Research using machine learning to solve scientific problems. Covers AI quality control, sea-surface temperature estimation, label-free cell analysis, and cyber defense.
Researchers designed a new AI model that accurately predicts treatment outcomes for stroke prevention in people with heart disease. The model, called CURE, was trained on a massive dataset of de-identified patient information and fine-tuned to focus on specific health conditions and treatments.
An interdisciplinary team has published guidelines for responsible use of machine learning in science, focusing on transparency and reproducibility. The checklist includes detailed descriptions of machine learning models, data, and experimental designs to ensure integrity and validity.
A machine learning tool called PheNet can identify patients with rare, undiagnosed diseases years earlier, improving outcomes and reducing cost and morbidity. By analyzing patterns in electronic health records, the tool ranks patients by likelihood of having a disorder like common variable immunodeficiency.
Researchers found that chatbots like ChatGPT can provide accurate information on climate change-related topics, but not all responses are reliable. The study compared the chatbots' answers to hazard risk indices generated by the Intergovernmental Panel on Climate Change and found more accuracy with tropical storms than droughts.
Researchers trained a quadruped robot using deep reinforcement learning to learn gait transitions on challenging terrain. The robot transitioned from walking to trotting and then to pronking to avoid falls, demonstrating the emergence of animal-like locomotion.
Researchers developed a device with AI algorithms to analyze suckling strength and pattern in newborns. The system showed improved accuracy over subjective clinician assessments, identifying abnormal patterns that may indicate the need for surgical intervention or improved feeding practices.
Researchers created new maps that accurately describe phosphorus levels in Amazonian soil using artificial intelligence, revealing the region's low concentration of the mineral. This information is crucial for understanding how tropical forests will react to climate change.
A new study by UNC Charlotte Professor Abigail Leavitt LaBella and colleagues reveals that yeasts do not fit the 'jack of all trades, master of none' adage. They found that yeasts with a broad ability to metabolize different carbon sources are actually efficient growers.
A machine learning system called AI-MARRVEL, developed by Baylor College of Medicine, has shown promising results in diagnosing rare Mendelian disorders. The system consistently ranked diagnosed genes as the No. 1 candidate in twice as many cases than other benchmark methods.
A new branch of AI, causal machine learning (ML), can estimate treatment outcomes better than traditional methods. This enables personalized treatment strategies and improved patient health.
A new study from the University of Illinois explores the use of hyperspectral imaging and explainable artificial intelligence to assess sweet potato attributes, leading to more informed decision-making and higher-quality products. The results can help industry professionals understand the significance of different features in predictin...
Researchers developed an AI technique to estimate brain age based on EEG scans, making early screening for degenerative brain diseases more accessible. The technology uses machine learning to analyze brain waves and can be used as a low-cost screening tool for large numbers of people.
Using AI software called SLEAP, scientists at Salk Institute are designing climate-saving plants with optimized root systems that can store more carbon. This approach enables researchers to analyze plant features and connect desirable traits to targetable genes, accelerating the development of carbon-capturing plants.
Researchers from Drexel University have developed a machine-learning algorithm that can extract and recognize digital fingerprints of AI-generated video, including Stable Video Diffusion, Video-Crafter, and Cog-Video. The algorithm can be trained on just a few examples of new AI generators to detect them.
A Cleveland Clinic study uses reinforcement learning to design antibiotic regimens that minimize resistance and maximize susceptibility. The AI model predicts the most efficient treatment plans for multiple strains of E. coli, informing hospital-wide infection management.
Researchers at the University of Pennsylvania are working on a decision-support system to help clinicians navigate conflicting clinical trials. The system will use artificial intelligence to analyze study results and provide recommendations, aiming to improve patient outcomes and inform treatment decisions.
Researchers at ETH Zurich have developed an AI algorithm that can design new active pharmaceutical ingredients by analyzing protein structures. The algorithm generates blueprints for potential drug molecules that increase or inhibit protein activity, reducing the need for lengthy discovery processes and minimizing side effects.
Researchers built more accurate cardiovascular risk models using machine learning, finding that women are underdiagnosed due to sex-neutral criteria. The study used the UK Biobank dataset and found that electrocardiogram (EKG) tests were most effective in improving detection of cardiovascular disease in both men and women.
A new framework mitigates bias in machine classification by evaluating under different fairness metrics and inferring specific bias terms from data. The framework substantially reduces bias in classification outcomes while preserving accuracy across seven datasets and 21 classifiers.
