Research using machine learning to solve scientific problems. Covers AI quality control, sea-surface temperature estimation, label-free cell analysis, and cyber defense.
A recent study identified over 182,000 small seismic events in South Korea, with 135,000 related to mining explosions. The researchers used machine learning techniques to analyze data from 421 seismic stations and found distinct patterns that allowed them to distinguish between microseismic events and earthquakes.
Researchers successfully applied reinforcement learning to protein design, creating proteins with improved antibody generation and accurate nano-structures. The approach may lead to more potent vaccines and novel applications in regenerative medicine.
Researchers developed machine-learning algorithms to generate proteins with specific structural features, enabling the creation of biologically inspired materials. The models can produce millions of new protein ideas in a few days, allowing scientists to explore unique applications.
Researchers at Bar-Ilan University have discovered that efficient learning on artificial shallow architectures can achieve the same classification success rates as deep learning architectures, but with less computational complexity. This breakthrough has significant implications for the development of unique hardware and advanced GPU t...
Researchers developed a machine learning model that predicts strokes more accurately than current systems, using data available at hospital arrival. The model achieved high precision and sensitivity, outperforming existing scales.
A recent study found that the Epic sepsis model's accuracy in predicting sepsis onset depends on hospital factors such as sepsis incidence and multiple health conditions. The model performed worse in hospitals with higher rates of these conditions, suggesting it may be more useful in lower-acuity settings.
Researchers have developed a computer-assisted method to automate the assessment of speech severity in ataxia patients, achieving an 80% hit rate. The new methodology leverages artificial intelligence and could simplify procedures for determining ataxia severity, facilitating research and clinical practice.
Using high-throughput experiments and machine learning-based algorithms, researchers forecast the behavior of hybrid perovskites with high accuracy. The study aims to find materials that combine high-efficiency performance with resilience to environmental conditions.
A machine learning program can spot risky conversations on Instagram by analyzing metadata clues, such as conversation length and participant engagement. The system was 87% accurate in identifying risky chats using sparse and anonymous details from over 17,000 private chats.
A new Dartmouth study examines how changes in precipitation and temperature due to global warming affect streamflow and flooding in the Northeast. The research finds that a warmer climate will lead to increased streamflow and higher flood risk, particularly if soils become wetter and more prone to heavy rainfall events.
A new machine learning study found that habit formation varies in time for different behaviors, such as gym-going and hand-washing. The study analyzed data from over 30,000 gymgoers and 3,000 hospital workers, revealing factors like past behavior and time since last visit played significant roles.
Researchers developed ProTac, a soft robotic link with multimodal perception to improve human-robot interactions. The device incorporates tactile and proximity sensing capabilities, enabling robots to react safely and predictably to physical contact.
A machine learning model has been developed to predict a patient's risk for a sleep disorder using demographic, lifestyle, physical exam results, and laboratory values. The model identified depression, weight, age, and waist circumference as the greatest predictors of a sleep disorder diagnosis.
Researchers at the University of Cambridge designed a soft robotic hand that can grasp a range of objects using passive movement and tactile sensors. The hand successfully grasped 11 of 14 objects in tests, including a peach, computer mouse, and roll of bubble wrap, demonstrating its ability to predict when it might drop an object.
A new VR locomotion system, Seamless-walk, offers a natural and comfortable experience without equipment or body pose recording. It uses high-resolution foot pressure imprints and machine learning to estimate the user's direction and movement speed.
Researchers at the University of São Paulo used artificial intelligence and Twitter to develop a database and models that can detect depression and anxiety before clinical diagnosis. The study found that BERT performed best in predicting depression and anxiety, with a statistically significant difference from LogReg.
A new study from Weill Cornell Medicine has identified four clinically distinct groups of people with autism spectrum disorder, each with unique brain connection patterns and behavioral traits. The findings highlight the potential for personalized therapies tailored to individual subgroups, which may lead to more effective treatments.
A new machine learning model estimates optimal treatment timing for sepsis, taking into account vital signs and lab test results to predict patient survival. The model was trained on a dataset of over 14,000 individuals with sepsis and showed improved outcomes when actual treatment matched the recommended timeline.
A Swansea University study reveals the potential of machine learning in identifying Ankylosing Spondylitis (AS) patients, reducing diagnosis delays from eight years to earlier. The research uses a national data repository to develop a predictive model for AS detection, empowering GPs to refer patients more efficiently.
