Research using machine learning to solve scientific problems. Covers AI quality control, sea-surface temperature estimation, label-free cell analysis, and cyber defense.
A machine learning model has been developed to distinguish the composition ratio of solid mixtures of chemical compounds using only photographs. The model was trained on a small dataset and achieved accuracy roughly twice that of human experts.
Researchers have developed a novel neural network approach to design brand new proteins with unique arrangements and dynamic functionalities. The method combines attention neural networks with graph neural networks to predict existing protein properties and envision new proteins that nature has not yet devised.
Researchers at Singapore Management University aim to create a robust machine learning system capable of correctly identifying Singapore's multiracial food. The project focuses on addressing biases in current systems and developing an algorithm that can recognize a wide range of dishes, including those beyond popular online trends.
Scientists have successfully measured the speed of molecular charge migration in a carbon-chain molecule, revealing a movement of several angstroms per femtosecond. The study used a two-color high harmonic spectroscopy scheme with machine learning reconstruction to achieve a temporal resolution of 50 as.
A new study by the University of Illinois and USDA-Agricultural Research Service has identified the key factors influencing sweet corn yield. The analysis found that seed source is a significant variable, with processors having a choice over which hybrids to use, and high nighttime temperatures also impact yield.
Researchers at MIT developed a method to simplify the process of whole-body manipulation for robots, enabling them to reason efficiently about moving objects. The technique uses AI and smoothing to reduce the number of decisions required, making it possible for robots to adapt quickly in complex environments.
The University of Missouri is launching a five-year, $3 million doctoral training program to prepare the next generation of scientists and engineers for emerging fields like materials science and data science. The program aims to empower future workers with both technical expertise and data-driven insights.
A new AI tool predicts certain forms of esophageal and stomach cancer at least three years prior to diagnosis. The K-ECAN tool uses basic EHR data to identify high-risk patients, who may benefit from earlier screening and preventative measures.
A new study found ChatGPT to be nearly 72 percent accurate across all medical specialties and phases of clinical care. It was also 77 percent accurate in making final diagnoses. However, the model struggled with differential diagnosis, which is a crucial aspect of medicine.
Researchers developed a deep-learning model to assess CXR images for probable COVID-19 severity. The model achieved an area under the receiver operating characteristic curve of 0.78 when predicting intensive care need within 24 hours.
Researchers developed a modified bandit Q-learning algorithm that aims to learn optimal Q values for every state-action pair, balancing exploitation and exploration. The scheme relies on photonic systems to enhance learning quality, accelerating parallel learning through conflict-free decision-making.
A new study reveals that using big data and machine learning can improve antimicrobial resistance surveillance in livestock production. The research found correlations between environmental variables, microbial communities, and antimicrobial resistance, suggesting multiple routes for improving surveillance.
A team led by Dr. Zixiang Xiong at Texas A&M University aims to understand the fundamental limits of learned source coding, a machine learning-based data compression method. They hope to develop more powerful compression methods for efficient use of wireless communication and less energy consumption by mobile devices.
Researchers developed a novel method using Google Trends to assess player popularity and demonstrated its improvement in predicting market value. The method involves calculating six indicators of popularity that can be compared among players, improving accuracy when incorporated with other factors.
Researchers at the University of Pennsylvania School of Engineering and Applied Science have discovered dozens of small protein sequences with antibiotic qualities in extinct organisms like Neanderthals and Denisovans. They then synthesized these molecules using artificial intelligence and tested their efficacy against pathogens.
Researchers used AI and mobility data to enhance air pollution models, improving accuracy by an average of 17.5% and identifying hotspots with high PM2.5 levels. This integrated approach can inform targeted health alerts and safety measures for areas with poor air quality.
Researchers propose a hypothesis that astrocytes, non-neuronal cells in the brain, can perform core computation as transformers, providing insights into human brain function and machine learning success. This discovery could spark future neuroscience research and help explain transformer performance across complex tasks.
A new study shows that a rule-based natural language processing tool successfully identified patients with unstable access to transportation, food insecurity, social isolation, financial problems, and signs of abuse or exploitation. The tool performed better than deep learning algorithms in identifying these social determinants of health.
Scientists at Max-Planck-Institut für Eisenforschung developed a machine learning model that enhances predictive accuracy in alloy design, uncovering new corrosion-resistant compositions. The model combines numerical and textual data, enabling the identification of optimal alloy formulas.
Researchers used AI to accurately classify four subtypes of Parkinson's disease from patient-derived stem cells, with one subtype reaching an accuracy of 95%. The study suggests that personalized medicine and targeted drug discovery could be possible using this approach.
