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 found that a commercial machine learning tool was moderately successful in predicting hospitalization-related kidney injury, but struggled to identify high-risk patients. The tool performed better for low-risk patients and Stage 1 HA-AKI cases.
Researchers at North Carolina State University are developing a suite of performance metrics to standardize the evaluation of self-driving labs in chemistry and materials science. These metrics aim to compare different lab technologies and identify areas for improvement, ultimately advancing the field and accelerating discovery.
A KAUST research team has developed a machine-learning approach that balances privacy preservation and model performance using ensemble privacy-preserving algorithms. The approach, called PPML-Omics, achieves better model performance while keeping the same level of privacy protection compared to previous methods.
A study published in Oncotarget has identified specific mutational and therapeutic landscapes of pancreatic cancer in the Russian population. By applying machine learning models to full exome individual data, researchers received personalized recommendations for targeted treatment options for each clinical case.
A new study reveals that online images reinforce powerful gender stereotypes, with female and male associations being more extreme among Google Images than within text. The study also found that bias in images is more psychologically potent than in text, leading to stronger biases even three days later.
Researchers from Argonne National Laboratory and the University of Illinois Urbana-Champaign used generative AI to quickly assemble over 120,000 new MOF candidates for carbon capture. The approach combines AI with high-throughput screening, molecular dynamics simulations and theory-based design to identify optimal materials.
Researchers used data from 9,300 miles of Greek roads to develop a machine-learning model predicting crash sites. The model identified key features such as abrupt speed limit changes and incomplete lane markings as predictors of crashes. The study's findings have implications for improving road safety globally.
Researchers at UC Santa Cruz find that popular link prediction metrics are flawed and do not accurately measure algorithm performance. They recommend using a new metric, VCMPR, to benchmark link prediction tasks and highlight the importance of accurate metrics in machine learning decision-making.
Researchers use machine learning to combine mismatched datasets and reduce variation by over 95%, retaining meaningful differences. The approach has potential to provide deeper understanding of normal metabolism and identify biomarkers for disease.
A new depth from focus/defocus approach, DDFS, combines model-based and learning-based strategies to achieve notable improvements in performance and applicability. The proposed method outperformed state-of-the-art methods in various metrics for several image datasets.
Researchers developed sensors using aerogels to detect formaldehyde, a common indoor air pollutant, with real-time detection capabilities. The sensors require minimal power and can distinguish between different gases.
A new paper argues that LLMs can interpret and analyze neuroscientific data, unlocking new insights and potential treatments. Lead author Danilo Bzdok suggests that scientists may not always fully understand the mechanism behind biological processes discovered by LLMs.
Researchers have created a genAI model called 'drugAI' that can generate unique molecular structures for potential drugs with high binding affinity and efficacy. The model outperforms traditional methods in terms of speed and cost, opening up new possibilities for disease treatment.
A machine learning study of 22,000 surgical cases found that models can inform individual prognosis and aid in decision-making to reduce ineffective spine care. The findings suggest that these models may improve outcomes for patients undergoing lumbar disc herniation surgery.
Researchers developed a machine learning framework that encodes images like a retina, reducing sensory encoding challenges in neural prostheses. The actor-model approach produced images eliciting a neuronal response more akin to the original image response.
Assistant Professor Santiago Segarra at Rice University has won the NSF CAREER Award to develop a new approach for AI-powered climate prediction by leveraging structural properties in real-world data. The research aims to create more effective learning algorithms for structured domains.
A new study uses GPT-3 to simplify chemical analysis, achieving accuracy surpassing state-of-the-art models. The approach fine-tunes the language model with curated Q&As, enabling easy and fast discovery in low-data chemistry.
Researchers developed an AI algorithm that evaluates potential working dogs' personalities using data from nearly 8,000 C-BARQ responses. The algorithm clusters responses into five personality types and can help shelters reduce animal returns by matching them with suitable adoptive families.
Researchers from the University of Rochester's Laboratory for Laser Energetics demonstrated an effective 'spark plug' for direct-drive methods of inertial confinement fusion (ICF), achieving a plasma hot enough to initiate fusion reactions. The successful experiments use the OMEGA laser system, with the goal of eventually producing fus...
Researchers have developed a novel approach to determine the age of mosquitoes, which could help improve pesticide strategies and reduce the spread of diseases like malaria. The method uses surface-enhanced Raman spectroscopy (SERS) to analyze biomolecules in mosquito water extract.