Researchers at CZ Biohub SF developed a new diagnostic method for LRTI in children, leveraging machine learning and genomics to identify pathogens with high accuracy. The method uses metagenomic sequencing data to analyze gene expression and microbial abundance, providing a more holistic approach to diagnosing the condition.
A new study published in Neurology shows that younger brain age is associated with superior post-stroke outcomes, suggesting a potential biomarker for stroke rehabilitation. The research team used multi-site data sets and 3D brain structural MRIs to analyze the relationship between brain age and stroke recovery.
Researchers found that simultaneous learning of two tasks enhances a deep-learning model's performance on retrieving precipitation information from satellite data. The new framework uses multi-task learning to improve current estimates of precipitation, outperforming existing approaches.
Max Planck scientists explore the possibilities of artificial intelligence in materials science, discussing how combining physics-based modeling with AI can unlock complex material designs. The research focuses on overcoming limitations of traditional methods and handling sparse, noisy data.
A new study published in The European Heart Journal – Digital Health found that smart watch data can predict a higher risk of developing heart failure and irregular heart rhythms. Researchers used machine learning to analyze ECG recordings from wearable devices and identified extra beats as indicators of increased cardiovascular risk.
Researchers developed a high-speed prediction model combining physical simulations and machine learning, achieving high accuracy without compromising computation time. The technology uses correspondence between input physical conditions and abstract data space handled by machine learning algorithms.
A team of experts at the University of Manchester has received €2.5million funding to design smarter robots that can understand humans' inner feelings through language. The project aims to develop robots that can engage meaningfully with older generations, helping them in everyday life.
MIT researchers discovered unconventional activation functions that enable optimal neural network performance, leading to better classification on various datasets. The findings suggest that selecting the correct activation function can significantly improve data accuracy in machine learning applications.
A recent study found that genetic risk scores are more predictive of Alzheimer's disease in adults over 65 than age. The study used machine learning models to rank risk factors, including household income, which emerged as an important risk factor. The findings suggest considering genetic information when working on Alzheimer's disease.
Researchers at the University of Washington developed GlucoScreen, a new system that leverages smartphone capacitive touch sensing to measure blood glucose levels. The system's accuracy is comparable to standard glucometer testing, making it potentially less costly and more accessible for widespread screening.
A new AI tool, CLEAN, can predict enzyme functions based on amino acid sequences, outperforming leading state-of-the-art tools in accuracy and reliability. The tool was developed using contrastive learning and verified experimentally with both computational and in vitro experiments.
A recent study suggests that cardiac sphericity, or the roundness of the heart, may be a useful indicator for diagnosing cardiovascular conditions. The research used machine learning to analyze medical images and found a link between increased sphericity and future heart troubles.
A machine learning model developed by Colorado State University researchers has been tested with forecasters at the Storm Prediction Center to improve medium-range severe weather forecasts. The tool provides a probabilistic measure of hazardous weather events, such as tornadoes and hail, four to eight days in advance.
Researchers used explainable AI to analyze gut microbiomes for CRC biomarkers, discovering four distinct subgroups with unique bacterial profiles. This approach shows promise for a more personalized microbiome exploration and potential disease subgroup identification.
Researchers developed a deep learning approach to recognize and predict motion using vector-based relative change in position. The method, VecNet+LSTM, scored higher than other frameworks in recognizing motion and predicting future movements. This study has implications for machine learning in video analysis and artificial intelligence.
Researchers at NIST have developed a new method of digitally simulating hurricanes using AI techniques, accurately representing the trajectory and wind speeds of real storms. The simulations can help develop improved guidelines for building design in hurricane-prone regions.
A team led by Professor Yoshihiro Yamazaki from Kyushu University discovered the chemical innerworkings of a perovskite-based electrolyte developed for solid oxide fuel cells. By combining synchrotron radiation analysis, large-scale simulations, machine learning, and thermogravimetric analysis, they found that protons are introduced at...
Researchers developed a VR imaging system to measure neural activity in mouse brains during behavior, revealing abnormalities in cortical functional network dynamics associated with autism. The system successfully distinguished between autism model mice and wild-type mice based on their brain network patterns.
A new technique combines machine learning with short-wave infrared fluorescence imaging to detect precise tumor boundaries with higher accuracy than traditional methods. The approach achieved a remarkable per-pixel classification accuracy of 97.5 percent and demonstrated robustness against changes in imaging conditions.
