Text-to-image generation models such as Stable Diffusion have recently attracted increasing interest due to their broad applications in creative design and content creation. However, they remain vulnerable due to persistent weaknesses in deep neural networks. While previous studies have investigated adversarial attacks against such models, they often introduce noticeable changes to the original inputs, reducing the imperceptibility of the attacks. In this paper, we propose the Descriptive Term Insert (DesInsert) along with two variants DesInsert-White and DesInsert-Black, designed for white-box and black-box settings, respectively. The key idea is to introduce an optimized descriptive term into the input text, thereby preserving the semantics and enabling imperceptible attacks. Specifically, DesInsert-White leverages a discretized softmax approach to enhance the white-box search process, enabling more efficient discovery of descriptive terms for stealthier attacks. Meanwhile, DesInsert-Black employs a novel genetic encoding strategy in the word space to generate semantically coherent adversarial examples. Extensive experiments demonstrate that DesInsert can consistently distort generated images over popular text-to-image models with minimal perceptible changes: It generates adversarial samples with perplexity (PPL) values that are less than half of those produced by existing baselines. Moreover, it achieves a 21.98% higher success rate and over 10× faster generation speed in white-box settings, and outperforms in black-box settings with higher success rates and fewer query costs. Our work reveals the critical vulnerabilities in current text-to-image systems and highlights the need for developing more robust generative models. The code is available at https://github.com/FanyuBu/DesInsert.
暂无摘要（点击查看详情）
There is a commonly held assumption that feelings such as pain are causes of behaviour. We say we withdrew our hand from the hotplate because it hurt or that we flinched at the needle because it stung. The causal role of pain is widely implicated in theories of learning and decision-making. But what if this commonsense idea that feelings cause behaviour is just wrong? To date, there is no known mechanism for how subjectively experienced pain directly modulates neural activity and it is hard to see how there could be. There is no known mechanism by which pain could directly gate ion channels. On this basis, we contend that the real cause of behaviour is neural activity and that feelings of pain have no known causal role. This raises the question of whether pain has any function at all-i.e., whether it has causal powers or is merely epiphenomenal. Epiphenomenalism faces the intractable problem of explaining how such an attention-consuming feeling as pain could be epiphenomenal and yet still have survived evolutionary selection. In response, we infer from the available neuroscientific evidence that the best explanation is that pain has a novel, non-causal function and that decisions to act are instead caused by an internal decoding process involving threshold detection of accumulated evidence of pain rather than by pain per se. Because pain is necessarily implicated in the best explanation of subsequent decision-making, we do not conclude that pain is epiphenomenal or functionless even if it has no causal influence over decisions or subsequent actions. On this view, pain functions to mark neural pathways that are the causes of behaviour as salient, serving as a ground but not a cause of subsequent decision-making and action. This perspective has far-reaching implications for diverse fields including neuropsychiatry, biopsychosocial modelling, robotics, and brain-computer interfaces.
暂无摘要（点击查看详情）
The figure of the "medical clown" is a theologically and anthropologically charged vocation necessary for moral critique and healing in contemporary medicine. Drawing on the Oglala Sioux Heyoka and the circus clown, or holy fool, the authors distinguish between external critics-philosophers and theologians who unmask medicine's principalities through folly-and internal actors-clinicians who subvert dehumanizing logics from within. Against corporate, efficiency-driven models that deform the healing vocation, both roles embody prophetic reversal, exposing absurdities and reorienting practice toward possibilities of redemption that could not otherwise be seen from within the logic of medicine's current priorities and principalities. The holy fool destabilizes the status quo through provocative critique; the Heyoka, grounded in communal trust, rouses through loving contrariness. Together, they witness to medicine's eschatological promises: healing as relational and resistant to commodification. By reclaiming the sacred work of "walking backwards," these clowns and fools invite a re-formation of medical imagination, challenging and shaping practitioners who dwell with suffering to resist the seductions of a system that is forgetting its calling.
