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 a machine learning-based method to estimate chemical attributes of spice extracts by measuring fluorescence emitted by polyphenols and flavonoids. The approach integrated data from multiple dilution levels, achieving highly accurate results.
A combination of 11 proteins can predict long-term disability outcomes in multiple sclerosis, making it possible to tailor treatments to individual patients. The study identified a specific protein, neurofilament light chain, as a reliable biomarker for disease activity.
A University of Oxford study has used machine learning to bridge the 'reality gap' between predicted and observed behavior in quantum devices. The approach enables accurate predictions and informs compensation approaches to mitigate unwanted effects of material imperfections.
A new app, MindEar, has shown promising results in reducing tinnitus symptoms in just weeks through a combination of cognitive behavioral therapy, mindfulness, and relaxation exercises. The app is now available for individuals to trial on their smartphones, offering hope for millions affected by tinnitus.
Researchers developed a deep-learning-based algorithm to identify tissue cellular neighborhoods (TCNs) in breast and colorectal tumors, revealing new fibroblast-enriched and granulocyte-enriched TCNs associated with high-risk disease subtypes. The study aims to better understand cancer evasion mechanisms and potential therapeutic targets.
Boleslaw Szymanski and his team developed a clustering method called SpeakEasy2: Champagne to group molecular data, which showed consistent performance across diverse applications. The method was tested on bulk gene expression, single-cell data, protein interaction networks, and large-scale human networks data, revealing its effectiven...
A recent study investigates how deceased individuals' consent affects societal acceptance of digital resurrection. The research found that explicit consent increases acceptability by two points, while 59% of respondents disagree with their own digital resurrection, highlighting the need for clear legal regulations on the subject.
A new study uses machine learning and satellite imagery to create the first global map of large vessel traffic and offshore infrastructure, finding a remarkable amount of activity previously unknown. The analysis reveals that industrial fishing and transport activities are concentrated around Africa and south Asia.
The iStar tool uses advanced techniques to capture both detailed views of individual cells and broader tissue patterns, enabling doctors to diagnose cancers that might otherwise go undetected. It also predicts gene activities at near-single-cell resolution, paving the way for molecular disease diagnosis.
A prognostic study of 1,000 very preterm infants found predictive modeling can identify those at risk for cognitive impairment. This allows for targeted interventions to be implemented early, potentially improving outcomes.
A study of 100 adults found a significant link between poor sleep quality and higher systolic blood pressure, kidney function impairment, and liver function issues. The research highlights the potential of home-based electroencephalography for assessing sleep quality.
A study of 70 autistic youths in psychiatric hospitals found wearable biosensing and machine learning can identify impending aggressive behaviors. The findings suggest a potential for developing adaptive intervention systems to prevent aggression.
A new study from the University of Ottawa and Copenhagen Business School finds that removing human bias from organizational processes can lead to autonomous systems that create their own environments. This could limit human ability to recognize automation biases, notice environmental shifts, and take action.
A new AI system, Coscientist, has demonstrated its ability to autonomously learn about Nobel Prize-winning chemical reactions and design successful laboratory procedures. The system achieved this in just a few minutes, outperforming human chemists in some cases.
Deep learning models have been developed to analyze X-ray diffraction data, improving the search for new materials. The models can sift through large amounts of data generated by X-ray diffraction techniques, providing valuable insights into material structure and properties.
Researchers are leveraging AI/ML to improve health outcomes, but human judgment is still crucial for model selection and data quality. Explainable AI can enhance transparency and acceptance in clinical practice.
A study found that clinicians can be fooled by biased AI models even with provided explanations, leading to serious declines in accuracy. While accurate AI models improved diagnostic decisions for some demographics, biased models worsened decisions for others.
Researchers used AI to analyze electroretinogram signals from children's eyes, identifying unique features associated with autism spectrum disorder (ASD). The test, completed within 10 minutes, shows promise for diagnosing ASD more accurately and efficiently than current methods.
A study by Virginia Tech researchers found that ChatGPT can identify location-specific environmental justice challenges in large, high-density population areas. However, the tool was limited in its ability to provide contextualized information on local environmental justice issues.
A team of MIT researchers has created a computational model that can calculate the structures of transition states in chemical reactions much more quickly than traditional methods. The new model uses machine learning and can generate accurate predictions for thousands of reactions, enabling chemists to design new catalysts and fuels.