Researchers developed a molecular predictor of radiation response using cell line data and machine learning-based approaches, capturing a wider range of biological processes. The new gene signature has the potential to aid decision-making, personalize treatments, and improve outcomes for various types of cancers.
Researchers at University of Virginia Health System developed a new approach to machine learning that identifies drugs minimizing harmful scarring after heart attacks. The tool predicts and explains drug effects for other diseases as well.
A machine learning technique called LASSO was used to analyze blood samples from six countries, identifying seven genes that can predict the risk of developing a secondary respiratory bacterial infection. The findings aim to guide clinicians in making more informed decisions about antibiotic use.
The team proposed a novel machine learning model with data augmentation, which accurately predicts the plastic anisotropic properties of wrought Mg alloys. The model showed significantly better robustness and generalizability than other models, paving the way for improved design and manufacturing of metal products.
Researchers at Tohoku University and Shanghai Jiao Tong University developed a machine learning method to predict the growth of carbon nanostructures on metal surfaces. The approach combines theoretical models with data from chemistry experiments to control the dynamics of material growth, leading to improved quality and efficiency.
Researchers developed a risk calculator to predict individualized risk of right heart failure after left heart pump surgery, tailoring care to each patient. The STOP-RVF calculator uses machine learning and considers variables such as pre-existing health conditions, medications, and demographic information.
Researchers have developed an AI-driven test that accurately diagnoses ovarian cancer in women clinically classified as normal, improving detection of early-stage disease. The test uses machine learning and blood metabolite information to assign a probability of disease presence or absence, offering a more clinically informative approach.
Lehigh University researcher Dominic DiFranzo is developing an AI-powered platform, Social Media TestDrive, to equip youth with skills to navigate social media safely. The platform uses conversational AI tools to provide instant feedback and interactive simulations, empowering students to be 'upstanders' against cyberbullying.
Researchers from the University of Cambridge have developed a robotic sensor that reads Braille at twice the speed of humans, achieving 87% accuracy. The breakthrough uses machine learning algorithms to 'deblur' images and recognize letters, paving the way for potential applications in robotics and prosthetics.
Researchers at UTA developed a novel learning-based framework to predict Alzheimer’s disease progression. The DETree strategy can pinpoint clinical status within the disease spectrum, allowing for more accurate planning and potential applications in other diseases with multiple stages of development.
Researchers developed a predictive model to detect users and content related to Islamic State extremists on social media, identifying potential propaganda messages and their characteristics. The study's findings can help social media companies and law enforcement agencies track and prevent the spread of extremist propaganda.
Researchers have developed a novel optical neural network architecture that achieves nonlinear optical computation by precisely controlling ultrashort pulse propagation in multimode fibers. This approach streamlines the need for energy-intensive digital processes, achieving comparable accuracy with significantly reduced parameters.
The American College of Radiology has issued a joint statement with four other radiology societies to address the development and use of AI tools in radiology. The statement emphasizes the need for increased monitoring of AI utility and safety, advocating for collaboration among developers, clinicians, purchasers, and regulators.
Researchers developed a novel deep learning method to study crystal structure and molecular interactions of perchlorate salts. The analysis revealed that the explosives' nature is linked to chemical bonding and intermolecular interactions.
The TRAILS AI Institute has awarded eight seed grants totaling $1.5 million to advance AI design, development and governance. The funded projects include developing AI chatbots for smoking cessation and designing animal-like robots for autism support.
Researchers propose a novel method to balance data utility and privacy in IoT devices, minimizing added noise for effective privacy protection. The approach considers inherent errors in measurements, optimizing differential privacy standards.
Researchers developed an AI model using CT images to predict lymph node metastasis in non-functional pancreatic neuroendocrine tumors. The model achieved high accuracy, even for smaller tumors, and can help guide surgical decisions.
A new study by George Washington University researchers predicts an escalation of daily bad-actor AI activity globally by mid-2024, posing a significant threat to election results. The study provides the first quantitative analysis of the misuse of AI for disinformation during elections.
A new study by Northwestern University found that lab-trained AI models are easily misled by tissue contaminants, resulting in errors in diagnoses and vessel damage detection. The researchers suggest improving the problem of quantifying and addressing biological impurities in AI models to enhance accuracy.
