Search Results (948)

Symmetry and asymmetry between past and future knowledge are at the heart of continual learning. Deep neural networks typically lose the temporal symmetry that would preserve earlier knowledge when the network is trained sequentially, a phenomenon known as catastrophic forgetting. Dynamically expandable networks (DENs) attempt to restore symmetry by allocating a dedicated module—such as a feature extractor or a task token—for every new task while freezing all previously learned modules. Although this strategy yields high average accuracy, we observe a pronounced asymmetry: earlier tasks still degrade over time, indicating that frozen modules alone do not guarantee knowledge conservation. Moreover, feature bias, arising from the imbalance between old and new samples, further exacerbates the forgetting issue. This raises a fundamental challenge: how can multiple feature extractors be coordinated more effectively to mitigate catastrophic forgetting while enabling the robust acquisition of new tasks? To address this challenge, we propose two asymmetric, contrastive auxiliary losses that exploit rich information from previous tasks to guide new task learning. Specifically, our approach integrates features extracted by both frozen and current modules to reinforce task boundaries while facilitating the learning process. In addition, we introduce a feature adjustment mechanism to alleviate the bias caused by class imbalance. Extensive experiments on benchmarks, including DyTox and MCG, demonstrate that our approach reduces catastrophic forgetting and achieves state-of-the-art performance on ImageNet-100. Full article

ESG has emerged as a key non-financial indicator, drawing significant investor focus. Disparities in ESG ratings may skew investor perceptions, potentially endangering stock values and financial market stability. This paper examines the link between ESG rating divergences and stock price crash risk, drawing on data from six Chinese and global ESG rating agencies. Focusing on Shanghai and Shenzhen A-share listed firms, it analyzes information from 2015 to 2022 within the theoretical contexts of information asymmetry and external monitoring. This study finds that ESG rating divergence markedly elevates stock price crash risk, a relationship that persists through a series of robustness checks. Specifically, the mechanisms operate through two key pathways: increased reputational damage risk due to information asymmetry and reduced external monitoring due to weakened external governance. The results of the heterogeneity analysis indicate that ESG rating divergence exacerbates stock price crash risk more significantly for non-state-owned firms, firms with low levels of marketization, and firms in high-pollution industries. This study provides clear actionable strategic paths and policy intervention points for investors to avoid risks, firms to optimize management, and regulators to formulate policies. Full article

Understanding how media narratives frame the Sustainable Development Goals (SDGs) is essential for global sustainability governance. This study presents a novel, data-driven analysis of 135,000 news articles mapped to SDGs 1–17 across 100 countries. Using polarity-based sentiment aggregation and principal component analysis (PCA), we reduce high-dimensional SDG sentiment profiles into a two-dimensional space and identify emergent clusters of countries using K-means. To contextualize these clusters, we integrate national-level indicators like Human Development Index (HDI), GDP per capita, CO₂ emissions, and press freedom scores, revealing robust correlations between sentiment structure and developmental attributes. Countries with higher HDI and freer media environments produce more optimistic and diverse SDG narratives, while lower-HDI countries tend toward more polarized or crisis-framed coverage. Our findings offer a typology of SDG discourse that reflects geopolitical, environmental, and informational asymmetries, providing new insights to support international policy coordination and sustainability communication. This work contributes a scalable methodology for monitoring global sustainability sentiment and underscores the importance of narrative equity in achieving Agenda 2030. Full article

The integration of artificial intelligence (AI) into environmental, social, and governance (ESG) disclosure represents a critical frontier for corporate transparency in emerging markets. This study investigates the relationship between AI adoption in ESG reporting, disclosure quality, and financial performance among 180 Saudi-listed companies (2021–2024) within Vision 2030’s transformative context. Using the System Generalized Method of Moments (GMM) estimation with panel unit root and cointegration testing to ensure stationarity assumptions and addressing endogeneity through bounding analysis, the study finds that AI adoption intensity significantly enhances ESG disclosure quality (β = 0.289, p < 0.001), with coefficient significance assessed through t-tests using firm-clustered robust standard errors. Enhanced disclosure quality translates into meaningful financial performance improvements: 0.094 percentage points in return on assets (ROA), 0.156 in return on equity (ROE), and 0.0073 units in Tobin’s Q. Mediation analysis reveals that 73% of AI’s total effect operates through improved ESG quality rather than direct operational benefits. The findings demonstrate parametric bounds robust to macroeconomic confounders, suggesting AI-enhanced transparency creates substantial shareholder value through strengthened stakeholder relationships and reduced information asymmetries. Full article

