain reliable energy spectra for low-energy muons (10-100 MeV). We also employ simulation data from other research groups to estimate muon-induced SEE cross sections for transistors at different process nodes, including bulk, FD-SOI, and FinFET technologies. By incorporating solar energetic particle spectra associated with GLE events, we assess muon-induced SEE risks under both static and GLE conditions. Our results show that under static conditions, flight control systems with 1 MB memory using advanced nodes below 45 nm and bulk transistors face non-negligible muon-induced SEE risks in all Chinese cities. In contrast, systems with FD-SOI transistors can effectively mitigate these risks. For large-memory systems (1 GB), redundancy or other hardening measures are essential regardless of the process technology. Regarding GLE events, we introduce the concept of muon hazard levels to evaluate regional risk variations. During GLEs, the increase in muon-induced SEE risk is negligible in mid-to-low latitude regions but becomes significant at high latitudes.
Science Journals
arXiv:2607.13204v1 Announce Type: cross
Abstract: Medical imaging systems such as CT, MRI, PET, and SPECT do not directly acquire images. Instead, they measure physical signals that encode anatomical or physiological information, and image reconstruction recovers the underlying image by solving an inverse problem. Although these imaging modalities are governed by different imaging physics, they share a common computational framework that naturally connects medical physics, linear algebra, probability, numerical optimization, and efficient computing. As medical imaging systems acquire increasingly large and higher-dimensional datasets, image reconstruction has become one of the primary computational bottlenecks in modern medical imaging. Advanced reconstruction methods, including analytical reconstruction, iterative optimization, and statistical model-based reconstruction, substantially improve image quality while reducing radiation dose or scan time, but at significantly increased computational cost. Efficient computing has therefore become essential for achieving clinically practical reconstruction times. This chapter presents a unified computational perspective on medical image acquisition and reconstruction across CT, MRI, PET, and SPECT. It first reviews the imaging physics and data acquisition process for each modality and derives a generalized mathematical framework for image reconstruction. Building on this framework, the chapter discusses analytical, iterative, and statistical reconstruction methods together with their computational characteristics. Finally, it examines efficient computing considerations, including optimization algorithms, physics-aware forward operators, memory-efficient implementations, and parallel computing strategies. Together, these topics demonstrate how the integration of imaging physics, mathematical modeling, and efficient computing enables accurate and scalable medical image reconstruction.
arXiv:2607.13408v1 Announce Type: cross
Abstract: Recent text-to-audio models generate high-quality audio, but often fail to follow instructions involving multiple sound events and temporal order. This gap arises because existing evaluation and training signals mainly emphasize global similarity or perceptual quality, with limited supervision on instruction-level correctness. We propose an instruction-level framework that uses audio-aware large language models (ALLMs) as fine-grained judges to verify target event presence and temporal relations in generated audio. After validating ALLM judgments on benchmarks and through human verification, we use their feedback to construct preference pairs for direct preference optimization. We further introduce S3Bench, a narrative benchmark for evaluating multi-event temporal instruction following. Experiments show that our method improves event completeness, temporal ordering, and joint instruction-following accuracy across existing benchmarks and S3Bench, while maintaining audio quality.
arXiv:2607.13395v1 Announce Type: new
Abstract: The pursuit of autonomously self-improving models has attracted growing interest in the era of large-scale foundation models. Drawing inspiration from the concept of "enlightenment" or "aha moment" in human brain, we hypothesize that large models exhibit an analogous enlightenment phenomenon-a latent capacity for sudden capability boost. Then, we propose Enlightenment, a novel training-free post-tuning paradigm for large-scale models. Our approach modifies shortcuts for key modules/layers without weight updates, while existing training-free ones predominantly manipulate attention weights. We introduce two architecture-specific instantiations: i) For large language models, we propose attention head-mixing shortcuts that recalibrate attention weights by linking the initial attention head's output to all other target heads, modulated by an adaptive scaling factor initialization strategy. ii) For vision-language models, we apply a lightweight scalar-modulated factor to residual connections in the decoder layers, regulating information flow. Extensive experiments show that Enlightenment efficiently unlocks the latent potential of pre-trained networks, yielding remarkable performance improvements across diverse benchmarks and models.
