Browse Articles
Discover research articles across all indexed journals
Echoes in AI: Quantifying lack of plot diversity in LLM outputs
With rapid advances in large language models (LLMs), there has been an increasing application of LLMs in creative content ideation and generation. A critical question emerges: can current LLMs provide ideas that are diverse enough to truly bolster collective creativity? We examine two state-of-the-art LLMs, GPT-4 and LLaMA-3, on story generation and discover that LLM-generated stories often consist of plot elements that are echoed across a number of generations. To quantify this phenomenon, we introduce the Sui Generis score, an automatic metric that measures the uniqueness of a plot element among alternative storylines generated using the same prompt under an LLM. Evaluating on 100 short stories, we find that LLM-generated stories often contain combinations of idiosyncratic plot elements echoed frequently across generations and across different LLMs, while plots from the original human-written stories are rarely recreated or even echoed in pieces. Moreover, our human evaluation shows that the ranking of Sui Generis scores among story segments correlates moderately with human judgment of surprise level, even though score computation is completely automatic without relying on human judgment.
Effectiveness of educational intervention based on the protection motivation theory in promoting brucellosis preventive behaviors in ranchers
Visual processing oscillates differently through time for adults with ADHD
ADHD is a neurodevelopmental disorder affecting 3–4% of Canadian adults and 2.6% of adults worldwide. Its symptoms include inattention, hyperactivity and impulsivity. Though ADHD is known to affect several brain functions and cognitive processes, little is known regarding its impact on perceptual oscillations. This study compared the temporal features of visual processing between ADHD and neurotypical individuals in a visual word recognition task through the use of a temporal sampling technique, the outcome of which are classification images reflecting processing effectiveness according to the temporal properties of the stimulus. These temporal features were sufficiently different across groups while at the same time sufficiently congruent across participants of the same group that a machine learning algorithm classified participants in their respective groups with a 91.8% accuracy using only a small portion of the available features. Secondary findings showed that individuals with ADHD could be classified with high accuracy (91.3%) regarding their use of psychostimulant medication. These findings suggest the existence of strong behavioral markers of ADHD as well as of regular medication usage on visual performance which can be uncovered by random temporal sampling.
Parallel trade-offs in human cognition and neural networks: The dynamic interplay between in-context and in-weight learning
Human learning embodies a striking duality: Sometimes, we can rapidly infer and compose logical rules, benefiting from structured curricula (e.g., in formal education), while other times, we rely on an incremental approach or trial-and-error, learning better from curricula that are randomly interleaved. Influential psychological theories explain this seemingly conflicting behavioral evidence by positing two qualitatively different learning systems—one for rapid, rule-based inferences (e.g., in working memory) and another for slow, incremental adaptation (e.g., in long-term and procedural memory). It remains unclear how to reconcile such theories with neural networks, which learn via incremental weight updates and are thus a natural model for the latter, but are not obviously compatible with the former. However, recent evidence suggests that metalearning neural networks and large language models are capable of in-context learning (ICL)—the ability to flexibly infer the structure of a new task from a few examples. In contrast to standard in-weight learning (IWL), which is analogous to synaptic change, ICL is more naturally linked to activation-based dynamics thought to underlie working memory in humans. Here, we show that the interplay between ICL and IWL naturally ties together a broad range of learning phenomena observed in humans, including curriculum effects on category-learning tasks, compositionality, and a trade-off between flexibility and retention in brain and behavior. Our work shows how emergent ICL can equip neural networks with fundamentally different learning properties that can coexist with their native IWL, thus offering an integrative perspective on dual-process theories of human cognition.
A dual-stream deep learning framework for skin cancer classification using histopathological-inherited and vision-based feature extraction
Exploring the landing speed of digital society PPP projects: A continuous-time event history analysis of 300 projects
The Public-Private Partnership (PPP) model has become a viable alternative or supplement to traditional approaches in the development of a digital society. However, PPP projects in this domain often face significant landing (the launch of project implementation, typically marked by contract signing and the commencement of operational activities) challenges. Understanding the factors that influence the speed of project landing is thus of considerable practical importance. Based on the theoretical framework of Van Meter and Van Horn, we propose a causal mechanism linking the policy implementation system to the landing speed of digital society PPP projects. Using data from 300 projects, the study empirically tests this mechanism through continuous-time event history analysis. The results indicate that, compared with organizational and actor factors, material resources such as government fiscal resources and regional financial resources have a relatively weaker impact on landing speed. In contrast, internal goal consensus on innovation, internal government coordination, government-business relationship quality, leadership performance demand, and corporate social responsibility all significantly promote faster landing. Conversely, due to low project profitability and market barriers to social capital participation, information enterprise development is negatively associated with landing speed. We provide targeted policy recommendations to enhance the efficiency and timeliness of PPP-driven digital society initiatives.
