Self-trained vision transformers mimic human gaze with surprising precision
Peer-Reviewed Publication
Updates every hour. Last Updated: 12-Sep-2025 11:11 ET (12-Sep-2025 15:11 GMT/UTC)
Researchers from The University of Osaka have revealed that vision transformers can develop gaze patterns similar to humans, learning to prioritize faces without ever being taught what a face is. The findings suggest that self-supervised ViTs may serve as a practical tool in computer vision and a theoretical model for studying human visual processing. The research underscores the potential of biologically inspired AI systems to uncover the mechanisms of human cognition.
Researchers at University of Tsukuba have developed a method to predict when online conversations deviate from their original topics and escalate into personal attacks. Using large language models (LLMs) and a zero-shot prediction approach, the technique achieves high accuracy without requiring platform-specific training.
The molecular pathways involved in antiviral defenses and counter-defenses in host-pathogen systems remain unclear. Researchers from Japan have used Neurospora crassa as a model organism to explore how RNA editing influences fungal antiviral responses. They identified two neighboring genes—an RNA-editing enzyme (old) and a transcription factor (zao)—that regulate virus-induced gene expression. Their findings show how the old-zao module controls both asymptomatic and symptomatic infections, providing new insight into conserved antiviral mechanisms in fungi.
Diamond quantum sensors can be used to analyze the magnetization response of soft magnetic materials used in power electronics; report scientists based on collaborative research. Using a novel imaging technique, they developed quantum protocols to simultaneously image both the amplitude and phase of AC stray fields over a wide frequency range up to 2.3 MHz. Their results demonstrate that quantum sensing is a powerful tool for developing advanced magnetic materials across diverse applications.
Plants require phosphorus to grow and survive. In environments with low levels of soil phosphorus, some plants have adapted by forming cluster roots to extract the small amount of phosphorus available in the soil. Researchers recently performed an RNA-Seq experiment on cluster roots of Hakea laurina to identify important genes involved in the survival of the Proteaceae plant family in low-phosphorus soil. The team discovered a new secretion pathway in H. laurina cluster roots and that the absence of a suberized exodermis further enhances this trait. This knowledge may be applied to engineering crops that can thrive in phosphorus-poor soil.
An Osaka Metropolitan University researcher examined the spatial patterns and geographic characteristics of tourism-accommodation hotspots in Kyoto. The results of this study revealed hotspots in the areas around historic Kyoto through the tourism-accommodation intensity index.