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 trained on clinical text notes can effectively categorize patients into 10 risk groups, allowing for targeted care. The study found that patients in lower-risk groups had lower rates of lung inflammation and were less likely to receive antibiotics or chest X-ray referrals.
A new CAR T cell design approach using machine learning and artificial intelligence is being developed to improve cancer treatment. The project aims to create a hybrid knowledge- and data-driven approach to guide the design of immunotherapeutic cells.
Researchers at Duke University have discovered a class of compounds called argyrodites that could lead to the development of safer and more efficient solid-state batteries. The materials' unique crystalline structures allow for fast ion conduction, making them promising candidates for energy storage applications.
A team of researchers developed a deep-learning model called WearNet that uses Fitbit data to detect depression and anxiety. The study found that WearNet performed better than state-of-the-art machine learning models in detecting these conditions, producing individual-level predictions.
This field uses trial-and-error learning with natural selection to solve complex reinforcement learning tasks, but requires significant computational resources. Researchers can enhance its efficiency by improving encoding, sampling, search operators, algorithmic frameworks, and evaluation methods.
Researchers used machine learning to identify 'synthetic extreme' DNA sequences that are active in humans but not fruit flies. These rare sequences have potential practical applications in biotechnology and biomedical research.
A new machine learning-based model predicts individual cardiac surgery patient mortality risk with improved performance over current population-derived models. The model uses electronic health records to provide personalized risk assessments, offering a significant advantage over existing benchmarks.
A machine learning model uses patterns in mineral associations to predict previously unknown mineral occurrences, including geologically important minerals like uraninite and rutherfordine. The model also identified promising areas for critical rare earth element and lithium minerals.
New research from ESMT Berlin suggests that using machine-based predictions can improve overall accuracy of human decisions, but also increase the likelihood of certain errors and the human's cognitive effort. The study highlights the importance of collaboration between humans and machines to maximize complementary strengths.
An AI developed at TU Wien has shown to suggest appropriate treatment steps in cases of blood poisoning, outperforming human decisions. The AI can examine time-varying patient conditions and calculate treatment strategies, increasing cure rates by up to 3%. However, legal aspects and liability need discussion.
A new AI algorithm named CoDE-ACS can quickly and accurately rule out heart attacks in patients, improving diagnosis and reducing hospital admissions. The tool has the potential to reduce misdiagnosis and inequalities in diagnosis across different populations.
A Rensselaer researcher has used artificial intelligence to discover novel van der Waals (vdW) magnets with large magnetic moments. These two-dimensional vdW magnets have the potential to advance science and technology in data storage, spintronics, and quantum computing.
Researchers used AI to analyze speech patterns of patients with Parkinson's disease, finding they spoke in shorter sentences with more verbs and fewer common nouns. The study suggests potential early detection methods for the condition through conversational analysis.
Researchers found that machine-learning models trained with descriptive data label rule violations more harshly than humans, leading to potential serious implications in the real world. This study highlights the need for careful consideration of data labeling and training methods to ensure fairness and accuracy in AI decision-making.
A new advanced artificial intelligence system has been developed in the UK that can accurately identify protein patterns within individual cells. The HCPL system uses a deep-learning model to quickly and accurately determine subcellular structures where proteins are present.
A team of researchers developed an unsupervised entity alignment framework to improve knowledge graph search, avoiding human labor. The framework outperformed most competitors on precision and recall, scoring higher overall across multiple datasets.
A team of researchers has developed an AI tool called CRANK-MS that uses neural networks to analyze biomarkers in patients' bodily fluids and predict Parkinson's disease onset with an accuracy of up to 96%. The tool may help identify early warning signs for the disease, which can be challenging to diagnose.
A new video anomaly detection algorithm (COVAD) uses content-based attention to focus on objects in frames, improving performance over baseline models. The algorithm also refines the memory module for normal behavioral patterns.
Researchers developed EmbryoNet, an automated image analysis software that uses AI to detect and classify developmental defects in fish embryos. The software outperforms human experts in terms of speed and accuracy, making it a valuable tool for investigating the mechanisms of drug action and studying embryonic development.
Researchers develop a new online learning algorithm that enables the training of larger spiking neural networks with six million neurons. This allows for faster and more efficient processing of tasks such as speech recognition and object detection.
Dr. Miaomiao Zhang's research aims to automate late mechanical activation detection from cardiac magnetic resonance images using machine learning and AI techniques. This could lead to more accurate placement of CRT electrodes and improved patient outcomes.