arXiv:2607.13476v1 Announce Type: cross
Abstract: Quantum process tomography, the task of learning an unknown quantum channel from black-box access, is a central problem in quantum information. In this setting, protocols with quantum memory can coherently store and jointly process quantum information obtained from multiple channel uses, whereas protocols without quantum memory must measure after each use and retain only a classical transcript of the measurement outcomes. A fundamental open question is whether quantum memory provides a query-complexity advantage even when protocols without quantum memory may adapt their experiments based on all previous outcomes with unbounded classical computational power. In this work, we show that it does. We determine the optimal query complexity of quantum process tomography without quantum memory up to a constant factor to be $\Theta(d_{\mathrm{in}}^3 d_{\mathrm{out}}^3/\varepsilon^2)$, where $d_{\mathrm{in}}$ and $d_{\mathrm{out}}$ are the channel input and output dimensions, respectively, and $\varepsilon$ is the target diamond-norm accuracy. More precisely, we prove that any incoherent protocol for this task, including adaptive protocols, requires $\Omega(d_{\mathrm{in}}^3 d_{\mathrm{out}}^3/\varepsilon^2)$ queries, even when each channel use may be assisted by arbitrary fresh ancilla, and we present a non-adaptive, ancilla-free incoherent protocol achieving the matching upper bound $O(d_{\mathrm{in}}^3 d_{\mathrm{out}}^3/\varepsilon^2)$. Our results thereby generalize the optimal sample-complexity bounds for single-copy state tomography, recovered as the special case $d_{\mathrm{in}}=1$. By contrast, coherent protocols with quantum memory achieve query complexity $\Theta(d_{\mathrm{in}}^2 d_{\mathrm{out}}^2/\varepsilon^2)$. Hence, our results establish a rigorous learning separation between quantum process tomography with and without quantum memory.
arXiv:2607.13067v1 Announce Type: new
Abstract: Teleoperating remotely operated vehicles (ROVs) in flooded, cluttered infrastructure is fundamentally limited by narrow 2D egocentric views and subsea communication latency. We present a multimodal teleoperation architecture built on a ROS-Unity framework that decouples proactive spatial planning from reactive boundary avoidance. The system replaces static camera feeds with a Dynamic Adaptive Viewpoint System (DAVS), which uses continuous optimization and real-time 3D Gaussian Splatting (3DGS) to synthesize an occlusion-free exocentric viewpoint from onboard state estimation. To further reduce sensory workload, a torso-mounted vibrotactile suit maps local obstacle clearance to intuitive haptic proximity cues. The architecture was evaluated in a controlled human-subject study (N = 30) using a BlueROV2 navigating a complex simulated underwater facility. A 3 x 4 repeated-measures design compared three interaction modalities (Egocentric, Haptic, Exocentric) under four communication delays (0.0-1.0 s). Performance was quantified using behavioral measures and functional near-infrared spectroscopy (fNIRS) to assess task-evoked prefrontal activation. Results show that reactive haptic feedback improves path adherence under minimal delay, whereas the 3DGS-driven exocentric visualization provides superior resilience under severe latency (0.5-1.0 s), significantly outperforming the other modalities. fNIRS further revealed a cognitive disengagement effect: increasing latency during conventional egocentric teleoperation overloaded working memory and reduced prefrontal activation, whereas the proactive spatial context provided by DAVS sustained executive control. These findings demonstrate that spatially grounded, multimodal assistance can substantially improve operator performance and cognitive endurance during latency-degraded underwater teleoperation.
