Research using machine learning to solve scientific problems. Covers AI quality control, sea-surface temperature estimation, label-free cell analysis, and cyber defense.
Researchers developed an AI model to optimize the macronutrient content of pooled human donor milk recipes, decreasing production time by 60%. The model improved protein and fat levels in milk bank products without compromising bacterial safety, benefiting preterm and sick babies.
A new AI program created by researchers at UF and NVIDIA can generate medical records so well that human physicians couldn't tell the difference. The GatorTronGPT model uses a large language model to mimic natural human language, overcoming hurdles such as protecting patients' privacy and being highly technical.
Researchers at Universitat Oberta de Catalunya used machine learning to simulate patient conditions based on questionnaire answers, predicting stress test performance and identifying symptoms. This approach provides a warning for early diagnosis and referral to specialized units.
A novel machine learning model, FIREANN, accurately simulates system-field interactions for complex chemical, biological, and material systems. The model correlates response properties like dipole moment and polarizability with potential energy changes under external fields.
A team of researchers, including York University, used a mouse model to test how the brain learns new sensory input patterns. They found that the brain's response to image patterns that violate expectations evolves differently over time, suggesting a distinct role in sensory learning.
A new tool called Facemap uses deep neural networks to relate mouse facial movements to neural activity in the brain. This allows researchers to track and quantify movements and correlate them with brain activity, bringing them one step closer to understanding how the brain uses persistent, widespread signals.
Researchers argue for a 'human-centered AI' approach to co-creativity, balancing automation with human control. They emphasize the need for interdisciplinary research on creativity, ethics, and intellectual property rights in human-AI collaboration.
The study used a machine learning approach called FUN-PROSE to predict how fungi react to different environmental conditions. The model was able to accurately predict the expression of genes in baker's yeast and two less studied fungi, with limitations noted for organisms with more complex gene regulation.
Researchers at Rice University are developing a machine learning framework to improve decision-making processes in military communication networks. The goal is to enable rapid, adaptive action across a broad range of scenarios by combining local data in the most effective manner.
Scientists at PNNL introduced a new way to evaluate AI system recommendations by incorporating human experts' insights. Human expertise improved the accuracy of predictions and boosted confidence scores, indicating better decision-making capabilities for machine learning systems.
A recent study published in Nature reveals that machine learning algorithms designed to diagnose bacterial vaginosis in women show diagnostic bias among ethnic groups. The research found that Hispanic women were more likely to receive false-positive diagnoses, while Asian women received the most false-negative diagnoses.
A recent study by researchers at Charité – Universitätsmedizin Berlin highlights the limitations of large language models like ChatGPT in precision medicine. Human experts were found to be more accurate in identifying personalized treatment options for fictitious cancer patients.
Scientists at the University of Copenhagen and University of Victoria have developed an AI formula to predict rogue waves, which can split apart ships and damage oil rigs. The new knowledge can make shipping safer by identifying the likelihood of being struck by a monster wave at sea.
Researchers used machine learning to guide high-throughput experimental screening of small molecules, finding ones that improve vaccine response and reduce inflammation. The team discovered a molecule that outperforms the best immunomodulators on the market, with potential applications in cancer treatment.
Researchers developed a deep learning model that can identify previously unknown quasicrystalline phases in multiphase crystalline samples. The model achieved a prediction accuracy of over 92% and successfully detected an unknown phase in Al-Si-Ru alloys.
A new USC study identifies two metabolites that may predict which young Latino people are most likely to develop prediabetes. The research found that allylphenol sulfate and caprylic acid were the most predictive of prediabetes when combined with other risk factors.
The Python code library snnTorch, developed by UC Santa Cruz's Jason Eshraghian, has surpassed 100,000 downloads and is used in various projects. A new paper published in the Proceedings of the IEEE documents the library and offers a candid educational resource for students and programmers interested in brain-inspired AI.
A novel robotic system developed by USC researchers can help clinicians accurately assess a patient's rehabilitation progress. The method generates an 'arm nonuse' metric using machine learning and a socially assistive robot to track how much a patient is using their weaker arm spontaneously.
Researchers developed a new 3D inkjet printing system that works with a wider range of materials, including slower-curing materials. The system utilizes computer vision to automatically scan the print surface and adjust the amount of resin deposited in real time.
A team of researchers has developed an atom-predicting model similar to the GPT models that support applications like ChatGPT. The new model focuses on small organic molecules with relevance to energy storage and conversion applications.