Scientists have identified distinct mechanisms by which people remove information from their working memory and found that forgetting requires much effort. They also discovered four brain networks that activate when memories are maintained or purged through different mechanisms.
Scholars developed an AI program using symbolic regression to improve mass inferences of galaxy clusters. The new equations extract jelly at the center and concentrate on doughy outskirts for reliable mass inferences.
A team of researchers used a genetic algorithm to discover an organic catalyst for the Morita–Baylis–Hillman reaction, which outperformed traditional catalysts. The computational method suggested new molecular structures that were not present in the initial population, leading to a novel discovery.
A new study using ChatGPT highlights the challenges of AI in academia, including concerns over academic dishonesty. However, researchers believe that AI can also be leveraged to improve education by automating administrative tasks and adapting assessments.
Researchers used artificial intelligence to analyze electrocardiogram results from 1.6 million patients, predicting mortality risk with 85% accuracy. The study aims to improve individual care and create a learning health-care system that feeds data back into the healthcare system.
An international team developed a novel method for evaluating AI interpretability methods to decipher the basis of AI reasoning and possible biases. The approach helps users understand what influences AI results and whether they can be trusted, especially in medical applications where AI-powered decisions can impact health and lives.
Researchers developed an image analysis algorithm that can automatically measure Arabidopsis thaliana stomatal aperture with high accuracy and speed. The technology also includes a portable imaging device for non-destructive observation using intact plants, allowing for rapid measurement of subtle changes in stomatal aperture.
Researchers develop unsupervised machine learning algorithm to classify osteosarcoma at diagnosis based on gene expression modules. This approach enables personalized treatment strategies for osteosarcoma patients.
According to Li, machine intelligence is based on a combination of matter, energy, structure, and time, which he calls
Researchers developed a machine learning model that maps graphene-gas molecule van der Waals complex bonding evolution for selective gas detection. The model achieved 100% accuracy in distinguishing between different atmospheric environments, showcasing its potential for environmental monitoring and non-invasive medical diagnosis.
A low-cost, open-source device called Flatburn can measure air pollution using 3D printing or inexpensive parts. Researchers have tested and calibrated it against existing state-of-the-art machines, making it suitable for community groups and individual citizens to track local air quality.
Researchers developed a self-driven lab, AlphaFlow, that uses AI to optimize complex chemical reactions and discover new materials. The system significantly reduces the time needed to develop new chemistries from months to hours.
The TOAST project, led by Aarhus University, aims to develop the Tactile Internet technology and upscale the talent pool for this emerging field. Researchers will focus on edge intelligence, haptic communication, and machine learning to create immersive user experiences.
A University of Illinois project uses AI-powered object recognition to quantify kernel damage in wheat, enabling faster disease analysis and improved resistance. The technology has shown promising results, with potential for an online portal to automate scoring and support breeders in their efforts to eliminate fusarium head blight.
Researchers developed classification models using machine learning to accurately sort different types of compostable and biodegradable plastics from conventional ones. The technique achieved perfect accuracy for larger samples but showed some variability with smaller ones, highlighting its potential for industrial-scale implementation.
Researchers have created a transformer model for Metal-Organic Frameworks (MOFs), allowing for faster results and less data. The MOFTransformer model predicts key properties such as hydrogen storage capacity with improved accuracy.
Researchers developed a deep learning-based AI model to automate cirrhosis identification using large amounts of data from EHRs. The model successfully identified patients with cirrhosis with a precision of 97%, offering potential for early diagnosis and improved management of the disease.
A cross-disciplinary team developed a convolutional neural network to analyze microscopy images of chromosomes with cohesion defects. The algorithm achieved 73.1% accuracy in classifying new images, streamlining experiments with chromosome analysis.
A University of California, Riverside study found that people's memories for celebrity faces are sharper but less accurate than those for non-celebrity faces. This 'variance-bias tradeoff' suggests that humans prioritize precision over accuracy to accommodate the natural variability in face recognition.
Researchers at the University of Rochester have created a nitrogen-doped lutetium hydride that exhibits superconductivity at 69 degrees Fahrenheit and 10 kilobars of pressure. This breakthrough material has the potential to enable practical applications, as it reduces the required pressure for superconductivity to occur.
A new AI study suggests dementia prevalence rates in India could be up to 8.44%, affecting 10.08 million older adults, higher than previous estimates and comparable to countries like the US and UK
The UTSA ScooterLab will collect data on riders' mobility, context and environment to improve sustainable transportation solutions. The project aims to transform the way we think about micro-mobility.