arXiv:2607.13425v1 Announce Type: new
Abstract: Learning effectively from limited data is critical in domains like security where labeled examples are scarce. Large language models (LLMs) have demonstrated some capabilities for data-efficient learning, especially through parameter-efficient adaptation methods, but continue to struggle when faced with few samples for difficult tasks. To meet this challenge, we propose Attention Head Reweighting (AHR), a data-efficient method that adapts LLMs to new text-classification tasks by learning only a single scalar per attention head. This drastically reduces the number of parameters that need to be learned by making use of the functional specialization of individual attention heads. Experiments on diverse open-source text classification datasets show that AHR can outperform standard baselines like LoRA when learning from limited samples, despite having 200-1000x fewer trainable parameters, as our AHR only modifies ~0.0001% of the model's parameters. In addition, our learned weights are easy to interpret and can be analyzed to better understand the mechanisms and attention heads responsible for in-context learning abilities in LLMs.
arXiv:2607.13430v1 Announce Type: new
Abstract: Translating complex biomedical data into patient-friendly narratives is central to modern biomedical informatics. This study presents a comparative analysis of training small language models (SLMs) in specialized biomedical datato-text generation tasks. We explore widely adopted post-training methods including supervised fine-tuning (SFT), direct preference optimization (DPO), odds ratio preference optimization (ORPO), and group relative policy optimization (GRPO) with Qwen-based SLMs on a medicine package leaflets dataset. To assess cross-dataset generalizability, we also curated drug label data from openFDA. We evaluate models using both standard lexical overlap metrics like ROUGE as well as semantic similarity measures. Across our experiments, the results show that (1) the aligned SLMs outperform proprietary models like GPT-5; (2) ORPO outperforms the SFTbaselines; (3) GRPO yields the most robust cross-dataset performance among the alignment methods tested as well as GPT-5.
arXiv:2607.13834v1 Announce Type: cross
Abstract: Scalable quantum computing is limited by the dense network of electrical interconnects linking cryogenic quantum processors to room-temperature control electronics. To overcome this bottleneck, considerable effort has focused on cryogenic CMOS electronics and microwave-to-optical transduction, aiming to reduce wiring complexity and thermal loading. Wireless interconnects have recently emerged as a promising complementary approach, yet their compatibility with superconducting quantum hardware remains largely unexplored. Here, we demonstrate the wireless excitation of a superconducting microwave resonator of the type routinely employed for qubit readout, operating at millikelvin temperatures inside a dilution refrigerator. By directly comparing wired and wireless operation within the same cryogenic environment, we show that wireless coupling preserves the intrinsic resonator response while revealing parasitic electromagnetic pathways arising from stray radiation within the cryostat enclosure. These results establish a framework for the co-design of wireless interconnects, cryogenic packaging and superconducting quantum hardware.
arXiv:2607.13990v1 Announce Type: cross
Abstract: We prove an effective-resistance bound for fixed-rank external-field measures. Let $d\ge2$ be an integer, let $m\in\{1,\ldots,d-1\}$. Let $w\in(0,+\infty)^d$, and let $\mathsf S$ be an $m$-element random subset of $[d]$ distributed according to the rank-$m$ external-field measure with weights $w$, i.e., \[\mathbb P(\mathsf S=S)=\frac{\prod_{i\in S}w_i}{e_m(w)},\qquad S\subseteq\{1,\dots,d\},\quad|S|=m,\] where \[e_m(w):=\sum_{\substack{T\subseteq\{1,\dots,d\}\\|T|=m}}\prod_{\ell\in T}w_\ell\] is the $m$th elementary symmetric polynomial in $w_1,\ldots,w_d$.
Let $X:=(X_1,\dots,X_d)^\top$ be its indicator vector, i.e., \[X_i=\mathbb I\{i\in\mathsf S\},\qquad i\in\{1,\dots,d\}.\]
Let $\Sigma:=\operatorname{Cov}(X)$, put $v_i:=\Sigma_{ii}$ for each $i\in\{1,\dots,d\}$, and let $\mathbf e_1,\ldots,\mathbf e_d$ denote the standard basis of $\mathbb R^d$.