Researchers used machine learning algorithms to analyze ChatGPT-generated and human-written Japanese texts, finding that the AI's style could be distinguished with high accuracy. The study suggests a new method for detecting AI-generated content in academic papers written in Japanese.
A team of researchers from the University of Cambridge developed a way to incorporate human error into machine learning systems, improving their performance in handling uncertain feedback. However, they found that even with uncertainty accounted for, hybrid systems still perform worse than standalone machine learning models.
Researchers have developed a new explainable AI model to reduce bias and enhance trust in machine learning-generated decisions. The Pattern Discovery and Disentanglement (PDD) model can predict medical results with rigorous statistics and explainable patterns, leading to more reliable diagnoses and better treatment recommendations.
The new tool, SnorCall, analyzes unsolicited calls to shed light on robocall trends and types of scams. It extracted information from over 232,000 robocalls, including phone numbers used in scams, helping regulators and law enforcement take action.
A machine learning model found that background parenchymal enhancement (BPE) on breast MRI is an indicator of breast cancer risk in women with extremely dense breasts. Women with dense breasts are at a higher risk of developing breast cancer compared to those with fatty breasts.
Researchers found that clinical strains of Aspergillus fumigatus differ significantly from environmental strains in amino acid synthesis. The fungus appears to shape the lung microbiome to its advantage, surviving on vital metabolites produced by other microorganisms.
Researchers developed an AI model called OncoNPC that can analyze genetic data to predict cancer type and origin. The model accurately classified at least 40% of tumors with unknown origin, leading to a 2.2-fold increase in eligible patients for targeted treatments.
Critically ill children on ventilator support can experience patient-ventilator asynchrony (PVA), which worsens outcomes. Researchers are using machine learning to develop a common set of definitions and measurements for PVA in pediatric patients, aiming to minimize risks and improve clinical outcomes.
Researchers created a self-supervised AI model called GedankenNet that learns physics laws and thought experiments to reconstruct microscopic images. The model successfully reconstructed human tissue samples and Pap smears from holograms without relying on real-world experiments or data.
AI-powered clinical decision support systems can enhance patient care, but doctors lack necessary skills to interpret and act on risk predictions. To address this gap, medical education and training need to incorporate explicit coverage of probabilistic reasoning tailored to CDS algorithms.
A radiomic-based model using T2-weighted MRI data achieved high accuracy in diagnosing pediatric Crohn disease, outperforming expert radiologists. The model was ensembled with clinical data to further improve performance.
Researchers developed a generative AI tool, AniFaceDrawing, to assist users in creating high-quality anime portraits. The tool uses a sketch-to-image framework and employs stroke-level disentanglement to match raw sketches with latent vectors of the generative model.
Researchers at University College London found that humans can only reliably detect fake speech 73% of the time, and this ability improves only slightly with training. The study's findings raise concerns about the potential for deepfakes to be used by criminals and nation-states to cause harm.
Researchers have discovered a way to utilize nonlinear scattering media for optical computing and machine learning. They created a novel theoretical framework involving third-order tensors, which can represent the complex relationships between input and output signals. This breakthrough has potential applications in real-world settings...
A machine learning system capable of learning diverse tennis skills from broadcast video footage has been created by a research team led by Simon Fraser University's Jason Peng. The system can generate long-lasting matches with realistic racket and ball dynamics between two physically simulated characters.
Researchers propose leveraging high-value information to overcome statistical and computational challenges in reinforcement learning. By accessing valuable observations, agents can improve strategies without trial and error, making the learning process more efficient and effective.
A study by University of Toronto researchers found that child language development and language evolution share a common cognitive foundation, based on a core knowledge base. The team built a computational model that predicts word meaning extension patterns across languages and time scales, highlighting the role of visual, associative,...
Researchers designed machine learning models to identify children at risk of self-harm, finding that incorporating more data points and diagnostic codes improved detection rates. The models were particularly effective for detecting underrepresented groups, such as Black and Latino youth.
A new study uses machine learning to analyze data from DrugAge, a database of chemical compounds modulating lifespan in model organisms. The researchers create four types of datasets to predict whether or not a compound extends the lifespan of C. elegans, using features such as compound-protein interactions and Gene Ontology terms.
An analysis of English Twitter data reveals a 17-fold increase in daily FGM conversations on International Day of Zero Tolerance, suggesting opportunities for social media education. At least 200 million women and girls have undergone FGM, leading to short- and long-term health consequences.
