搜索结果：Hacked

Recent advances in image synthesis have been propelled by powerful generative models, such as Masked Generative Transformers (MaskGIT), autoregressive models, diffusion models, and rectified flow models. A common principle behind their success is the decomposition of complex synthesis tasks into multiple tractable steps. However, this introduces a proliferation of step-specific parameters to be configured for modulating the iterative generation process (e.g., mask ratio, noise level, or temperature at each step). Existing approaches typically rely on manually-designed scheduling rules to manage this complexity, demanding expert knowledge and extensive trial-and-error. Furthermore, these static schedules lack the flexibility to adapt to the unique characteristics of each individual sample, yielding sub-optimal performance. To address this issue, we present AdaGen, a general, learnable, and sample-adaptive framework for scheduling the iterative generation process. Specifically, we formulate the scheduling problem as a Markov Decision Process, where a lightweight policy network is introduced to adaptively determine the most suitable parameters given the current generation state, and can be trained through reinforcement learning. Importantly, we demonstrate that simple reward designs, such as FID or pre-trained reward models, can be easily hacked and may not reliably guarantee the desired quality or diversity of generated samples. Therefore, we propose an adversarial reward design to guide the training of the policy networks effectively. Finally, we introduce an inference-time refinement strategy and a controllable fidelity-diversity trade-off mechanism to further enhance the performance and flexibility of AdaGen. Comprehensive experiments across five benchmark datasets (ImageNet-256 × 256 & 512 × 512, MS-COCO, CC3M, and LAION-5B) and four distinct generative paradigms validate the superiority of AdaGen . For example, AdaGen achieves better performance on DiT-XL with $\mathbf {\sim 3\times }$∼3× lower inference cost and improves the FID of VAR from 1.92 to 1.59 with negligible additional computational overhead.

The rapid digitalization of radiation oncology has enabled transformative advances but also introduced new vulnerabilities for millions of cancer patients. While most reported cyberattacks in healthcare have involved ransomware, the potential compromise of linear accelerators (LINACs) would pose a far greater risk to patient safety. This review examines cybersecurity challenges in radiation oncology with a focus on LINACs. We analyze the anatomy of potential attacks and their implications to raise awareness of stealth threats. We describe LINACs typical network topology and relevant clinical workflow to identify critical assets, vulnerabilities, and attack vectors. We then explore defence mechanisms and real-world best practices, emphasizing that although radiotherapy benefits from mature safety practices and strong clinical safeguards, proactive measures are still needed to anticipate cyber threats and strengthen system resilience. Immediate adoption of a multilayered zero-trust security architecture, supported by robust vendor and regulatory measures, is essential to protect patient safety and maintain clinical continuity in an evolving threat landscape.

Adaptive deep brain stimulation (aDBS) is a neuromodulation technology that enables real-time monitoring and automatic adjustment of stimulation parameters. While enhancing treatment precision, this technology also introduces more intricate ethical challenges than conventional DBS (cDBS). Current discussions of autonomy in aDBS often appeal to different definitions of autonomy, making it difficult to fully capture the distinctive problems posed by aDBS. To address this gap, this article proposes a multidimensional analytical framework centred on control, derived from aDBS's two distinctive features: 'automatic unperceived operation' and 'patient-device shared control of neural activity'. Within this framework, control is categorised along two dimensions: source and perception. Using this framework, the study identifies three categories of ethical challenges: (1) unperceived external control by third parties (such as hacking) constitutes a potential manipulation of individual autonomy; (2) unperceived external control arising from system autonomy may become conflated with patients' internal control, blurring human-machine boundaries and creating dilemmas of self-identity and responsibility attribution; and (3) perceived external control may foster patient dependence or even addiction to the technology, undermining individual capacities for internal control. This mapping provides a more explanatory conceptual toolkit for ethical assessment and policy-making around aDBS, aiming to ensure that technological innovation promotes human well-being while safeguarding individual dignity and autonomy.