Our main result is that, for every $i\ne j$, \[(\mathbf e_i-\mathbf e_j)^\top\Sigma^\dagger(\mathbf e_i-\mathbf e_j)\le\frac1{v_i}+\frac1{v_j},\] where $\Sigma^\dagger$ is the Moore-Penrose pseudoinverse of $\Sigma$. As a consequence, if \[v:=(v_1,\ldots,v_d)^\top,\qquad D:=\operatorname{diag}(v),\qquad V:=\sum_{i=1}^dv_i,\] then, as a corollary, we obtain \[\Sigma\succeq\frac12\left(D-\frac{vv^\top}{V}\right),\] which establishes a factor-two relaxation of the normalized covariance bound conjectured by Anari, Haqi, and Ma.
As a further corollary, combining our theorem with the recent framework of Anari, Haqi, and Ma yields a constant-stretch guarantee for correlated sampling on the hypersimplex without relying on the still-open normalized covariance conjecture assumed in their conditional result.
Our result improves the logarithmic-in-$k$ stretch bound of Naor, Raju, Shetty, Srinivasan, Valieva, and Wajc to a constant and resolves the open question posed in their work.
arXiv:2307.01412v4 Announce Type: replace
Abstract: The sliding suffix tree (Fiala \& Greene, 1989) is a suffix tree that is maintained for a sliding window $W_i = T[i..i+d-1]$ of size $d$ that shifts over an input text $T$ of length $n$ from left to right, for increasing $i = 1, \ldots, n-d+1$. It is known that the sliding suffix tree can be maintained in $O(n \log \sigma)$ time with $O(d)$ space, where $\sigma$ is the alphabet size. Updating the sliding suffix tree from $W_i = T[i..i+d-1]$ to $W_{i+1} = T[i+1..i+d]$ requires the following three major tasks: (1) Delete the leaf that represents the longest suffix $W_i$, (2) Insert new leaves that represent the suffixes of $W_{i+1}$ that appear exactly once in $W_{i+1}$, and (3) After the leaf deletion due to Task (1) and each leaf insertion due to Task (2), maintain the label $\langle \ell, r \rangle$ of every edge as a valid pair in the new window $W_{i+1}$, such that $i+1 \leq \ell \leq r \leq i+d$. In this paper, we present the first algorithm that performs Task (3) in $O(1)$ worst-case time per node deletion/insertion, which leads to another alternative to efficient sliding suffix tree construction. This is an improvement over the existing algorithms by Larsson (1996, 1999) and by Senft (2005) both of which can only perform Task (3) in $O(1)$ amortized time. Our key data structure is a non-trivial extension of leaf pointers, which were originally proposed by Brodnik and Jekovec (2018) for pattern matching with sliding suffix trees.
arXiv:2311.06563v3 Announce Type: replace
Abstract: We study \textsc{Monotone 3-Sat-$(\leq k,1)$}, a restricted variant of the \textsc{Satisfiability} problem where clauses consist of three variables and are monotone (every clause contains either only unnegated or only negated variables) with up to $k$ positive and exactly one negative occurrence per variable in the formula. We resolve a challenge posed by Darmann and D\"ocker (On simplified NP-complete variants of \textsc{Monotone} 3-\textsc{Sat}, Discrete Applied Mathematics 292:45--58, 2021) by proving that for~$k\in \{3,4\}$, the problem is trivial in the sense that every instance satisfying the given restrictions is satisfiable. This result closes the remaining gap in a dichotomy theorem: Triviality for $k\in \{1,2\}$ follows by a result by Tovey (A simplified NP-complete satisfiability problem, Discrete Applied Mathematics 8(1):85--89, 1984), while NP-completeness for~$k\geq 5$ was shown by Darmann and D\"ocker. To obtain our result, we introduce the notion of \emph{color structures} and show that a satisfying assignment can always be constructed in $\mathcal{O}(n \cdot m)$ time, where $n$ and $m$ denote the number of negative and positive clauses of the input formula, respectively.