Researchers developed DIRFA, an AI-based program that generates realistic videos with facial animations synchronized to spoken audio, showcasing improvements over existing approaches. The tool has potential applications in healthcare, education, and entertainment, enhancing user experiences.
Researchers found that smaller subsets of data can be just as effective in training AI models, reducing the need for massive computing power. The study suggests that information richness is more important than dataset size.
Researchers at NC State University developed an autonomous system called SmartDope to synthesize 'best-in-class' materials for specific applications in hours or days. It uses a self-driving lab to manipulate variables, characterize optical properties, and update its understanding of the synthesis chemistry through machine learning.
Researchers developed a deep convolutional neural network to pinpoint cardiac catheter tip locations in photoacoustic images, achieving high precision and recall. The approach has the potential to replace fluoroscopy during cardiac interventions, leading to safer procedures.
Recent study by University of Bonn researchers reveals that machine learning models in drug discovery research are not as effective as thought, relying heavily on memorized data. The findings suggest that AI applications in this field are overrated and should be supplemented with chemical knowledge and simpler methods.
The UTSA MATRIX AI Consortium has received a $2 million grant to create new AI models that rapidly learn, adapt, and operate in uncertain conditions. The team aims to bridge the gap between human brain processing efficiency and current AI limitations, enabling more efficient and adaptive AI systems.
Yu Yang's NSF-funded research aims to reduce vehicle emissions and promote the use of electric bikes and scooters by developing socially informed traffic signal control systems. The project involves a three-pronged method that uses low-cost mobile air-quality sensing, spatial-temporal graph diffusion learning, and reinforcement learnin...
Scientists have developed an AI system that accurately maps the surface area and outline of giant icebergs in one-hundredth of a second. This technology surpasses manual interpretation methods, which can take several minutes to delineate an iceberg's outline, and offers insights into their impact on the polar environment.
Researchers developed an AI system that can scan through college application essays to identify evidence of key personal traits, such as leadership and perseverance. The system aims to reduce algorithmic bias and provide more holistic admissions decisions.
Researchers from the University of Cambridge have developed a virtual reality application that allows users to build figures and shapes without interacting with menus. The 'HotGestures' system uses machine learning to recognize hand gestures, providing fast and effective shortcuts for tool selection and usage.
A recent study published in Nature Communications validates MSIntuit CRC, an AI-driven digital pathology diagnostic, as a reliable pre-screening tool for colorectal cancer. The diagnostic accurately rules out nearly 50% of MSS patients while correctly classifying over 96% of MSI patients.
A University of Houston research team integrated machine learning with SHAP analysis to identify the city's air pollution sources more accurately. The study found that the oil and gas industry had the highest impact on emissions, while shortwave radiation and relative humidity were key influencing factors for overall ozone concentration.
A Lancaster University academic argues that AI and algorithms contribute to polarization, radicalism, and political violence, posing a threat to national security. The paper examines how AI has been securitized throughout its history, highlighting the need for better understanding and management of its risks.
Researchers at Johns Hopkins Medicine created a machine learning model to calculate percent necrosis in osteosarcoma patients after chemotherapy. The model achieved an 85% positive correlation with musculoskeletal pathologist results, increasing accuracy to 99% when one outlier was removed. This could help provide patients with earlier...
Researchers are combining biology, physics, computer science, and engineering to design electric circuits that mimic the brain's adaptive behavior. The goal is to create a more efficient AI application that can learn from history and adapt without significant energy consumption.
Researchers at the University of Sydney have developed a physical neural network that can learn and remember data in real-time, using nanowire networks to mimic brain-inspired learning and memory functions. The network achieved high accuracy in benchmark image recognition tasks and demonstrated its capacity for online learning.
A new study found that high-peace countries are characterized by an increased prevalence of words related to optimism for the future and fun, while low-peace countries feature more references to control and fear. The research used a machine learning model to identify these linguistic patterns in media articles from 18 countries.
The Portuguese team TWIZ from NOVA School of Science and Technology secured 1st Place in the Alexa TaskBot Challenge 2 with a multimodal conversational agent. The winning team was led by João Magalhães and included CMU Portugal Affiliated Ph.D. students Diogo Tavares and Diogo Silva, who improved their visual interface as their biggest...
A team of scientists discovered two types of neurons in fruit flies and mice that enable them to identify distinct smells. With experience, these animals can learn to differentiate between very similar odors, a process that could improve machine-learning models and AI systems.
A new project aims to help robots assess risks and make autonomous decisions. The research focuses on quantifying ambiguity in robot perception to improve safety and efficiency.