We present datasets from three large-scale human-subject experiments involving red-team hacking in a cyber range in the Guarding Against Malicious Biased Threats (GAMBiT) project. Across Experiments 1-3 (July 2024-March 2025), 19-20 skilled attackers per experiment conducted two 8-hour days of self-paced operations in a simulated enterprise network (SimSpace Cyber Force Platform) while collecting multi-modal data: self-reports (background, demographics, psychometrics), operational notes, terminal histories, key logs, network packet captures (PCAP), and NIDS alerts (Suricata). Each participant began from a standardized Kali Linux VM and pursued realistic objectives (e.g., target discovery and data exfiltration) under controlled constraints. Derivative curated logs and labels are included. The combined data release supports research on attacker behavior modeling, bias-aware analytics, and method benchmarking. Data are available via IEEE DataPort entries for Experiments 1-3.

Social investments are strategic allocation of resources to welfare programs, such as education or health care, expected to generate social and economic returns. In Europe, social investment has become a well established policy paradigm, yet research has rarely examined how it impacts how state-citizen relationships are understood. This article explores how the social investment discourse functions as a frame of interaction within welfare programs for young people with drug problems in Denmark. Theoretically, the article draws on Foucault's concept of discourse, Goffman's theory of frames and Hacking's notion of looping effects. Empirically, it presents two contrasting cases of young people with drug problems who position themselves as, respectively, good and bad investments. The social investment discourse provides some youths and their relatives with means to legitimize requests for help; it enables a selfpresentation as "a good investment", which can strengthen their sense of entitlement to welfare. In contrast, others come to view themselves as "bad investments". For them, the discourse frames programs as a waste of money, which risk to undermine the sense of entitlement among individuals who fear they might fail to benefit from welfare programs. Social investment is a policy paradigm; it is not designed as a frame of self-presentation. Nevertheless, once its rationality enters public discourse, it produces new interactional frames and subject positions. An unintended but significant effect is that the discourse may stigmatize marginalized citizens, undermine their sense of entitlement to welfare, and weaken the universal, normative and rights-based principles underpinning the Nordic welfare states.

This cross-sectional study (N&#xa0;=&#xa0;572) examined how chronological age and individual risk perceptions influence the perceived usefulness of disruptive implant and wearable technologies, controlling for sex. Previous research has frequently examined age effects in technology acceptance but rarely employed differentiated measures of perceived risk, especially for body-integrated technologies. To address this gap, we developed and validated the Implant Risk Scale (IRS) to assess specific risk dimensions associated with implantable technologies. Ten potential risk dimensions were identified from prior research, and 50 items were developed based on the relevant scientific literature. Exploratory factor analysis of this item pool identified four distinct risk factors: ethical, physiological, psychological, and data hacking risk, accounting for 85&#xa0;% of the total variance. Each risk could reliably be assessed with three items. In addition to age (&#x3b2;&#xa0;=&#xa0;-0.22, p&#xa0;<&#xa0;.01), physiological risk (&#x3b2;&#xa0;=&#xa0;-0.27, p&#xa0;<&#xa0;.01) and psychological risk (&#x3b2;&#xa0;=&#xa0;-0.16, p&#xa0;<&#xa0;.01) had independent negative effects on perceived usefulness. Moreover, moderation analyses indicated that each implant risk dimension enhanced the negative effect of age on perceived usefulness (&#x394;R2s between 0.01 and 0.02, ps&#xa0;<&#xa0;0.01). Differentiated risk dimensions should be considered when examining technology adoption behavior. The validated IRS provides a concise framework for capturing these risk assessments. Different demographic groups seem to be affected to varying degrees by these risk dimensions. Social inequalities between demographic groups due to the use of disruptive technologies might be widened by perceived risk.