The article explores the Trickster as a symbol of liminality, transgression, and creativity, analyzing its characteristics through mythology, folklore, and literature. The Trickster (a prankster and a "boundary-crosser") embodies ambiguity, deception, and hidden wisdom, operating at the intersection of order and disorder, of sacred and profane. Through myths such as that of Hermes, as well as the tales of fools, jesters, and clowns, the Trickster is portrayed as an agent of transformation, subversion, and innovation. It subverts social and cultural categories through play, irony, hyper-sexualization, and the absence of shame. Turner's theory of liminality and Freud's notion of the uncanny are compared to highlight how the Trickster traverses both psychosocial thresholds and the boundaries between consciousness and the repressed-generating ambiguity, anxiety, and estrangement. The Trickster's archetypal function, as elaborated by Jung, is expressed as both a collective and personal shadow-a symbol of productive chaos and unconscious creativity. This figure proves essential for understanding processes of cultural and psychic change, acting as a catalyst for innovation and symbolic renewal. This article emphasizes the importance of the Trickster as a symbol of psychic and collective processes of innovative adaptation, creativity, and transformation. But at the same time, the Trickster points to the associated risks of disorientation, unease, subversion and suffering. Through this ambivalence, the Trickster provides insights for interdisciplinary inquiry across cultural psychology, anthropology, and psychoanalysis.
Deep learning underpins most of the currently advanced natural language processing (NLP) tasks such as textual classification, neural machine translation (NMT), abstractive summarization and question-answering (QA). However, the robustness of the models, particularly QA models, against adversarial attacks is a critical concern that remains insufficiently explored. This paper introduces QA-Attack (Question Answering Attack), a novel word-level adversarial strategy that fools QA models. Our attention-based attack exploits the customized attention mechanism and deletion ranking strategy to identify and target specific words within contextual passages. It creates deceptive inputs by carefully choosing and substituting synonyms, preserving grammatical integrity while misleading the model to produce incorrect responses. Our approach demonstrates versatility across various question types, particularly when dealing with extensive long textual inputs. Extensive experiments on multiple benchmark datasets demonstrate that QA-Attack successfully deceives baseline QA models and surpasses existing adversarial techniques regarding success rate, semantics changes, BLEU score, fluency and grammar error rate.
暂无摘要（点击查看详情）
Consider the classic Ebbinghaus illusion in figure 1: the orange circle on the right seems bigger, even though both orange circles have the same size. Such size illusions occur in the real world too: in fact, a common dieting trick is to serve the same food on a smaller plate, which makes it appear larger, so you end up eating less. Why do we get fooled by such size illusions?
The plantaris muscle typically originates at the lower portion of the lateral supracondylar line of the femur and oblique popliteal ligament of the knee joint and inserts medial to the Achilles tendon at the calcaneal tuberosity. Functionally, this muscle is unclear, as it could be for proprioception of the leg or flexion of the knee and ankle. Additionally, it is absent in some individuals. If present, the tendon is often mistaken for the tibial nerve, which is why it is known as the "fool's nerve." This case study involves the discovery of a bilateral plantaris muscle duplication in an 88-year-old Japanese woman during a routine cadaveric dissection. This deviates from the more common unilateral presence of an accessory plantaris muscle that is predominantly reported in previous studies. Additionally, both the accessory origin and insertion differ between the two. The right accessory muscle belly originated lateral to the primary, and the tendon terminated in the tendon of the lateral head of the gastrocnemius. Conversely, the left accessory muscle belly originated superomedial to the primary muscle, where the corresponding tendon terminated in the tendon of the medial head of the gastrocnemius. Anatomical variations, including duplication, of the plantaris muscle remain an area of ongoing research due to its clinical relevance with Achilles tendinopathy, tendon grafts, and surgeries. This case reports bilateral asymmetrical accessory plantaris muscles and tendons, which can alert physicians to avoid anesthetic complications when performing popliteal or tibial nerve blocks.