This paper describes a crowdsourced experiment in which participants were asked to judge which of two simultaneously presented facial images (one real, one AI-generated) was fake. With the growing presence of synthetic imagery in digital environments, cognitive systems must adapt to novel and often deceptive visual stimuli. Recent developments in cognitive science propose that some mental processes may exhibit quantum-like characteristics, particularly in their context sensitivity. Drawing on Tezzin’s “generalized fair coin” model, this study applied Contextuality-by-Default (CbD) theory to investigate whether human judgments of human faces exhibit quantum-like contextuality. Across 20 trials, each treated as a “generalized coin”, bootstrap resampling (10,000 iterations per coin) revealed that nine trials demonstrated quantum-like contextuality. Notably, Coin 4 exhibited strong context-sensitive causal asymmetry, where both the real and synthetic faces elicited inverse judgments due to their unusually strong resemblance to one another. These results support the growing evidence that cognitive judgments are sometimes quantum-like contextual, suggesting that adopting comparative strategies, such as evaluating unfamiliar faces alongside known-real exemplars, may enhance accuracy in detecting synthetic images. Such pairwise methods align with the strengths of human perception and may inform future interventions, user interfaces, or educational tools aimed at improving visual judgment under uncertainty. Full article

The remarkable success of autoregressive Large Language Models (LLMs) is predicated on the causal attention mechanism, which enforces a static and rigid form of informational asymmetry by permitting each token to attend only to its predecessors. While effective for sequential generation, this hard-coded unidirectional constraint fails to capture the more complex, dynamic, and nonlinear dependencies inherent in sophisticated reasoning, logical inference, and discourse. In this paper, we challenge this paradigm by introducing Dynamic Asymmetric Attention (DAA), a novel mechanism that replaces the static causal mask with a learnable context-aware guidance module. DAA dynamically generates a continuous-valued attention bias for each query–key pair, effectively learning a “soft” information flow policy that guides rather than merely restricts the model’s focus. Trained end-to-end, our DAA-augmented models demonstrate significant performance gains on a suite of benchmarks, including improvements in perplexity on language modeling and notable accuracy boosts on complex reasoning tasks such as code generation (HumanEval) and mathematical problem-solving (GSM8k). Crucially, DAA provides a new lens for model interpretability. By visualizing the learned asymmetric attention patterns, it is possible to uncover the implicit information flow graphs that the model constructs during inference. These visualizations reveal how the model dynamically prioritizes evidence and forges directed logical links in chain-of-thought reasoning, making its decision-making process more transparent. Our work demonstrates that transitioning from a static hard-wired asymmetry to a learned and dynamic one not only enhances model performance but also paves the way for a new class of more capable and profoundly more explainable LLMs. Full article

This paper investigates the incentive mechanism for dual-channel healthcare service supply chains, where doctors simultaneously undertake both offline and online medical tasks, based on the common agency theory. Considering the geographical distance between online patients and public hospitals, we construct common agency, game-theoretic models under two scenarios: without spillover effects and with spillover effects. Through analytical solutions, we derive the equilibrium outcomes for both scenarios and conduct comparative and numerical analyses. The findings reveal that as follows: (1) Compared to the scenario without spillover effects, the incentive intensity for offline healthcare increases when spillover effects are considered, and doctors exert higher effort levels in offline healthcare. (2) The incentive intensity for online healthcare may decrease, yet doctors’ effort levels in the online channel do not decline accordingly and may even increase; (3) Non-economic incentives (e.g., online reputation) exhibit a substitution effect on economic incentives; (4) Online reputation not only influences decision-making in the online healthcare channel but also affects decisions in the offline channel through spillover effects. These findings provide valuable insights for public hospitals and online healthcare platforms to optimize incentive structures and for doctors to allocate efforts effectively across dual-channel healthcare services. Full article

Tomicus yunnanensis, a notorious forest pest in southwest China, primarily employs infochemicals to coordinate mass attacks that overcome host tree defenses. However, secondary visual cues, particularly detection of host color changes, also aid host location. This study characterized the compound eye structure and vision of T. yunnanensis using electron microscopy and phototaxis tests. The apposition eye contains 224–266 ommatidia, with asymmetry between left and right. Quadrilateral facets occupy the dorsal third, while hexagonal facets dominate elsewhere. Each ommatidium comprises a large corneal lens, an acone-type crystalline cone from four cone cells, and an open-type rhabdom formed by eight retinular cells (R7–R8 centrally, R1–R6 peripherally), surrounded by two primary and at least seventeen secondary pigment cells. Dark/light adaptation alters cone size/shape and rhabdom cross-sectional area/outline (without pigment granule movement) to regulate light reaching the photoreceptors. Behavioral observations showed peak flight activity occurs between 7:00–11:00 AM, with no nighttime activity. Phototaxis tests revealed females are highly sensitive to 360 nm, 380 nm, and 700 nm wavelengths, while males exhibit high sensitivity to 360 nm and 400 nm. This work enhances knowledge on the integration of visual and olfactory sensory information in beetles for host location and non-host avoidance. Full article