arXiv:2311.10896v2 Announce Type: replace
Abstract: We prove and collect numerous explicit and computable results for the fractional Laplacian $(-\Delta)^s f(x)$ with $s>0$ as well as its whole space inverse, the Riesz potential, $(-\Delta)^{-s}f(x)$ with $s\in\left(0,\frac{1}{2}\right)$, subject to row-specific parameter and integrability conditions. Choices of $f(x)$ include weighted classical orthogonal polynomials such as the Legendre, Chebyshev, Jacobi, Laguerre and Hermite polynomials, or first and second kind Bessel functions with or without sinusoid weights. Some higher dimensional fractional Laplacians and Riesz potentials of generalized Zernike polynomials on the unit ball and its complement as well as whole space generalized Laguerre polynomials are also discussed. The aim of this paper is to aid in the continued development of numerical methods for problems involving the fractional Laplacian or the Riesz potential in bounded and unbounded domains -- both directly by providing useful basis or frame functions for spectral method approaches and indirectly by providing accessible ways to construct computable synthetic problems on which to test new numerical methods.
arXiv:2405.17366v3 Announce Type: replace
Abstract: We present a novel machine-learning (ML) approach (EM-GANSim) for real-time electromagnetic (EM) propagation that is used for wireless communication simulation in 3D indoor environments. Our approach uses a modified conditional Generative Adversarial Network (GAN) that incorporates encoded geometry and transmitter location while adhering to the electromagnetic propagation theory. The overall physically-inspired learning is able to predict the power distribution in 3D scenes, which is represented using heatmaps. We evaluated our method on 15 complex 3D indoor environments, with 4 additional scenarios later included in the results, showcasing the generalizability of the model across diverse conditions. Our overall accuracy is comparable to ray tracing-based EM simulation, as evidenced by lower mean squared error values. Furthermore, our GAN-based method drastically reduces the computation time, achieving a 5X speedup on complex benchmarks. In practice, it can compute the signal strength in a few milliseconds on any location in 3D indoor environments. We also present a large dataset of 3D models and EM ray tracing-simulated heatmaps. To the best of our knowledge, EM-GANSim is the first real-time algorithm for EM simulation in complex 3D indoor environments. We plan to release the code and the dataset.
arXiv:2408.04118v5 Announce Type: replace
Abstract: Much energy has been devoted to developing a matroid's computational properties, yet parallel algorithm design for matroid optimization seems less understood. Specifically, the current state of the art is a folklore reduction from optimization to the search based on methods originating in [KUW88]. However, while this reduction adds only constant overhead in terms of \emph{adaptive complexity}, it imposes a high cost in \emph{query complexity}. In response, we present a new reduction from optimization to search within the class of \emph{binary matroids} which, when $n$ and $r$ take the size of the ground set and matroid rank respectively, implies a novel optimization algorithm terminating in $\mathcal{O}(\sqrt{n}\cdot\log r)$ parallel rounds using only $\mathcal{O}(rn\cdot\log r)$ independence queries. This is a significant improvement in query complexity when the matroid is sparse, meaning $r \ll n$, while trading off only a logarithmic factor of the rank in the adaptive complexity. At a technical level, our method begins by observing that a basis is optimal if and only if it is the set of points of minimum weight in any cocircuit. Importantly, this certificate reveals that simultaneous tests for \emph{local optimality} in cocircuits is a general paradigm for parallel matroid optimization. By combining this idea with connections between bases and cocircuits we obtain our reduction, whose efficiency follows by analyzing the lattice of flats. A primary goal of our study is initiating a finer understanding of parallel matroid optimization. And so, since many of our techniques begin with observations about general matroids and their flats, we hope that our efforts aid the future design of parallel matroid algorithms and applications of lattice theory thereof.