Hang aims to develop general-purpose robots that can handle complex physical interactions without requiring perfect input from sensors or extensive instructions. His project seeks to improve robotic manipulation tasks by reducing assumptions about how the robot acts in real-world conditions.
Researchers at Monash University developed a co-training AI algorithm that can effectively mimic human oversight in medical imaging. The algorithm achieved an average improvement of 3% compared to state-of-the-art approaches using limited annotated data, enabling AI models to make more informed decisions and uncover accurate diagnoses.
Researchers have developed a computational tool to compare large datasets and predict immune responses to disease, potentially leading to better vaccines. The new algorithm, designed by La Jolla Institute for Immunology scientists, uses machine learning to identify underlying patterns in immune system data.
The Fengyun-4A satellite in collaboration with a machine learning model generated a detailed PV resource map for China, providing new insights into the country's solar energy potential. This advancement sets a new standard for solar resource mapping, empowering decision-makers to make informed choices for a sustainable future.
A new editorial explores the potential of machine learning to enhance early cancer detection in primary care, leveraging extensive patient data and improving risk stratification accuracy. The authors emphasize the need for responsible implementation, collaboration, and validation across diverse populations.
Researchers developed a tool called SKILL that enables AI agents to learn 102 distinct tasks by sharing knowledge in parallel, reducing the time needed to master new skills. The technology has potential applications in medicine, education, and other fields where vast knowledge is required.
The study evaluates recent research on artificial intelligence-generated molecular structures from the perspective of medicinal chemists, recommending guidelines for assessing novelty and validity. Insilico Medicine's recommendations aim to improve the process of generating and evaluating novel AI-generated drugs.
Researchers have developed a new study showing that targeted electrical stimulation in patients with traumatic brain injury improved memory recall by 19%. The technology delivers the right stimulation at the right time, informed by the wiring of the individual's brain and that individual's successful memory retrieval.
A new geometric deep learning model called GFCN has been developed to detect stroke lesions in brain imaging scans. The model leverages rich geometric information to segment brain tissue and achieves higher segmentation performance than other neural network architectures.
Researchers at the University of Alabama created a detailed assessment of vulnerability to natural hazards across the continental US. The study found significant differences in vulnerability between neighboring blocks and identified top 10 states with high vulnerability rates, including Minnesota and Ohio.
Lero is recruiting 16 top international post-doctoral researchers for a €2.9 million fellowship program focusing on privacy, trust, inclusion and fairness in software expertise. The program will provide discipline-specific skills training and enhance career development.
A new AI approach has confirmed that the Maldivian government effectively enforced its ban on open trash burning and single-use plastics, eliminating toxic smoke plumes from satellite imagery. The tool, trained using transfer learning and image segmentation, achieved 88% accuracy in identifying plumes.
A joint research team has developed a novel approach combining machine learning with quantum-classical computational molecular design to accelerate the discovery of efficient OLED emitters. The optimal OLED emitter discovered is a deuterated derivative of Alq₃, which is both extremely efficient at emitting light and synthesizable.
Researchers at NIST and their colleagues used machine learning to identify abnormal cardiac rhythms in firefighters, achieving 97% accuracy. The Heart Health Monitoring model could lead to a portable heart monitor to detect early warning signs of heart trouble and prevent fatal cardiac events.
A new AI technology has been developed to generate artificial scientific data, allowing for faster and more efficient detection of material features. The AI uses generative adversarial networks to incorporate background noise and experimental imperfections into the generated data, making it virtually indistinguishable from real data.
Researchers used machine learning algorithms to explain how calpain-1 activity is modified on the molecular level, finding that lipid peroxidation products like MDA and HNE can increase or decrease activity. The study provides new insights into protein modification and its role in meat tenderness.
A Lehigh University professor has received $4 million in NIH grants to develop an AI-driven approach for precision mental health diagnosis and care. The project aims to identify biomarkers in the brain that can predict treatment response and personalize interventions for patients with depression and other mental disorders.
Researchers at Flinders University use machine learning to identify species interactions and predict which species are most likely to go extinct. By analyzing species traits and interactions, the algorithm can help plan interventions before extinctions occur.
A machine learning model developed by Duke Health researchers can differentiate normal cognition from mild cognitive impairment using retinal images from the eye. The model achieved a sensitivity of 79% and specificity of 83%, identifying specific features in OCT and OCTA images that signal cognitive impairment. This non-invasive metho...
Researchers at Cornell University developed a new method that uses machine learning to visualize nanotextures in thin-film materials. This technique overcomes the challenge of preserving the sample, allowing for dynamic study of thin films and discovery of new morphologies.