Researchers developed a machine-learning model to predict the risk of visual impairment in people with severe shortsightedness. The study used a dataset of 967 Japanese patients and found that the logistic regression-based model performed well at predicting visual impairment at 5 years.
The EXPLORE toolkit offers interactive visual analytics and machine learning to analyze galaxy data, identify unusual stars, and visualize the lunar surface. Users can create immersive experiences, including 3D models of the Moon and interactive sky maps.
A new study from MIT shows that computational models trained on auditory tasks display an internal organization similar to the human auditory cortex. Models trained on diverse tasks and background noise more closely mimic brain activation patterns.
The São Paulo School of Advanced Science on Precision Livestock Farming aims to empower graduate students and early-career researchers with machine learning, statistical tools, and database systems. The school will provide a platform for them to process vast amounts of data and develop a multidisciplinary approach to PLF.
Researchers have developed an AI-powered system to diagnose autism spectrum disorder (ASD) in children using a single flash of light to the eye. The system uses electroretinography (ERG) to identify specific features that classify ASD, providing a faster and more accurate method for diagnosis than existing tests.
Researchers have developed an AI-powered algorithm that can detect the risk of atrial fibrillation (AFib) with high accuracy, even in people with infrequent AFib episodes. This new tool has the potential to reduce the incidence of stroke and heart failure by identifying patients at increased risk and guiding targeted interventions.
The São Paulo School of Advanced Science on Quantum Materials will select and support 100 graduate students and young researchers to focus on fundamental, theoretical, and experimental aspects of quantum materials. The program will cover topics like superconductivity, electronic topology, and complex magnetic configurations.
Researchers developed a method to measure microvascular changes in the skin using AI and optoacoustic imaging technology, enabling non-invasive assessment of diabetes severity. The study identified 32 significant changes in blood vessels, which can be used to monitor disease progression.
Researchers from the University of Technology Sydney have developed a portable, non-invasive system that can decode silent thoughts and turn them into text. The technology has been shown to achieve state-of-the-art performance in EEG translation, with an accuracy score of around 40% on BLEU-1.
Researchers developed a background-resistant model to predict wheat Leaf Area Index (LAI) across diverse soil backgrounds, showing substantial improvement in prediction accuracy. The model demonstrated good estimation accuracy for different soil backgrounds and reliably captured seasonal LAI dynamics under various treatments.
Researchers developed a weeklong high school curriculum that teaches color chemistry and AI, improving students' knowledge and motivation. The curriculum uses machine learning to analyze pH levels, showing the connection between chemistry and AI in a practical application.
Researchers developed an adaptable machine learning algorithm to measure driver 'workload' using driving performance signals, enabling real-time adjustments to in-vehicle systems for enhanced safety and user experience. The system can respond to changes in the driver's behavior, status, road conditions, or characteristics.
Researchers create Automatic Surface Reconstruction framework to estimate all possible variations of material surfaces, providing detailed information on catalysts, semiconductors, and battery components. The method reduces human intuition and provides dynamic information on surface properties over time.
Researchers developed a geoknowledge-guided GPT model that extracted location data from Hurricane Harvey tweets with an accuracy rate 76% better than default models. The model recognized complete location descriptions, enabling first responders to reach victims more quickly and potentially saving lives.
Researchers analyzed 13,000 hours of audio data from Okinawan forests before, during, and after typhoons, finding that ecosystems responded differently than expected. The study suggests that developed sites were more resilient to extreme weather than anticipated, but climate change may push these ecosystems to their limits.
A new AI-powered satellite analysis technique reveals the economic conditions of regions with limited data, such as North Korea. The approach combines human input with machine learning to provide detailed economic maps and monitor progress towards Sustainable Development Goals.
Researchers use AI to develop dynamic modeling of brain graphs, capturing dynamics in continuous time for more accurate predictions and personalized treatment of brain diseases. The project aims to track disease development in individual patients and identify biomarkers associated with brain disorders.
Novel Dice loss functions, t-vMF Dice loss and Adaptive t-vMF Dice loss, have been developed to improve image segmentation accuracy in medical images. These new functions outperform conventional formulations and show great potential for critical fields like medical imaging and diagnosis.
Researchers used brain imaging and machine learning to identify distinct patterns of brain connectivity in people with autism spectrum disorder (ASD), taking into account individual differences. The study reveals that certain brain features are shared across subtypes, while others are unique to specific individuals.