arXiv:2411.19142v3 Announce Type: replace
Abstract: The General Data Protection Regulation (GDPR) is considered as the benchmark in the European Union (EU) for privacy and data protection standards. Since before its entry into force in 2018, substantial research has been conducted in the software engineering (SE) literature investigating the elicitation, representation, and verification of GDPR privacy requirements. Software systems deployed anywhere in the world must comply with GDPR as long as they handle personal data of EU residents. Mobile applications (apps) are no different in that regard. With the growing pervasiveness of mobile apps and their increasing demand for personal data, privacy concerns have acquired further interest within the SE community. Despite the extensive literature on GDPR-relevant privacy concerns in mobile apps, there is no secondary study that describes, analyzes, and categorizes the current focus. Research gaps and persistent challenges are thus left unnoticed. This article aims to provide a comprehensive overview of the existing research on GDPR privacy concerns in the context of mobile apps. To do so, we conducted a systematic literature review of 60 primary studies. Our findings show that existing studies predominantly address three key GDPR-related privacy concerns: (i) the direct collection of personal data from users, (ii) the sharing of personal data with external entities (e.g., third parties) beyond the mobile apps, and (iii) the analysis of user consent as a legal basis for collecting personal data. Our study highlighted research gaps, calling for further research to better understand: (i) the indirect collection of personal data, e.g., data exposed to mobile apps through, e.g., permission requests, (ii) the impact of legal bases beyond consent and how they may affect the development of mobile apps, and (iii) the required implementation details pertinent to data subject rights.
arXiv:2412.19791v3 Announce Type: replace
Abstract: This paper is concerned with high-order numerical methods for hyperbolic systems of balance laws. Such methods are typically based on high-order piecewise polynomial reconstructions (interpolations) of the computed discrete quantities. However, such reconstructions (interpolations) may be oscillatory unless the reconstruction (interpolation) procedure is applied to the local characteristic variables via the local characteristic decomposition (LCD). Another challenge in designing accurate and stable high-order schemes is related to enforcing a delicate balance between the fluxes, sources, and nonconservative product terms: a good scheme should be well-balanced (WB) in the sense that it should be capable of exactly preserving certain (physically relevant) steady states. One of the ways to ensure that the reconstruction (interpolation) preserves these steady states is to apply the reconstruction (interpolation) to the equilibrium variables, which are supposed to be constant at the steady states. To achieve this goal and to keep the reconstruction (interpolation) non-oscillatory, we introduce a new LCD of equilibrium variables. We apply the developed technique to the fifth-order Ai-WENO-Z interpolation implemented within the WB A-WENO framework recently introduced in [S. Chu, A. Kurganov, and R. Xin, Beijing J. of Pure and Appl. Math., 2 (2025), pp. 87--113], and illustrate its performance on a variety of numerical examples.
arXiv:2501.05396v4 Announce Type: replace
Abstract: Large language models (LLMs) are increasingly used in high-stakes decisions such as hiring and college admissions, making their social bias a critical concern. While LLMs are trained to refuse explicitly biased requests, bias can be leaked implicitly during LLM planning and reasoning process. As code becomes the primary medium for LLM internal logic-writing, we introduce FairCoder, a benchmark that frames decision-making as coding tasks to systematically probe LLM bias across employment, education, and healthcare domains, covering multiple fairness definitions. Considering that existing metrics may fail when LLMs frequently refuse the request, we propose FairScore, a metric that jointly captures refusal behavior and group-level outcome diversity. Experiments with a 1k-sample dataset on powerful LLMs reveal consistent and previously underexplored bias patterns, such as prioritizing applicants from high-income families in college admissions. Our findings highlight the risks of deploying LLMs as decision-making agents and provide a comprehensive evaluation framework for future research.