Lithium niobate materials, which have the potential to fabricate lasers, modulation devices, and photodetectors, are widely used in quantum information processing due to their exceptional optical and electro-optical properties. However, the photorefractive effect in lithium niobate materials may cause the quantum device to deviate from its ideal operation model, which is an important assumption for the security of quantum key distribution. Here, we demonstrate the practical security of the continuous-variable quantum key distribution protocol under the photorefractive effect, where the eavesdropper Eve injects a 488 nm visible light into the source-side variable optical attenuator and excites the photorefractive effect phenomenon in a lithium-niobate-based dual-waveguide. In particular, we derive the photorefractive-effect-induced intensity change of a variable optical attenuator and then the corresponding parameter estimation process, showing the effectiveness of this attack for various irradiation powers and waveguide technologies (including proton-exchanged and annealed-proton-exchanged waveguides). To show its effect on practical continuous-variable quantum key distribution systems, we present the composable finite-size security of one-way continuous-variable quantum key distribution and continuous-variable measurement-device-independent quantum key distribution protocols. Numerical results show that this induced-photorefraction attack can break the security even with a low irradiation power, such as 3 [Formula: see text] (or about 0.21 [Formula: see text]). In addition, we find that Eve's optimal attack strategy against continuous-variable measurement-device-independent quantum key distribution is related to the information reconciliation process by comparing three different attack strategies. Finally, we discuss some possible countermeasures to resist induced-photorefraction attacks to enhance the security of the system.

Swimming-induced pulmonary oedema (SIPE) is a potentially fatal condition associated with open-water swims. We sought to quantify the contribution of cardiac dysfunction to SIPE. We aimed to assess the incidence of SIPE during an endurance cold water swim. We determined associations between SIPE and changes in cardiac function through a SIPE questionnaire, lung ultrasound (LUS), cardiac biomarkers (N-Terminal pro-B-type natriuretic peptide (NT proBNP) and cardiac troponin I (cTnI)) and transthoracic echocardiograms (TTE). Twenty open-water swimmers (10 males) underwent a TTE, LUS for pulmonary oedema and cardiac biomarkers before, 2-hours and 24-hours after an 8-hour swim. Swimmers had an additional LUS and rated their breathlessness upon leaving the water. Participants with breathlessness and 3 or more B-lines present in two or more LUS views were considered SIPE positive. Five swimmers (25&#x202f;% of cohort) presenting with post-event breathlessness and evidence of lung water were considered SIPE positive. SIPE had no demonstrable effect on left systolic function (LV ejection fraction, global longitudinal strain [GLS]), diastolic function (left atrial volume, E/e') and right ventricular (RV) function (RV fractional area change and RV free wall GLS). SIPE was associated with a small increase in troponin post-swim (at 2&#x202f;hours SIPE+ 32.1&#x202f;ng/L, SIPE- 12.6&#x202f;ng/L, p&#x202f;=&#x202f;0.004: at 24&#x202f;hours SIPE+ 12.6&#x202f;ng/L, SIPE- 4.8&#x202f;ng/L, p&#x202f;=&#x202f;0.04) but had no impact on NT proBNP. SIPE is common in open-water swimmers following an endurance swim and is identified using LUS. Whilst SIPE was associated with a small increase in post-swim troponin levels, no further evidence of cardiac dysfunction was identified at two hours after the swim to explain the pulmonary oedema.

Recent approaches in text-to-image (T2I) generation have actively adopted reinforcement learning (RL) techniques for human preference alignment. However, existing approaches primarily rely on a single reward function, which can lead to overfitting on specific metrics, resulting in issues such as reward hacking and imbalanced optimization among multiple objectives. To address this, we propose Flow-Multi: a flow-matching multi-reward framework for text-to-image generation. Our method builds upon flow-matching-based group-relative policy optimization (GRPO) learning. Each sample is evaluated by four reward models-based on text-to-image alignment, human preference, aesthetic quality, and GenEval-to create a multi-dimensional reward vector. We then utilize the Pareto dominance relationship to remove dominated samples and update the policy using only the non-dominated set. Additionally, we introduce advantage masking during training to suppress the contribution of low-reward samples, ensuring that only high-quality rewards are reflected in policy optimization. Experimental results demonstrate that Flow-Multi achieves balanced improvements across multiple reward criteria compared to the existing Flow-GRPO, validating the effectiveness of the multi-reward reinforcement learning framework for stable alignment in text-to-image generation.