Researchers from Integrated Biosciences developed an AI platform to discover novel senolytic compounds, a class of molecules targeting age-related processes. The platform identified three highly selective and potent compounds with favorable medicinal chemistry properties.
Researchers at EPFL have computationally designed novel protein binders that attach seamlessly to key targets, including the SARS-CoV-2 spike protein, using deep learning-generated 'fingerprints' to characterize millions of protein fragments. This method demonstrates therapeutic potential for rapidly designing protein-based therapeutics.
Nearly half of COVID-19 patients develop secondary bacterial pneumonia, which can lead to death. Machine learning analysis found that this complication offsets the mortality rate from the viral infection itself.
The UMD-led TRAILS institute will develop AI technologies that promote trust and mitigate risks through broader participation, new technology development, and informed governance. The institute aims to create AI systems that align with values and interests of diverse groups, leading to increased transparency, reliability, and accountab...
Researchers at Drexel University developed a machine learning model to predict Philadelphia's future energy use based on zoning decisions and building characteristics. The model uses two machine learning programs to tease out patterns from massive datasets and make projections about future energy consumption.
The National Science Foundation has awarded Columbia University a $20 million grant to establish the AI Institute for Artificial and Natural Intelligence (ARNI), an interdisciplinary center focused on connecting AI systems to brain research. ARNI will bring together top researchers from across the U.S. to advance neuroscience, cognitiv...
The CEBRA algorithm captures brain dynamics with high accuracy, allowing for the prediction of complex information such as visual stimuli and arm movements in primates. By analyzing brain signals alone, researchers can decode hidden structure in data, enabling new insights into how the brain processes information.
A study by University of Maryland School of Medicine found that AI algorithms used in medical imaging lack demographic data and bias evaluations. This can exacerbate existing health disparities worldwide. Researchers encourage future competitions to prioritize fairness among different groups.
A new machine learning model can automatically translate Akkadian text written in cuneiform into English, with the first version using Latin transliteration achieving satisfactory results. The program is effective for translating short sentences and can be used as part of a human-machine collaboration to correct and refine its output.
Researchers developed GAME-Net, a graph neural network that rapidly evaluates adsorption energy for large molecules like plastics and biomass. The model achieves accuracy comparable to density functional theory (DFT) while utilizing simple molecular representations.
Three high school students co-authored a paper using AI engine PandaOmics to discover new therapeutic targets for glioblastoma multiforme, a common and aggressive malignant brain tumor. The study identified three genes strongly correlated with both aging and glioblastoma as potential therapeutic targets.
Scientists measured brain waves in participants and artificial intelligence systems to reveal similarities in how the brain interprets speech. The study provides a window into the operation of AI systems, which have been advancing rapidly but remain largely opaque.
Researchers found that using positive trigger words can retrain large language models and result in less biased responses. The team analyzed GPT-2's responses to user prompts about different countries worldwide and found a significant impact on the types of adjectives used to describe citizens.
A new AI-powered model, PLPNet, has been developed to accurately diagnose tomato leaf diseases from digital photographs. The model achieved an accuracy of 94.5% while maintaining a speed of over 25 frames per second.
A deep learning model has been developed to classify cancer cells into distinct types, enabling accurate prediction of metastatic potential. The tool achieves high accuracy and is simple to use, making it a promising solution for medical practitioners.
A deep neural network developed by researchers at the University of California - Santa Cruz has been shown to accurately classify particle signals with 99.8% accuracy in real-time. The system can identify weak or noisy signals and pinpoint their source, making it suitable for point-of-care applications.
Researchers at the University of Pennsylvania School of Engineering and Applied Science have created a photonic device that provides programmable on-chip information processing without lithography. This breakthrough enables superior accuracy and flexibility for AI applications, overcoming limitations of traditional electronic systems.
A team of scientists at SLAC and Argonne National Laboratory has developed an algorithm that pairs machine-learning techniques with classical beam physics equations to precisely predict a particle beam's distribution of positions and velocities. This detailed information will help improve experimental reliability, especially at higher ...
A team of researchers from the Complexity Science Hub and Central European University created more-detailed poverty maps for Sierra Leone and Uganda, identifying poor areas with greater accuracy. The maps use a combination of survey information, satellite imagery, and social media data to provide a more accurate picture of wealth distr...