Correction for Jankovic et al., Wireless arm-worn bioimpedance sensor for continuous assessment of whole-body hydration
Bioherbicidal and cytogenotoxic potential of nanoemulsions containing essential oils from Piper amalago and Piper dilatatum
Personalized models of disorders of consciousness reveal complementary roles of connectivity and local parameters in diagnosis and prognosis
The study of disorders of consciousness (DoC) is very complex because patients suffer from a wide variety of lesions, affected brain mechanisms, different severity of symptoms, and are unable to communicate. Combining neuroimaging data and mathematical modeling can help us quantify and better describe some of these alterations. The goal of this study is to provide a new analysis and modeling pipeline for fMRI data leading to new diagnosis and prognosis biomarkers at the individual patient level. To do so, we project patients’ fMRI data into a low-dimension latent-space . We define the latent space’s dimension as the smallest dimension able to maintain the complexity, non-linearities, and information carried by the data, according to different criteria that we detail in the first part. This dimensionality reduction procedure then allows us to build biologically inspired latent whole-brain models that can be calibrated at the single-patient level. In particular, we propose a new model inspired by the regulation of neuronal activity by astrocytes in the brain. This modeling procedure leads to two types of model-based biomarkers (MBBs) that provide novel insight at different levels: (1) the connectivity matrices bring us information about the severity of the patient’s diagnosis, and, (2) the local node parameters correlate to the patient’s etiology, age and prognosis. Altogether, this study offers a new data processing framework for resting-state fMRI which provides crucial information regarding DoC patients diagnosis and prognosis. Finally, this analysis pipeline could be applied to other neurological conditions.
Using large language models to categorize strategic situations and decipher motivations behind human behaviors
By varying prompts to a large language model, we can elicit the full range of human behaviors in a variety of different scenarios in classic economic games. By analyzing which prompts elicit which behaviors, we can categorize and compare different strategic situations, which can also help provide insight into what different economic scenarios might induce people to think about. We discuss how this provides a step toward a nonstandard method of inferring (deciphering) the motivations behind the human behaviors. We also show how this deciphering process can be used to categorize differences in the behavioral tendencies of different populations.
Deep computer vision with artificial intelligence based sign language recognition to assist hearing and speech-impaired individuals
Sonar image denoising based on clustering and Bayesian sparse coding
Side-scan sonar image (SSI) are often affected by a combination of multiplicative speckle noise and additive noise, which degrades image quality and hinders target recognition and scene interpretation. To address this problem, this paper proposes a denoising algorithm that integrates non-local similar block clustering with Bayesian sparse coding. The proposed method leverages cross-scale structural features and noise statistical properties of image patches, and employs a similarity metric based on the Equivalent Number of Looks (ENL) along with an improved K-means clustering algorithm to achieve accurate classification and enhance intra-class noise consistency. Subsequently, a joint training strategy is used to construct dictionaries for each cluster, and Bayesian Orthogonal Matching Pursuit (BOMP) is applied for sparse representation. This enables effective modeling and suppression of mixed noise while preserving structural details. Experimental results demonstrate that the proposed method outperforms several classical approaches in both objective metrics such as PSNR and SSIM, and in visual quality, particularly in preserving target edges and textures under severe noise conditions.
Abiotic synthesis during the interaction of ferrous chloride–rich silicic fluids with marble under high-grade metamorphic conditions
Ferrous chloride–rich silicic fluid and melt infiltration led to the decarbonation of dolomitic marble in the Chinese Sulu ultrahigh-pressure metamorphic terrain under temperatures ranging from 670 to 800 °C, pressures from the aragonite + albite to calcite stability fields and the oxygen fugacities between the hematite–magnetite and pyrite–pyrrhotite–magnetite buffers, resulting in the formation of olivine marble and diopsidite. The inclusions in zircons trapped during the decarbonation process suggest that the H 2 -producing reaction 3FeCl 2 (aq) + 3CaCO 3 + H 2 O = Fe 3 O 4 (magnetite) + 3CaCl 2 (aq) + 3CO 2 + H 2 occurred and that it induced magnetite-catalyzed Fischer–Tropsch-type synthesis, as indicated by the presence of whewellite, disordered carbonaceous material, CH 4 , and CO in the inclusions. The results of this study highlight the role of aqueous Fe in generating H 2 and magnetite and have far-reaching implications for carbon speciation and solubility in deep fluids and for endogenic abiotic synthesis, which may be pivotal in the prebiotic occurrence of organic compounds on Earth.