Diffusion models (DMs) have achieved state-of-the-art generative performance but suffer from high sampling latency due to their sequential denoising nature. Existing solver-based acceleration methods often face significant image quality degradation under a low-latency budget, primarily due to accumulated truncation errors arising from the inability to capture high-curvature trajectory segments. In this paper, we propose the Ensemble Parallel Direction solver (dubbed as EPD-Solver), a novel ODE solver that mitigates these errors by incorporating multiple parallel gradient evaluations in each step. Motivated by the geometric insight that sampling trajectories are largely confined to a low-dimensional manifold, EPD-Solver&#xa0;leverages the Mean Value Theorem for vector-valued functions to approximate the integral solution more accurately. Importantly, since the additional gradient computations are independent, they can be fully parallelized, preserving low-latency sampling nature. We introduce a two-stage optimization framework. Initially, EPD-Solver&#xa0;optimizes a small set of learnable parameters via a distillation-based approach. We further propose a parameter-efficient Reinforcement Learning (RL) fine-tuning scheme that reformulates the solver as a stochastic Dirichlet policy. Unlike traditional methods that fine-tune the massive backbone, our RL approach operates strictly within the low-dimensional solver space, effectively mitigating reward hacking while enhancing performance in complex text-to-image (T2I) generation tasks. In addition, our method is flexible and can serve as a plugin (EPD-Solverplugin) to improve existing ODE samplers. Extensive experiments demonstrate the effectiveness of EPD-Solver. On validation benchmarks, at the same latency level of 5 NFE, the distilled EPD-Solver&#xa0;achieves state-of-the-art FID scores of 4.47 on CIFAR-10, 7.97 on FFHQ, 8.17 on ImageNet, and 8.26 on LSUN Bedroom, surpassing existing learning-based solvers by a significant margin. On T2I benchmarks, our RL-tuned EPD-Solver&#xa0;significantly improves human preference scores on both Stable Diffusion v1.5 and SD3-Medium. Notably, it outperforms the official 28-step baseline of SD3-Medium with only 20 steps, effectively bridging the gap between inference efficiency and high-fidelity generation.

暂无摘要（点击查看详情）

Oncotype DX has played a critical role in guiding treatment decisions for hormone receptor (HR)-positive, HER2-negative early-stage breast cancer. Clinically, a subset of patients with low Oncotype recurrent score (RS) will still progress on standard therapy and ultimately develop metastasis. Our goal was to explore potential molecular mechanisms, including specific genetic alterations and pathway activity associated with disease progression. We retrospectively reviewed a small series of low RS breast cancers with subsequent metastasis and analysed the clinicopathological characteristics and comprehensive genomic profiling (CGP) data from tumour tissue and circulating tumour DNA (ctDNA) by liquid biopsy. These tumours demonstrated a range of histopathologic features and molecular profiles. Common findings included enrichment of PIK3CA and TP53 mutations and treatment-emergent ESR1 mutations, observed in both tissue and ctDNA. CDKN2A, SPEN, KIT, CTNNB1, MYC, EMSY, KMT2C, MAP3K1 gene alterations were only found in low RS group in low frequency. Copy number amplifications events were less common in low RS group. In cases with both tissue and ctDNA data, tissue CGP proved useful baseline for identifying driver mutations such as PIK3CA and for contextualizing ctDNA findings, and ctDNA analysis was adequate for disease monitoring and tracking molecular evolution over time. Using real-world CGP of tumour tissue and ctDNA, we identified key molecular features associated with endocrine resistance in patients with low RS who later developed metastases. PIK3CA mutation and other ER group-related mutations contributed to the low RS. Tissue CGP provides baseline for interpreting ctDNA, and ctDNA monitoring PIK3CA, TP53, ESR1 and other pathogenic or driver mutations in the early course of low RS cases may represent an excellent non-invasive option for identifying targets and early intervention to prevent disease progression, though a large validation study is needed.