Fake news generation and propagation occurs in large volumes, at high speed, in diverse formats, while also being short-lived to evade detection and counteraction. Despite its role as an enabler, Artificial Intelligence (AI) has been effective at fake news detection and prediction through diverse techniques of both supervised and unsupervised machine learning. In this article, we propose a novel Artificial Intelligence (AI) approach that addresses the underexplored attribution of information asymmetry in fake news detection. This approach demonstrates how fine-tuned language models and emotion embeddings can be used to detect information asymmetry in intent, emotional framing, and linguistic complexity between content creators and content consumers. The intensity and temperature of emotion, selection of words, and the structure and relationship between words contribute to detecting this asymmetry. An empirical evaluation conducted on five benchmark datasets demonstrates the generalizability and real-time detection capabilities of the proposed AI approach. Full article

Background: Variability and low volume yield in breast aesthetic outcomes utilizing fat grafting promoted a search for surgical technique improvement. Aim: Using evidence-based information to optimize a surgical technique for aesthetic breast augmentation using stem cell-enriched fat grafting. Methods: Retrospective study of consecutive women (n = 118) from 2008 to 2025 requesting breast fat grafting using centrifugation–filtration fat processing combined with platelet-rich plasma and autologous adipose-derived stem cell-enriched fat. Results: Most surgical indications were for primary breast augmentation (65.8%), followed by fat grafting after implant removal (13.6%), during or after mammoplasty (13.6%), or simultaneously with implant exchange (12.7%). The mean volume per breast of purified, enriched fat grafted was 192 to 206 cc. Each patient had fat grafted into the subcutaneous plane with some patients having additional fat placed submuscularly in those without a dual plane or submuscularly placed implant, or where an implant capsule was absent. Most patients were either very satisfied or satisfied (95.8%), with 4.2% dissatisfied. Those dissatisfied were mainly those with insufficient breast volume and one with a suspected atypical mycobacteria infection. There was a 11.9% complication rate, with seroma formation at the harvested site the most common at 5.1% (n = 6). Palpable fibrotic areas were second in frequency at 3.4% (n = 4), but with no instances of breast oil cyst formation. The average number of fat grafting sessions per indication was only one, with 6.8% requesting a second staged fat grafting procedure. The revision procedures were only in patients with a sole augmentation indication, except for one mastopexy patient with severe breast size asymmetry. An estimated 75–85% grafted volume take was confirmed by a previous diagnostic ultrasound measurement study. Conclusions: Breast fat grafting incorporating learned knowledge of optimal harvesting, processing, storing, enrichment, and administration techniques yielded superior consistent breast enhancement aesthetic outcomes with a high patient and surgeon satisfaction rate through increased adipocyte survival, while minimizing complications including a low incidence of fibrotic areas and no oil cyst formation. Full article

This study aims to investigate asymmetries in muscle activation and co-contraction of main lower limb muscles during flamenco Zap-3 footwork with consideration of the footwork speed and dancer proficiency. Twelve flamenco dancers participated, including six professionals and six amateurs. Each participant performed the Zap-3 sequence under three speed conditions: 160 beats per minute (bpm), 180 bpm and the fastest speed level (F). The normalized surface electromyography was recorded in the gastrocnemius medialis (GM), biceps femoris (BF), tibialis anterior (TA) and rectus femoris (RF) in the dominant (DL) and non-dominant leg (NDL). The co-contraction index was also calculated for selected muscle pairs. The results showed that significant asymmetries occurred only in professional dancers and exclusively at the F speed level. Specifically, the value of the GM in the NDL was higher than that of the DL (p < 0.05, d = 1.97); the value of the BF in the DL was higher than that of the NDL (p < 0.05, d = 1.86) and the co-contraction index of BF/RF in the DL was higher than that of the NDL (p < 0.05, d = 1.87). Understanding these asymmetries may help to inform individualized training strategies aimed at optimizing performance and reducing potential risks. Full article