Researchers from MIT and ETH Zurich developed a filtering technique to simplify a key intermediate step in MILP solvers, speeding up the process by 30-70% without compromising accuracy. A machine-learning model is then used to pick the best combination of algorithms for a specific optimization problem.
Scientists developed an AI method to track neurons in moving and deforming animals using convolutional neural networks with targeted augmentation. This breakthrough reduces manual annotation efforts by three times, enabling faster analysis of brain activity in model organisms like Caenorhabditis elegans.
A new study reveals AI tools are more vulnerable than thought to targeted attacks that force AI systems to make bad decisions. Researchers developed a software called QuadAttac K to test for vulnerabilities in deep neural networks.
Researchers developed two models to evaluate personal learning performance, combining biometric responses with demographic data. The simplified forecast model showed higher accuracy and reduced overfitting than the full forecast model.
Scientists at National University of Singapore developed a hybrid generative machine learning model to explore structural disorders in complex materials. The model unveiled pathways to material disorder, shedding light on factors affecting piezoelectric response. It also found evidence that domain boundaries maximize entropy.
Researchers used machine learning and mobile mapping systems to analyze the town of Sabbioneta's streets and pavements, identifying accessible trajectories and paths for citizens with motor disabilities. The study demonstrated the effectiveness of AI in assessing physical accessibility in historic urban contexts.
Researchers have developed an AI algorithm that uses people's flavor impressions to make accurate predictions of individual wine preferences. The algorithm combines data from wine labels, user reviews, and sensory tastings to provide personalized recommendations.
This review article surveys existing deep active learning approaches, applications, and challenges in the context of foundation models. Effective query strategies and model training methods are essential for optimizing joint performance.
A researcher has developed a chatbot with expertise in nanomaterials, leveraging document-retrieval method to provide accurate context. The bot uses embedding to categorize and link information quickly, generating factual responses sourced from trusted documents.
A new deep learning-based detection system has been developed by INU researchers to improve the detection capabilities of autonomous vehicles. The system, aided by IoT technology, generates bounding boxes and confidence scores for visible obstacles using point cloud data and RGB images as input.
The contribution grows the open-access resource that scientists use to invent new materials for future technologies. Researchers can now focus on promising materials with improved fuel economy in cars, more efficient solar cells, or faster transistors.
A recent study has uncovered a remarkable connection between individuals' musical preferences and their moral values, shedding light on the influence of music on morality. The research found that specific lyrics and audio features from favorite songs can predict moral values such as Care and Fairness and Loyalty.
Researchers analyzed seismic data from the region since 2014, detecting a 8-month long crustal seismicity transient suggesting a preparation process before the M 7.8 Kahramanmaraş earthquake. This highlighted high and increasing seismic hazard in the area.
Researchers developed GraphNovo, a program that provides accurate understanding of peptide sequences in cells, improving immunotherapy for unique cases. The AI model enhances de novo peptide sequencing accuracy, filling gaps with precise mass data.
A novel contamination-detection method enables faster and safer T-cell therapy production, reducing the risk for patients and speeding up treatment. The method uses cutting-edge technology to identify harmful microorganisms within 24 hours.
A team of researchers used AI to optimize thermal aging schedules for nickel-aluminum alloys, resulting in stronger materials at high temperatures. By analyzing unconventional heat treatment patterns, the team discovered a two-step schedule that outperformed conventional methods.
Researchers have developed a novel approach using tensor networks to bridge quantum concepts with machine learning, enabling efficient construction of probabilistic models from quantum states. This framework offers enhanced interpretability comparable to classical probabilistic machine learning.
Neurons in the ventrolateral prefrontal cortex (VLPFC) work together to process social interactions by combining facial and vocal information. The study found that individual neurons did not exhibit strong responses to expressions or identities, but population-level activity could be decoded to reveal the identity and expression in vid...
Researchers developed three diffractive deep neural networks using orbital angular momentum to recognize objects in images, achieving accuracy comparable to wavelength and polarization-based models. The technology has potential for real-time processing applications like image recognition and data-intensive tasks.
A study conducted at Carnegie Mellon University suggests that combining human curation with automated recommender technology can improve user engagement in online news outlets. The research found that algorithmic recommendations outperformed human-curated choices on average, but the human editor did better under certain conditions.
Researchers have successfully mapped the entire HLA class II landscape, predicting how pathogens are displayed on cell surfaces. The mapping reveals that multiple HLA variants play essential roles in autoimmune disorders and organ rejection, highlighting their potential for developing immunotherapy treatments.