Optimal estimation of power Chris-Jerry distribution parameters using ranked set sampling design with application
Kinematical error analysis and autonomous calibration of a 5PUS-RPUR parallel robot
Kinematic calibration is essential for improving the absolute accuracy of parallel robots, but conventional identification methods often struggle with the complex, non-linear coupling of their numerous geometric error parameters. This can lead to convergence to local rather than global optima, limiting the effectiveness of the calibration. To address this challenge, this paper proposes a novel self-calibration methodology based on a global optimization strategy. Taking the 5PUS-RPUR parallel robot as an example, its inverse kinematics is established based on screw theory. A sensitivity analysis is performed using the finite difference method to screen for and eliminate error sources with a negligible impact on the moving platform’s pose. Measurement points are then selected uniformly throughout the workspace using the farthest point sampling algorithm. An objective function for the GA is constructed by integrating the actuator displacement errors from each kinematic chain with the overall pose error of the moving platform. Non-linear constraints are handled using a penalty function approach. Based on measurement data from an onboard IMU and joint encoders, the identification results are obtained. The experimental results demonstrate that the proposed method significantly improves the robot’s positional accuracy across its entire workspace. The superiority and efficacy of this approach are further corroborated by a benchmark comparison with three recent, state-of-the-art calibration methodologies.
Profile of Johannes Lehmann
Soil scientist Johannes Lehmann has spent his career examining how to improve soil quality to secure agricultural production, regulate climate, and keep water clean. His work with dark earths in the Brazilian Amazon led to the discovery of the importance of biochar in soil fertility and nutrient recycling from excreta. For more than 24 years, Lehmann has held a faculty position at Cornell University, where his research initially focused on carbon and nutrient cycles and went on to challenge the existence of soil humus and delve into the role of functional complexity in soil management.
Single cell RNA seq reveals the pro-regenerative phenotype of thrombospondin-2 deficient dermal fibroblasts
The MCTOT app: A publicly available tool for statistical cycle-to-threshold analysis and inference of informative but uncertainly determined qPCR data
As a common experimental technique, qPCR (Quantitative Real-time Polymerase Chain Reaction) is widely used to measure levels of nucleic acids, e.g., microRNAs and messenger RNA. While providing accurate and complete data, researchers have inevitably encountered uncertainly determined qPCR data because of intrinsically low amounts of biological material. The presence of incomplete or uncertain qPCR data challenges interpretation accuracy. This study presents a web application that integrates two sophisticated statistical methods – a flexible regression approach and a two-group hypothesis testing technique – to enhance the accuracy and robustness of qPCR data analysis with informative but uncertainly determined observations. To demonstrate the versatility and efficacy of our MCTOT (Multi-Functional Cycle-To-Threshold Statistical Analysis Tool) application, this study presents two distinct examples employing two-group hypothesis testing. The first example delves into an analysis of pathogens in wastewater, an area gaining increasing relevance for public health surveillance. The second example illustrates an application in the realm of liquid biopsy, a rapidly evolving field in disease diagnostics, monitoring, and early treatment. Moreover, the application’s process is further exhibited through another liquid biopsy example, wherein the flexible regression method is employed to detect the hemolysis effect on a molecular target. These examples demonstrate the tool’s capacity to not only identify significant differences between groups but also to quantify the effect size, a crucial aspect in biomedical research. The MCTOT web application stands as a pioneering step toward empowering researchers to harness the full potential of qPCR data, especially when dealing with informative but uncertainly determined observations. It also paves the way for further development of web-based tools that adhere to the refined CTOT (Cycle-To-Threshold) methodology, opening new avenues in qPCR data analysis and interpretation. The developed application can be accessed online through Shinyapps.io at https://ctot.shinyapps.io/bioinformatics/ for open access.
Evolutionarily conserved grammar rules viral factories of amoeba-infecting members of the hyperdiverse <i>Nucleocytoviricota</i> phylum
Despite sharing fewer than 10 core genes, the hyperdiverse Nucleocytoviricota phylum (ranging from poxviruses to giant viruses) universally assembles viral factories (VFs) resembling biomolecular condensates. Regardless, it is unclear how these viruses achieve such a level of functional conservation without clear conserved genetic information. We demonstrate that the VFs produced by amoeba-infecting viruses have liquid-like properties and identify a conserved molecular grammar governing viral factory scaffold protein: charge-patterned intrinsically disordered regions that drive phase separation independently of sequence homology. This grammar predicts functional scaffold proteins across the 15 viral families, revealing evolutionary constraints invisible to sequence or structural analysis. Strikingly, VFs exhibit subcompartmentalization analogous to nuclei, segregating transcription and mRNA processing (inner condensates) from replication (interphase zones) and translation (host cytoplasm). Our work establishes phase separation as a fundamental organizational principle bridging extreme genomic diversity, explaining how biological complexity emerges without gene conservation. This grammar is likely also conserved in non-amoeba-infecting members of the phylum and thus may represent a primordial solution for organelle-like organization, with broad implications for antiviral targeting.