Repetitive transcranial magnetic stimulation (rTMS) is an emerging treatment for brain disorders, but its therapeutic mechanism is poorly understood. We developed a mouse model of rTMS with superior clinical face validity and investigated the neural mechanism by which accelerated intermittent theta burst stimulation (aiTBS), the first rapid-acting rTMS antidepressant protocol, reversed chronic stress-induced behavioral deficits. Using fiber photometry, we showed that aiTBS drives distinct patterns of neural activity in intratelencephalic (IT) and pyramidal tract (PT) projection neurons in dorsomedial prefrontal cortex (dmPFC). However, only IT neurons exhibited persistently increased activity during both aiTBS and subsequent depression-related behaviors. aiTBS reversed stress-related loss of dendritic spines on IT, but not PT neurons, further demonstrating cell type-specific effects of stimulation. Chemogenetically inhibiting dmPFC IT, but not PT neurons, during rTMS blocked the antidepressant-like behavioral effects of aiTBS. Thus, we demonstrate a prefrontal mechanism linking rapid aiTBS-driven therapeutic effects to cell type-specific circuit plasticity.

Controlling reactivity using light remains an important goal in chemistry. Since acid-base reactions are central to nearly all chemical phenomena, photoacids are important molecules for controlling reactivity. Conventional photoacids release protons upon excitation or generate a Lewis acid through light-induced structural changes. Here, we report a conceptually different mechanism of Lewis photoacidity that leverages the excited-state mechanisms of a Br&#xf8;nsted photoacid. It is well known that the Br&#xf8;nsted photoacid 2-naphthol releases a proton upon light absorption and converts to 2-naphtholate. We show that 2-naphtholate binds the Lewis-acidic CO2 in the ground state and releases it upon photoexcitation. Through experimental and computational studies, we demonstrate that the spectroscopic characteristics and F&#xf6;rster cycle of this process are remarkably similar to those observed for proton release from 2-naphthol. This strategy provides a new approach for designing systems for reversible CO2 capture and release and has potential for photochemistry of Lewis adducts beyond CO2.

The protection of high-value cell lines (assets) relies on physical security by limiting access to samples. We present a cybersecurity-inspired platform that protects biological assets at the genetic level. This technology uses a permutation lock design where an asset can only be decrypted using an authentication code r from a search space composed of n objects on a defined keypad. Here, the genetic asset is designed as a scrambled DNA sequence, and the code is a temporal pattern of small molecules that regulate sets of recombinases that can unscramble a DNA sequence into the desired final sequence. In this work, a "blue team" designed and built an encrypted (scrambled) DNA sequence, and a "red team" sought to break the code through an ethical hacking exercise. Two iterations of testing revealed a 0.2% (2 in 990) chance of gaining access to the asset by random search, which is on par with the theoretical goal of 0.1% (1 in 990).

暂无摘要（点击查看详情）

The rapid proliferation of digital technologies has transformed the landscape of gender-based violence globally. This quantitative study used an online survey to explore the experiences of women with disabilities in relation to technology-facilitated gender-based violence (TFGBV) in South Africa. Findings from 204 participants highlight patterns across age, province, education, employment, income, disability type, and forms of TFGBV experienced, and how TFGBV may differ at intersections of these factors. They show that cyberbullying, hacking, and hate speech were the most prevalent forms of TFGBV, disproportionately affecting women with various disabilities. The study further reveals how socio-economic disadvantage, manifested in limited access to secure technologies, digital literacy, and support systems, intensifies exposure to harm and constrains access to justice. The study calls for inclusive, power-conscious approaches to research, policy and interventions that centre lived experiences of women with disabilities. Addressing TFGBV in low- and middle-income countries (LMICs), therefore, requires not only legal reform and digital safety initiatives but also broader strategies for socio-economic empowerment and systemic transformation to end gendered-disability violence in both the material and virtual world.