arXiv:2502.07780v4 Announce Type: replace
Abstract: Large Language Models (LLMs) have achieved significant success across various NLP tasks. However, their massive computational costs limit their widespread use, particularly in real-time applications. Structured pruning offers an effective solution by compressing models and directly providing end-to-end speed improvements, regardless of the hardware environment. Meanwhile, different components of the model exhibit varying sensitivities towards pruning, calling for non-uniform model compression. However, a pruning method should not only identify a capable substructure, but also account for post-compression training. To this end, we propose DarwinLM, a method for training-aware structured pruning. DarwinLM builds upon an evolutionary search process, generating multiple offspring models in each generation through mutation, and selecting the fittest for survival. To assess the effect of post-training, we incorporate a lightweight, multistep training process within the offspring population, progressively increasing the number of tokens and eliminating poorly performing models in each selection stage. We validate our method through extensive experiments on Llama-2-7B, Llama-3.1-8B and Qwen-2.5-14B-Instruct, achieving state-of-the-art performance for structured pruning. For instance, DarwinLM surpasses ShearedLlama while requiring 5x less training data during post-compression training. Code is at: https://github.com/IST-DASLab/DarwinLM
arXiv:2607.13452v1 Announce Type: new
Abstract: Incremental object detection (IOD) aims to extend detectors to new categories while retaining previously acquired knowledge. Existing methods often adopt a class incremental learning perspective, separating feature spaces to sharpen decision boundaries. However, this separation-oriented paradigm may overlook object symbiosis in detection, where co-occurrence and occlusion introduce spatial and semantic dependencies that benefit from shared representations. Ignoring these dependencies distorts the shared representations, exacerbates confusion between old and new classes, and accelerates catastrophic forgetting. To address this, we propose Symbiosis-Inspired Knowledge Distillation (SIKD), which explicitly leverages object symbiosis at two complementary levels. Spatial Symbiosis Distillation (SpSD) focuses on symbiotic regions where the old model responds with high overlap to objects in the new task. It preserves generalizable old class cues, suppresses class-specific bias and redundancy, and distills the refined evidence to the new model at matched spatial locations with slot-aligned supervision. Semantic Symbiosis Distillation (SeSD) maintains class level structure by forming confidence weighted prototypes for old classes and aligning their inter class soft ranks over the old class logits, which stabilizes the semantic topology during adaptation. Extensive experiments demonstrate the effectiveness and superiority of the proposed method.
arXiv:2607.13453v1 Announce Type: new
Abstract: Artificial Intelligence (AI), especially Generative AI (GenAI), adoption has increased in industries significantly in recent years. However, the use of these models may also expose systems to new forms of cyberattacks by different malicious actors -- adversarial prompt attack (APA) being one of the most prominent examples of such threats. This paper presents the implementation of an Adversarial Prompting Framework (APF) for a comprehensive assessment of AI safety. The framework systematically evaluates the resilience of the AI model through the generation of structured adversarial prompts at multiple sophistication levels, from direct harmful requests to advanced encoding-based attacks. Our implementation demonstrates the practical application of this methodology in enterprise environments, providing automated testing capabilities with quantitative security assessment metrics. The results indicate significant variations in the model vulnerabilities across different attack vectors, with encoded prompts presenting the highest success rates in bypassing safety mechanisms.
arXiv:2607.13473v1 Announce Type: new
Abstract: In many practical scenarios the shapes of scatterers exhibit uncertain geometric variations arising from diverse physical or environmental factors. For inverse scattering problems which are inherently ill-posed, the presence of such geometric uncertainties may have a non-negligible impact on the recovery process. With the aim of recovering both obstacle geometry and statistics of the shape uncertainties, in this paper we study an inverse acoustic scattering problem for three-dimensional smooth star-shaped obstacles with random isotropic fluctuations. We propose an efficient Monte Carlo-based multi-frequency recursive linearization algorithm in which the far-field operator is linearized with respect to the geometry parameters and frequency continuation is employed to recover the unknown geometry from coarse to fine scales. Based on the reconstructed samples, we further estimate the reference geometry and key statistics of the shape fluctuation field including Karhunen--Lo\`eve eigenvalues, covariance hyper-parameters for Gaussian perturbations and covariance structure, representative marginal distributions for non-Gaussian perturbations. We also prove that the probability law of the far-field data uniquely determines the radial function in distribution which implies uniqueness of the reference shape and related statistics. Numerical experiments demonstrate the effectiveness of the proposed method in recovering both the scatterer shapes and the associated statistical information under Gaussian and non-Gaussian random variations.