Social media platforms are fighting misinformation by adding warning flags and suggesting related content. However, it is still unclear how users understand these flags and whether they influence users' willingness to share content or believe false information-especially on video platforms such as TikTok. Do users notice these flags? If so, do the flags change how they view the content? Inoculation theory suggests that if individuals are forewarned about the potential for misinformation and are exposed to weakened forms of such misinformation, they may become less susceptible to misinformation. Does inoculating users with flagged content reduce false acceptance and sharing intentions later? To address these questions, we conducted a user study (N = 322) utilizing a TikTok-like interface, employing a 2 (misinformation warning: absent, present) × 3 (counterargument in pre-suggested false-flagged content: absent, cue, action) between-subject experimental design. While users noticed flags as intended, their perceptions regarding the frequency and harm of misinformation remain unchanged. Furthermoe, neither warning nor counterargument messages effectively reduced users' acceptance of misinformation or their intention to share it, highlighting the boundary conditions of inoculation theory within immersive short video contexts. Interestingly, media literacy and TikTok dependency have emerged as significant predictors of false acceptance. This study discusses the practical implications for the design of ethical social media interfaces.
暂无摘要（点击查看详情）
The extensive study of carbon nanotube (CNT) toxicity stems from their widespread application across various fields. The toxicity of CNTs is commonly assessed using cell viability assays based on tetrazolium salts, such as the MTT assay. ISO 10993-5 outlines the MTT assay and related in vitro cytotoxicity tests as international standards. However, nearly two decades ago, it was observed that MTT interacts with CNTs, potentially yielding inaccurate results. Despite this, the MTT assay remains the most widely used method for studying CNT toxicity in vitro today. Here, we demonstrate that six commonly used tetrazolium salts in cell viability assays-MTT, MTS, INT, XTT, WST-1, and WST-8- interfere with both single-walled nanotubes (SWCNTs) and multiwalled carbon nanotubes (MWCNTs). According to ISO 10993-5, cell viability percentages below 70% indicate cytotoxicity. At the standard testing duration of 3 h, the absorbance values in the presence of 5 mg/mL of either SWCNT or MWCNT decreased to below 70% relative to the control. At a lower concentration of 0.5 mg/mL, the effect was less pronounced, with the absorbance decreasing to an average of 84% compared to the control. Our results suggest that none of these cell viability assays alone offers a fully reliable method for evaluating CNT toxicity, especially with high CNT concentrations. Therefore, it is essential to carefully assess which in vitro methods are truly suitable for CNT toxicity studies.
Deep neural networks (DNNs) have achieved satisfactory performance in multiple fields. However, recent studies have shown that DNNs can be easily fooled by adversarial examples. To mitigate the threats caused by adversarial attacks, a highly effective strategy is to design detectors to reject adversarial examples. This article proposes an unsupervised class- and classifier-free adversarial detection method. It only takes unlabeled clean data for training to discriminate illegal samples, and does not require any knowledge about the adversarial examples, sample classes, and the original classifier. More specifically, motivated by the idea that adversarial examples may differ significantly from benign data in terms of sample structural information, we develop an adversarial detector that can simultaneously capture the residual information and the variable-wise structural relationships of data. After that, we design an attribute called data identity (ID) that combines the extracted residual and structural information of data to identify adversarial examples. We validate the superiority of the proposed method through detecting adversarial attacks on CIFAR-10 and ImageNet datasets, and the experimental results demonstrate that the performance of our model is the best among various state-of-the-art adversarial detectors. Besides, we also conduct visualization experiments to illustrate the role of structural information in detecting adversarial examples.
This work investigates the generalization behavior of deep neural networks (DNNs), focusing on the phenomenon of "fooling examples," where DNNs confidently classify inputs that appear random or unstructured to humans. To explore this phenomenon, we introduce an analytical framework based on maximum likelihood estimation (MLE), without adhering to conventional numerical approaches that rely on gradient-based optimization and explicit labels. Our analysis reveals that DNNs operating in an overparameterized regime exhibit a collapse in the output feature space. While this collapse improves network generalization, adding more layers eventually leads to a state of degeneracy, where the model learns trivial solutions by mapping distinct inputs to the same output, resulting in zero loss. Further investigation demonstrates that this degeneracy can be bypassed using our newly derived "wormhole" solution. The wormhole solution, when applied to arbitrary fooling examples, reconciles meaningful labels with random ones and provides a novel perspective on shortcut learning. These findings offer deeper insights into DNN generalization and highlight directions for future research on learning dynamics in unsupervised settings to bridge the gap between theory and practice.