Researchers developed an autonomous measurement algorithm to optimize electrical resistance measurements in materials libraries. The new approach enables faster characterization of materials by actively selecting the next measurement area.
Researchers at Osaka University use a robotic system to automate key experimental processes, accelerating the search for new materials. They evaluate 576 thin-film semiconductor samples using photoabsorption spectroscopy, optical microscopy, and time-resolved microwave conductivity analyses.
Researchers found self-supervised models generate activity patterns similar to mammalian brains, suggesting an organizing principle. The models learn representations of the physical world to make accurate predictions, potentially unlocking human-labeled data limitations.
Researchers from the UMA developed an open-source platform called Open Twins to create more accessible and versatile digital twins. This platform enables the simulation of real-world assets based on virtual replicas, predicting future behaviors and detecting anomalies, leading to more efficient companies that make data-driven decisions.
Researchers developed an AI-powered method to measure urban decay using street view images, identifying object classes like potholes and graffiti. The model showed promise in detecting urban decline in cities like San Francisco and Mexico City, with potential applications for informing urban policy and planning.
Researchers at Osaka University have developed a novel platform that combines nanopore technology with artificial intelligence to detect different coronavirus variants quickly. The platform was tested on 241 saliva samples and detected the Omicron variant 100% of the time.
University of Alberta researchers have identified taurine as a key player in predicting poor clinical outcomes and treating long COVID. A predictive test and proposed supplement trial aim to minimize symptoms and improve patient outcomes.
Researchers at New York University developed a novel learning procedure called Meta-learning for Compositionality (MLC) that enables neural networks to make compositional generalizations. MLC outperforms existing approaches and is on par with, and in some cases better than, human performance.
A new property evaluation method for nanoparticles' shape anisotropy has been developed using deep learning, achieving classification accuracy of approximately 80% on single particle basis. This breakthrough solves a long-standing issue in nanoparticle evaluation dating back to Einstein's time.
NTU Singapore has expanded its research collaborations with French partners to push the boundaries of science. The university has inked six new partnerships and renewed existing collaborations across various fields, including quantum physics, nuclear energy, and sustainability.
A new AI-based tool estimates patient-specific cancer survival rates using machine learning, offering personalized prognosis for patients with breast, thyroid, and pancreatic cancers. The Cancer Survival Calculator differs from existing estimators by incorporating multiple factors beyond disease stage.
Researchers used Fourier-transform infrared spectroscopy and machine learning to predict adsorption capacity of pharmaceuticals and personal care products on long-term aged microplastics. The study successfully captured the complexity of the system with up to 98% accuracy, providing new insights into the interactions between microplast...
Researchers at MIT found that similarity-focused generative AI models falter when tasked with designing new products, highlighting the need to prioritize innovation in engineering tasks. By adjusting training objectives and metrics, AI can be an effective 'co-pilot' for engineers, enabling faster creation of innovative products.
A proposed AI-centric medical curriculum aims to educate future healthcare practitioners in digital technology, with a focus on technical concepts, validation, ethics, and appraisal. The curriculum caters to varying student levels, from consumers to developers, promoting interprofessional collaboration and adaptable learning.
Lerrel Pinto aims to create robots that can perform a variety of tasks using algorithms and data collection methods. He seeks to address the trade-off between dexterity and generality in robotics.
Scientists designed an mRNA nanovaccine using machine learning to overcome delivery barriers, promoting strong immune responses and activating the STING pathway to kill tumor cells. The therapeutic strategy demonstrated stronger anti-tumor effects in melanoma and colorectal cancer models.
Researchers from GIST propose a novel approach to mitigate overfitting in pretrained models used for voice pathology detection, achieving 12.36% and 15.38% improvement in recall using contrastive learning.
Lehigh University researchers have developed a technique using machine learning and advanced spectroscopy to characterize waste feedstocks for gasification-produced hydrogen. This process has the potential to eliminate hazards associated with stored coal waste and reclaim valuable resources, while also emitting fewer pollutants than tr...
Researchers from Mindgard and Lancaster University exposed vulnerabilities in major Large Language Models (LLMs) that can be copied for as little as $50. This 'model leeching' attack allows attackers to gain insights into how LLMs work, enabling them to launch targeted attacks with an increased success rate.
Researchers developed an AI-based method to estimate BMD from plain X-ray images using a hierarchical learning framework. The approach showed high performance and reliability in estimating BMD, with correlation coefficients of 0.88 and 0.92 compared to DXA and QCT.