This study explores the transformative potential of blockchain technology in overhauling conventional housing rental systems. It specifically addresses persistent issues, such as information asymmetry, fraudulent listings, weak Rental Agreements, and data breaches. A comprehensive review of ten academic publications highlights the architectural frameworks, underlying technologies, and myriad benefits of decentralized rental platforms. The intrinsic characteristics of blockchain—immutability, transparency, and decentralization—are pivotal in enhancing the credibility of rental information and proactively preventing fraudulent activities. Smart contracts emerge as a key innovation, enabling the automated execution of Rental Agreements, thereby significantly boosting efficiency and minimizing reliance on intermediaries. Furthermore, Decentralized Identity (DID) solutions offer a robust mechanism for securely managing identities, effectively mitigating risks associated with data leakage, and fostering a more trustworthy environment. The suitability of platforms such as Hyperledger Fabric for developing such sophisticated rental systems is also critically evaluated. Blockchain-based systems promise to dramatically increase market transparency, bolster transaction security, and enhance fraud prevention. They also offer streamlined processes for dispute resolution. Despite these significant advantages, the widespread adoption of blockchain in the rental sector faces several challenges. These include inherent technological complexity, adoption barriers, the need for extensive legal and regulatory adaptation, and critical privacy concerns (e.g., ensuring compliance with GDPR). Furthermore, blockchain scalability limitations and the intricate balance between data immutability and the necessity for occasional data corrections present considerable hurdles. Future research should focus on developing user-friendly DID solutions, enhancing blockchain performance and cost-efficiency, strengthening smart contract security, optimizing the overall user experience, and exploring seamless integration with emerging technologies. While current challenges are undeniable, blockchain technology offers a powerful suite of tools for fundamentally improving the rental market’s efficiency, transparency, and security, exhibiting significant potential to reshape the entire rental ecosystem. Full article

This theoretical study rigorously investigates how generative artificial intelligence reshapes decision-making in social entrepreneurship by modeling the opportunity search process through the lens of optimal stopping theory. Social entrepreneurs often face high uncertainty and resource constraints, requiring them to strategically balance the cost of continued searching with the chance of identifying socially impactful opportunities. This study develops a formal model that captures two core mechanisms of generative AI: reducing search costs and increasing the probability of mission-aligned opportunity success. The theoretical analysis yields three key findings. First, generative AI accelerates the optimal stopping point, allowing social entrepreneurs to act more quickly on high-potential opportunities by lowering cognitive and resource burdens. Second, the influence of increased success probability outweighs that of reduced search costs, underscoring the strategic importance of insight quality over efficiency in socially embedded contexts. Third, the benefits of generative AI are amplified in uncertain environments, where it helps navigate complexity and mitigate information asymmetry. These insights contribute to a deeper conceptual understanding of how intelligent technologies transform the cognitive and strategic dimensions of social entrepreneurship, and they offer empirically testable propositions for future research at the intersection of AI, innovation, and mission-driven opportunity pursuit. Full article

Recent theoretical and empirical studies highlight that information asymmetry and owner–manager conflict of interest can distort corporate investment decisions. Building on this premise, we hypothesize that superior environmental, social, and governance (ESG) performance mitigates these frictions by (H1) alleviating financing constraints and (H2) intensifying external analyst scrutiny. To test these hypotheses, we examine all Shanghai and Shenzhen A-share non-financial firms from 2009 to 2023. Using panel fixed-effects and two-stage least squares with an industry–province–year instrument, we find that higher ESG performance significantly reduces investment inefficiency; the effect operates through both lower financing constraints and greater analyst coverage. Heterogeneity analyses reveal that the improvement is pronounced in small non-state-owned, non-high-carbon firms but absent in large state-owned high-carbon emitters. These findings enrich the literature on ESG and corporate performance and offer actionable insights for regulators and investors seeking high-quality development. Full article

Public–private partnership (PPP) logistics infrastructure projects have become increasingly prevalent globally. Consequently, the effective management of these projects as their concession periods expire presents a crucial challenge for governments, vital for the sustainable management of PPP logistics infrastructure. This study addresses this challenge by focusing on the pivotal post-concession decision: whether the government should self-operate the mature logistics infrastructure or re-delegate its management to a private entity. Our theoretical model, built on a principal–agent framework, first establishes a social welfare baseline under government self-operation and then analyzes delegated operation under symmetric information, identifying efficiency frontiers. Under symmetric information, we find that government self-operation is more advantageous when its own operational efficiency is sufficiently high, irrespective of the private enterprise’s efficiency; conversely, delegating to an efficient private enterprise is optimal only when government operational efficiency is low. We also demonstrate that if the government can directly specify the demand quantity and service level and delegates operation via a fixed fee, the enterprise can be incentivized to align with the social optimum. However, under asymmetric information, potential welfare gains from delegation are inevitably offset by informational rent and output distortion. We further uncover non-monotonic impacts of parameters like the proportion of low-cost firms on social welfare loss and demonstrate how information asymmetry can indirectly compromise the long-term resilience of the infrastructure. Ultimately, our work asserts that delegation is only superior if its potential efficiency gains sufficiently offset the inherent losses stemming from information asymmetry. Full article