Researchers used machine learning to analyze over 1.09 million potential metabolite-receptor pairs and predict the likelihood that each interaction contributed to Alzheimer's disease. The study found a protective metabolite called agmatine interacting with a receptor called CA3R in brain cells, reducing CA3R levels and lowering phospho...
A new AI model generates realistic images of single cells, which are used as synthetic data to train an AI model for better cell segmentation. The researchers found that providing a more diverse dataset during training improves performance.
A team of researchers at MIT has developed a new method to model the irrational behavior of humans, which can be used to predict their future actions. By analyzing an agent's previous decisions, the technique infers its computational constraints and adapts to human collaborators' weaknesses.
A Wayne State University researcher is working to transform code review tools and workflows to address participants' challenges in understanding and communicating code changes. The goal is to produce high-quality feedback using empirical methods, machine learning, and natural language techniques.
A new neural network-based system called BONES improves video streaming quality by up to 13% while reducing download size, according to NJIT researcher Jacob Chakareski. The system uses a mathematical function to optimize data transmission and can be applied to various platforms, including popular video conferencing services.
A new AI-based approach, PERCEPTION, uses single-cell transcriptomics to predict patient response to cancer drugs and monitor resistance. The approach was validated in three clinical trials for multiple myeloma, breast, and lung cancer, with promising results.
Researchers at the University of Cologne found that training can optimize word recognition, leading to improved reading efficiency. The 'Lexical Categorization Model' uses behavioral findings to predict brain activation and separates known words from unknown letter combinations.
Scientists have found evidence of a Bragg glass phase in a crystal using machine learning and X-ray technology. The study provides insight into the nature of glasses and their unique properties, which could inform material design.
Researchers from the University of Cambridge used AI to identify compounds that block alpha-synuclein aggregation, a key step in treating Parkinson's disease. This breakthrough could lead to faster development of new treatments for the condition, which affects over six million people worldwide.
Researchers have created a probabilistic computer prototype that combines CMOS with stochastic nanomagnets, achieving superior computational performance and energy-efficiency. The new technology reduces area and energy consumption by four and three orders of magnitude compared to current CMOS circuits.
Researchers at NC State University developed a fabric-based touch sensor that can control electronic devices through touch, utilizing machine learning algorithms to improve accuracy. The device, integrated into clothing, activates and controls functions like mobile apps, passwords, and video games with gestures on the sensor.
Researchers have developed an active learning strategy to accelerate the synthesis of high-performance engineered biochar with enhanced CO2 uptake. The approach nearly doubled CO2 capture performance, showcasing its transformative impact.
Researchers found that ~60% of tissues exhibit a significant negative correlation between age and stemness score, indicating a pan-tissue decline in stemness. This study adds weight to the idea that stem cell deterioration contributes to human aging, with hematopoietic stem cells from older individuals showing higher stemness scores.
A research team has successfully created a new dimension in photonic machine learning by incorporating sound waves, enabling the creation of reconfigurable neuromorphic building blocks. This innovation has the potential to revolutionize computing tasks by providing high-speed and large-capacity solutions.
A new study developed two machine learning models to quantify CD8+ cell positivity and classify the immunophenotype of cancer specimens in patients with non-small cell lung cancer. The models hold promise for identifying patients who may benefit from immunotherapy.
Researchers at Bar-Ilan University developed a new AI confidence measure that distinguishes between high- and low-confidence decision making in deep learning architectures. This breakthrough enables the creation of safer and more reliable autonomous vehicles by prioritizing human intervention when confidence levels are lower.
Scientists used a neural network to analyze massive particle collision data from the ATLAS detector, marking the first use of this technique in a collider experiment. The method identified an anomaly that may indicate the existence of an undiscovered particle.
A novel machine learning model has been developed to characterize material surfaces, accurately predicting key electronic properties. The model, which employs artificial neural networks and transfer learning, shows great promise for exploring new materials with superior properties.
Researchers used fMRI and predictive modeling to decode emotional dimensions of spontaneous thoughts, revealing the involvement of default mode, ventral attention, and frontoparietal networks. The study's findings hold promise for daydream decoding and potential applications in mental health.
A recent machine learning study has discovered a surprising link between wildfires in the western United States and hailstorms in the central US. The research, led by Jiwen Fan, used ML algorithms to analyze vast datasets spanning two decades, predicting hail storms with remarkable accuracy.
Lehigh University researcher A. Emrah Bayrak explores best practices for human-AI collaboration in complex design tasks, aiming to maximize productivity and job satisfaction. His project uses models that predict human decision-making and combines it with AI's training data analysis to determine strategies for division of labor.