Researchers at Sainsbury Wellcome Centre found that instinctual exploratory runs enable mice to learn a map of the world efficiently. The study demonstrates how biological brains can learn faster and more efficiently than AI agents by focusing on salient objects.
Researchers used machine learning to identify bacterial environmental pH preferences from genomic data, enabling quicker growth of finicky bacteria in labs. This technique may also improve agricultural practices by ensuring inoculants are adapted to local pH levels, supporting native prairie restoration and plant growth.
A new MIT deep-learning system can analyze the internal structure and properties of materials based solely on their surface conditions. The technique uses vast amounts of simulated data to generate reliable predictions, offering a promising solution for engineers seeking non-invasive insights into material properties.
Scientists are developing a high-resolution landslides susceptibility map to forecast future landslides in eastern Oklahoma. The project uses remote sensing data, machine learning, and LiDAR topographic data to understand the causes, mechanics, and associated hazards of landslides.
Recent research on local holographic transformations has improved the tractability of counting problems, enabling new approaches to complexity classification. The study reveals the potential of these transformations as a tool for proving hardness in various frameworks.
A team of researchers developed a self-checking deep learning system that accurately extracts information from gravitational-wave data. The algorithm, called DINGO, has been trained to interpret real data and can cross-check its own results for accuracy.
A machine learning model combines clinical and genetic data to predict the number of eggs retrieved in patients undergoing ovarian stimulation. This study aims to improve IVF procedures by providing personalized predictions, potentially increasing success rates.
A team of researchers from Carnegie Mellon University has developed an AI-based system to help clinicians make decisions quickly and precisely in the ICU. The system, called the AI Clinician Explorer, provides recommendations for treating sepsis based on data from over 18,000 patients.
A team of IUPUI researchers has developed an AI-powered approach to classify insect species, tackling the challenge of discovering new species. The method uses deep hierarchical Bayesian learning to distinguish between known and unknown species, providing insight into their taxonomy and ecosystem impacts.
Researchers develop imaging-based computer algorithms to boost crop-breeding data using self-supervised contrastive learning methods, outperforming conventional supervised approaches. The study uses wheat as a model crop and finds that these new methods can improve plant phenotyping accuracy and scalability.
A recent study analyzing 34 Hollywood films found that stereotypical gender roles persist, with men depicted as aggressive and powerful, and women as loving and caring. However, the analysis also showed a significant increase in female representation over the past two decades.
A team of experts identified 29 sources of bias in AI/ML models for medical imaging, including data collection, preparation, and deployment. The study provides a comprehensive roadmap for mitigating these biases and ensuring fairness, equity, and trust in AI/ML models.
Researchers at Argonne National Laboratory have developed a self-driving laboratory called Polybot, which automates electronic polymer research and frees scientists' time to work on tasks only humans can accomplish. The tool combines AI and robotics to streamline experimental processes and accelerate discovery.
A machine learning framework predicts and quantifies chromosome synthesis difficulties, providing guidance for optimizing design and synthesis processes. The model achieved high accuracy and predictive ability, enabling the development of a Synthesis difficulty Index to explain causes of synthesis difficulties.
Researchers trained a machine learning program using data from over 5 million direct messages, annotated by 150 adolescents who experienced uncomfortable or unsafe conversations. The technology can quickly flag risky DMs and is intended to address rising trends of child sexual exploitation.
Researchers at EPFL have developed a novel imaging technique using cryogenic transmission electron tomography and deep learning to visualize the nanostructure of platinum catalyst layers in fuel cells. This breakthrough reveals the heterogenous thickness of ionomer, a crucial component that influences catalyst performance.
Researchers at the University of Georgia have confirmed evidence of a previously unknown planet outside our solar system using machine learning tools. The discovery highlights the potential for artificial intelligence to enhance scientists' work and speed up analysis, with the potential to dramatically expand exoplanet discoveries.
Researchers tracked social grooming behavior in wild baboons using collars-mounted accelerometers, identifying and quantifying giving and receiving grooming with high accuracy. The study's findings have important implications for the study of social behavior in animals, particularly non-human primates.
A large-scale registry study in Finland identified factors associated with COVID-19 vaccination uptake, including low labor income and mental health issues. The study found that individuals in vulnerable positions were the least likely to vaccinate, highlighting the need for targeted vaccination campaigns.
Researchers used a machine learning model to simulate the behavior of hydrogen atoms at high pressures, discovering a new phase that was missed by previous theories and experiments. The discovery has sparked further investigation into the properties of solid hydrogen under extreme conditions.