arXiv:2607.13475v1 Announce Type: new
Abstract: Surgical tissue retraction requires effective manipulation planning under partial and noisy perception. We study state estimation for deformable tissue retraction, where only sparse observations of the tissue surface are available at decision time. We propose a learned state estimator that reconstructs the full deformable mesh state from 40 noisy vertex observations. The estimator combines a multilayer perceptron with a low-dimensional PCA latent representation and is trained using geometry-aware regularization that encourages smooth and physically plausible deformations. We evaluate the approach in a 2D deformable sheet simulation using single-step and multi-step retraction planning. Results show that the learned estimator achieves 98.1% of oracle performance in multi-step retraction while supporting efficient inference. These results demonstrate that learned, geometry-regularized state estimation can support effective deformable manipulation under realistic perception constraints.
arXiv:2607.13820v1 Announce Type: new
Abstract: Large Language Models (LLMs) are increasingly being used in everyday software engineering tasks, particularly in automated code generation. Despite their widespread adoption, these models remain far from perfect, making systematic and fair evaluation essential to understand their strengths and limitations. In the context of code generation, existing benchmarks are limited: they often target a single programming language and rely primarily on unit test outcomes, while overlooking other critical dimensions such as the overall quality of the generated code and its closeness to a valid solution. To address these gaps, we introduce PROBE, an extensible benchmark framework that, unlike prior work, establishes a systematic structure built on diverse and well-defined metrics, representative workloads, varied prompt templates, and a robust experimental procedure. In practice, the code generated by the LLMs is evaluated along three complementary dimensions: functional correctness, proximity to valid solutions, and code quality, enabling a comprehensive assessment of performance. We use PROBE to evaluate four open-source and two proprietary models under three prompting strategies across five programming languages. We further complement this analysis with a study of common errors in the code and provide concrete examples, offering clearer insight into where LLMs tend to struggle. Our findings show that, while LLMs achieve promising results, they struggle with harder problems and, in the case of smaller models, with programming languages that have fewer available resources for training, and they often fail due to fundamental and easily avoidable errors that underscore the unreliability of automatically generated code.
arXiv:2504.02814v2 Announce Type: replace
Abstract: In this paper, we investigate the Markovian iteration method for solving coupled forward-backward stochastic differential equations (FBSDEs) with a fully coupled drift term of the form $b(t,X_t,Y_t,Z_t)$. An FBSDE system typically involves three stochastic processes: the forward process $X$, the backward process $Y$ representing the solution, and the $Z$ process corresponding to the scaled derivative of $Y$. Previous work by Bender and Zhang (2008) established convergence results for iterative schemes for $Y$-coupled FBSDEs. However, extending these results to equations with $Z$ coupling presents significant challenges, particularly in obtaining a uniform control of the Lipschitz constants of the decoupling fields across iterations and time steps within a fixed-point framework.
To overcome this issue, we propose a novel differentiation-based method for handling the $Z$ process. This approach enables better control of the Lipschitz constants of decoupling fields, facilitating the well-posedness of the discretized FBSDE system with fully coupled drift. We rigorously prove the convergence of our Markovian iteration method in this more complex setting. Finally, we develop an efficient algorithm for computing the resulting numerical scheme, and numerical experiments confirm the theoretical findings and demonstrate the effectiveness and accuracy of the proposed methodology.