Ultrasound (US) simulation helps train physicians and medical students in image acquisition and interpretation, enabling safe practice of transducer manipulation and organ identification. Current simulators generate realistic images from reference scans. Although physics-based simulators provide real-time images, they lack sufficient realism, while recent deep learning-based models based on unpaired image-to-image translation improve realism but introduce anatomical inconsistencies. We propose a novel framework to reduce hallucinations from generative adversarial networks (GANs) used on physics-based simulations, enhancing anatomical accuracy and realism in abdominal US simulation. Our method aims to produce anatomically consistent images free from artifacts within and outside the field of view (FoV). We introduce a segmentation-guided loss to enforce anatomical consistency by using a pre-trained Unet model that segments abdominal organs from physics-based simulated scans. Penalizing segmentation discrepancies before and after the translation cycle helps prevent unrealistic artifacts. Additionally, we propose training GANs on images in polar coordinates to limit the field of view to non-blank regions. We evaluated our approach on unpaired datasets comprising 617 real abdominal US images from a SonoSite-M turbo v1.3 scanner and 971 artificial scans from a ray-casting simulator. Data was partitioned at the patient level into training (70%), validation (10%), and testing (20%). Performance was quantitatively assessed with Frechet and Kernel Inception Distances (FID and KID), and organ-specific χ 2 $\chi ^2$ histogram distances, reporting 95% confidence intervals. We compared our model against generative methods such as CUT, UVCGANv2, and UNSB, performing statistical analyses using Wilcoxon tests (FID and KID with Bonferroni-corrected α = 0.01 $\alpha = 0.01$ , χ 2 $\chi ^2$ with α = 0.008 $\alpha =0.008$ ). A perceptual realism study involving expert radiologists was also conducted. Our method significantly reduced FID and KID by 66% and 89%, respectively, compared to CycleGAN, and by 34% and 59% compared to the leading alternative UVCGANv2 ( p ≪ 0.01 $p \ll 0.01$ ). No significant differences ( p > 0.008 $p>0.008$ ) in echogenicity distributions were found between real and simulated images within liver and gallbladder regions. The user study indicated our simulated scans fooled radiologists in 36.2% of cases, outperforming other methods. Our segmentation-guided, polar-coordinates-trained CycleGAN framework significantly reduces hallucinations, ensuring anatomical consistency, and realism in simulated abdominal US images, surpassing existing methods.
暂无摘要（点击查看详情）
暂无摘要（点击查看详情）
The advent of immunotherapies and targeted treatments has improved survival for some people with metastatic cancer but also increased prognostic uncertainty. To inform clinician-patient communication and supportive care, this study explored uncertainty-related coping among people with metastatic uveal melanoma (mUM) - a disease for which treatments have emerged especially suddenly. A qualitative approach was taken using semi-structured interviews. Participants with mUM were recruited through consumer organisations internationally. Interviews explored participant perspectives on the impacts of uncertainty and their related coping strategies. Analysis involved inductive coding followed by deductive coding against Mishel's (1988) theoretical framework of uncertainty in illness. Seventeen people participated, including 10 from Australia. Participants described experiencing uncertainty as disempowering but also leveraged the opportunity it presented for remaining hopeful. Some participants used meta-cognition- alluded to as 'tricking' or 'fooling' themselves - to manage inconsistency between hoping for an exceptional response and accepting that benefits were likely to be modest at best. Most participants were able to maintain everyday normalcy but struggled to discuss their illness and treatment with family and friends. Participants reported heightened anxiety in the lead-up to routine scans and while awaiting results. Coping with uncertainty in the era of immunotherapy and targeted treatments involves 'hoping for the best while preparing for the worst'. Supportive care is especially needed at the time of scans. Some patients may also benefit from help with talking to their social networks. Head-to-head comparisons are needed of differing psychological interventions.