Researchers developed a novel hybrid approach combining traditional mathematical methods and cutting-edge machine learning to improve EIT analysis of building structures. The new method, called AND, reduces errors in reconstructing foreign objects' position and size compared to conventional EIT methods.
The project aims to develop a standardized Next Generation Science Storyline that can be delivered in any high school classroom, increasing science literacy and critical thinking among students. Pop-omics, a popcorn-based curriculum, will also provide hands-on lessons on machine learning and AI, connecting with the national AITC program.
A new algorithm integrates deep learning and federated learning for accurate channel estimation, outperforming state-of-the-art models in sparse and dense scenarios. The algorithm's use of a user motivation scheme and federated learning framework provides robustness, adaptability, and scalability.
A groundbreaking study uses machine learning to predict treatment resistance in cervical cancer by analyzing genetic mutations. The algorithm identified molecular assemblies driving treatment resistance and pinpointed tumors most susceptible to therapy, resulting in improved patient outcomes.
Scientists at Kyushu University use machine learning to identify promising green energy materials, accelerating the search for hydrogen fuel cell efficiency and expanding material discovery capabilities. Two new candidate materials with unique crystal structures have been successfully synthesized.
CDMENet outperforms fully and semi-supervised counterparts in grape yield prediction using Mean Absolute Error (MAE) and Root Mean Square Error (RMSE). The algorithm's robustness is demonstrated with limited labeled data, highlighting its potential as a cost-effective tool in agricultural yield estimation.
Researchers at UB have developed a new deepfake detection algorithm that reduces biases in facial recognition, with one method classifying videos based on demographics and the other relying on features not visible to the human eye. The algorithms improved fairness metrics and reduced disparities in accuracy across races and genders.
The Lehigh University Plasma Control Group is working on advanced controls and machine learning to improve plasma dynamics simulation capabilities and stabilize superheated gases in future reactors. The goal is to address technological issues with ITER and FPP, ensuring safe and controllable operation.
A team of South Korean scientists used machine learning to discover the secrets of cell variability, revealing a parallel structure that reduces heterogeneity among cells. This finding has far-reaching effects on cancer treatment and improvement in chemotherapy efficacy.
Researchers used quantum support vector machines to classify flow separation and angle of attack with increased accuracy, solving complex problems faster and more accurately than classical methods.
Researchers developed a platform combining automated experiments with AI to predict chemical reactivity, greatly accelerating the design process for new drugs. A machine learning model predicts where molecules will react and how reaction sites vary under different conditions, enabling precise tweaks to complex molecules.
Researchers from HIRI have developed a new machine learning approach using data integration and AI to improve predictions of CRISPRi efficacy, revealing that gene features matter more than guide RNA itself. The study provides valuable insights for designing effective CRISPRi experiments.
A new transparent brain implant has been developed to read deep neural activity from the surface, providing a step closer to building a minimally invasive brain-computer interface. The technology enables high-resolution data about deep neural activity by using recordings from the brain surface and correlating them with calcium spikes i...
Researchers from Bar-Ilan University discover a possible mechanism underlying the brain's efficient shallow learning, enabling it to perform complex classification tasks with similar accuracy as deep learning. The study proposes a wider and higher architecture as a complementary mechanism to deep architectures.
A new European research project, SMARTS, aims to improve air travel efficiency by redesigning flexible airspace sectors using artificial intelligence. The €2million project will create accurate predictive models and develop innovative sector configuration plans to reduce passenger delays, increase productivity, and lower emissions.
Using machine learning and computational modeling, Washington State University researchers found six good candidates for solvents that can extract materials on the moon and Mars usable in 3D printing. The solvents, called ionic liquids, are salts in a liquid state.
A systematic analysis of cancer cells identifies 370 candidate priority drug targets across 27 cancer types. Researchers used machine learning methods to find promising targets and linked them to specific biological markers and genetic features.
The Internet-of-Batteries (IoB) system utilizes IoT principles to gather data from EV batteries, analyzing health and performance, identifying faults, and optimizing usage. Machine learning approaches enhance decision-making for improved battery performance, increased range, and reduced costs.
A new Dartmouth study finds that seasonal snowpacks have shrunk significantly over the past 40 years due to human-driven climate change. The sharpest global warming-related reductions are in the Southwestern and Northeastern United States, as well as in Central and Eastern Europe.
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.