Researchers created a system called Holodeck to generate interactive 3D environments, leveraging language models like ChatGPT to control it. The system outperformed earlier tools in evaluating realism and accuracy, with human evaluators preferring its outputs across various indoor environments.
Chemists develop new reactions using model systems and substrates to demonstrate versatility. A new computer-aided method reduces subjective bias by analyzing real pharmaceutical compounds' complexity and structural properties. This improves data quality and facilitates machine learning applications.
A study found a significant association between hospital readmission after fracture surgery and underlying medical conditions. Gait analyses also offered valuable insights into injury impact on locomotion and recovery, optimizing rehabilitation strategies.
A new machine-learning approach outperforms human testers in generating diverse prompts that trigger a wider range of undesirable responses from chatbots. The technique provides a faster and more effective way to ensure the safety of large language models.
A new machine learning method called scGHOST has been introduced to identify single-cell 3D genome subcompartments and connect them to gene expression patterns. This can reveal the spatial organization of chromosomes within the nucleus, shedding light on how DNA structure influences gene expression and disease processes.
A study from the University of Gothenburg found that patient images are severely lacking in scientific articles on atopic dermatitis. This lack of visual aids hinders patients' ability to make informed decisions about their care, as they struggle to understand complex medical terminology and figures. The absence of images also affects ...
A new study by Carey Morewedge and colleagues found that people are more likely to recognize bias in algorithmic decisions than their own. This is because algorithms can codify and amplify human bias, but also reveal structural biases in society. The research suggests ways to increase awareness of biases and correct them.
A new statistical-modeling workflow can quickly identify molecular structures of products formed by chemical reactions, accelerating drug discovery and synthetic chemistry. The workflow also enables the analysis of unpurified reaction mixtures, reducing time spent on purification and characterization.
A novel photonic computing architecture has been developed for tens-of-task lifelong learning, surpassing existing electronic neural networks in capacity and energy efficiency. The L2 ONN demonstrates extraordinary learning capability on challenging tasks, such as vision classification and medical diagnosis.
Researchers at UMass Amherst have identified the most effective AI method for monitoring insect populations using bioacoustics, with deep learning models achieving high accuracy. The study found that machine and deep learning are becoming the gold standards for automated bioacoustics modeling.
A novel machine learning approach reveals complex relationships between hotel service attributes and customer satisfaction, providing actionable insights. The study's IML-DAA model achieves unparalleled accuracy in predicting customer satisfaction, elucidating the impact of specific service attributes on overall guest contentment.
Researchers developed an AI model to detect viable tumor cells in osteosarcoma patients, improving prognosis predictions. The model showed comparable detection performance to pathologists and reduced inter-assessor variability, enabling timely assessment.
Researchers used AI to investigate MOF-based membranes for helium extraction, revealing critical factors influencing separation performance. The study identified pore limiting diameter and void fraction as key physical features determining membrane selectivity and helium permeability.
Researchers from the University of Tokyo have developed a physics-based predictive tool that quickly identifies stable intercalated materials for advanced electronics and energy storage devices. By analyzing over 9,000 compounds, the tool uses straightforward principles from undergraduate chemistry to predict host-guest stability.
Researchers from the University of Washington created an AI algorithm to analyze infant poses using limited training data. By leveraging generative AI, they were able to produce high-quality results, enabling parents to monitor their babies' daily activities and detect potential health issues early.
Researchers found that neural networks use a similar path to chart their way from ignorance to truth when presented with images, despite varying network designs and training recipes. This commonality holds the potential for developing more efficient image classification algorithms, reducing the computational power required by AI systems.
Researchers analyzed over 3,500 river basins worldwide, finding that precipitation was the sole determining factor in only 25% of flood events. Soil moisture and air temperature were decisive factors in around 10% and 3% of cases, respectively. The study suggests that more extreme floods are caused by multiple factors interacting.
Researchers develop AI-powered method to rapidly predict multiple protein configurations, understanding protein dynamics and functions. This breakthrough has the potential to revolutionize drug discovery by uncovering more targets for new treatments.
A new MIT-derived algorithm corrects coarse climate model predictions by 'nudging' them toward more realistic patterns, leading to more accurate forecasts of extreme weather events. The approach uses machine learning and dynamical systems theory to improve the resolution of large-scale climate models.
A study published in Critical Care identified eight different trauma phenotypes associated with lower in-hospital mortality when treated with tranexamic acid. The researchers used a machine learning model to analyze data from over 50,000 patients and found subgroups of patients who received no benefit from treatment.