Research using machine learning to solve scientific problems. Covers AI quality control, sea-surface temperature estimation, label-free cell analysis, and cyber defense.
Researchers have developed a novel AI technology, Swarm Learning, to analyze big data in decentralized fashion, enabling private and collaborative analysis of scientific data. The approach combines machine learning with blockchain technology, allowing for secure information exchange and optimized parameters.
According to a review article in Science Robotics, researchers are making progress in learned robot manipulation, which enables robots to adapt to changing stimuli. The authors propose nine promising areas for future exploration, including representation learning, modular design, and task/skill customization.
Researchers developed a machine learning model that accurately predicted cardiac arrest risk by combining timing and weather data. The results showed that Sundays, Mondays, public holidays, winter, and low temperatures were associated with higher risks of cardiac arrest.
Researchers developed a new multi-modal image fusion method based on supervised deep learning to enhance image clarity, reduce redundant features, and support batch processing. The method achieves state-of-the-art performance in visual quality and quantitative evaluation metrics, improving medical diagnosis accuracy.
Researchers developed a new method to test AI algorithms' decision-making processes by presenting them with carefully designed synthetic data. The technique, called Global Importance Analysis, revealed that AI models consider more factors beyond just sequence length, such as RNA folding and motif proximity.
Researchers at TU Darmstadt created an interactive typeface, AdaptiFont, that adjusts font styles to increase reading speed. The system uses machine learning to generate personalized fonts based on individual users' preferences.
A new AI model precisely replicates human touchscreen typing by simulating eye and finger movements, making it easier to optimize keyboard designs for better typing. The model can also account for different user types, including those with motor impairments, to develop personalized typing aids.
Researchers use aerodynamic levitation and laser heating to suspend small samples of refractory oxides in mid-air, allowing for precise data collection. Machine learning algorithms are then used to predict structural changes and interactions between atoms at high temperatures.
Researchers developed a modular robot that autonomously adapts to its environment, achieving optimal behavior without a central controller. The robot learned to navigate and maintain behavior even with damage, paving the way for miniaturized robotic materials for various applications.
A study found that deep neural networks can accurately predict lung cancer type from CT scans, identifying new associations between genes and imaging features. This approach increases radiologists' confidence in assessing tumor types, informing individualized treatment planning.
Federated learning, a new approach to training AI models, is found to have a significantly greener impact than traditional methods. By distributing training across multiple devices, the energy consumption and CO2 emissions are reduced. This method has important privacy benefits as well, keeping data local and secure.
A team of researchers developed an AI technique to accurately detect sarcasm in social media text. The model identifies patterns and cue words that indicate sarcasm, enabling more effective customer feedback analysis.
A team of scientists at IBEC and Stanford University reveals that the brain's ability to autonomously learn reflects nature more closely than previously thought. This discovery has implications for improving memory deficits in humans and building new AI systems with advanced memory capabilities.
A team from the University of Bristol's QETLabs developed an algorithm that uses machine learning to reverse engineer Hamiltonian models and formulate approximate models for quantum systems. This breakthrough enables the automated characterization of new devices, such as quantum sensors.
Researchers developed DeepShake, a deep spatiotemporal neural network trained on over 36,000 earthquakes, which analyzes seismic signals in real time and provides advanced warnings of strong shaking. The model was tested using the 2019 Ridgecrest earthquake, sending simulated alerts up to 13 seconds prior to high-intensity ground shaking.
Active learning is used to identify promising organic molecules for efficient solar cells by iteratively deciding which data to learn from. This approach allows the algorithm to efficiently explore a vast molecular space and continuously improve its performance.
A research group from Tohoku University has developed an AI model that can accurately identify flooded buildings using news media photos within 24 hours of a disaster. The model achieved an 80% estimation accuracy, showcasing the potential for rapid damage mapping and accelerated disaster response.
A neural network developed by Skoltech researchers improves credit scoring using transactional banking data, surpassing existing models. The EWS-GCN model processes large-scale temporal graphs directly and aggregates information to predict target client credit ratings.
Researchers at KAUST developed a new technology that increases machine learning speed on parallelized computing systems by five-fold. This 'in-network aggregation' method uses readily available programmable network hardware to provide dramatic speed improvements.
A new system enables robots to recognize human workers and predict their poses, providing a safer and more efficient working environment. This allows robots to work side-by-side with humans on assembly lines without unnecessary interruptions.
Researchers at Rice University have optimized artificial intelligence software to run on commodity processors and train deep neural networks up to 15 times faster than top GPU trainers. The 'sub-linear deep learning engine' (SLIDE) uses hash tables to solve the search problem of matrix multiplication, reducing training time for AI models.
Researchers used X-ray microscopy to analyze lithium batteries and employed machine learning to speed up the learning curve about process that shortens battery life. Infrared microscopy also goes off-grid with new technique, enabling time-sensitive experiments and broadening biological spectromicroscopy scope.
A team of scientists developed a deep learning model that improves air quality estimates by combining satellite and ground-based observations. The model achieved higher spatial and temporal resolution, enabling more accurate predictions of nitrogen dioxide levels in the Los Angeles area.
Southwest Research Institute is developing an AI-based integrated corridor management decision support system for the Tennessee Department of Transportation. The system uses machine learning algorithms to optimize traffic flow and improve collaboration among transportation agencies.
A machine learning study of rock art in Arnhem Land, Australia, has reconstructed the chronology of artistic styles using over 14 million images. The analysis revealed a link between style similarity and time, showing that styles closer in age were also more similar in appearance.
A recent study by researchers at the University of Johannesburg shows how AI can forecast municipal solid waste in a large African city. By using machine learning algorithms and combining data from various sources, including census data and landfill site records, the team was able to predict the city's waste management needs until 2050...
Researchers developed an AI model to detect COVID-19 in chest X-rays with high accuracy, using a large dataset of other X-ray images as a starting point. The tool has the potential to assist doctors in identifying, measuring the severity and classifying the disease.
Researchers developed microswimmers that can change direction by heating tiny gold particles, then learned to navigate through a virtual environment via external control and virtual rewards. The findings suggest an optimal speed is key to navigation, with implications for autonomous tasks and collective behavior in biological systems.
Osaka University researchers use deep learning to improve mobile mixed reality generation, enabling the automatic removal of obstructions and addition of greenery. This technology may revolutionize green architecture and city revitalization by providing real-time visualizations.
MCG faculty adapted traditional curriculum to provide an online Pandemic Medicine Elective, covering topics from SARS-CoV-2 to professional responsibility. Students logged over 6,000 service hours through projects like mask delivery and contact tracing.
A new educational curriculum in longevity medicine for physicians has been developed, outlining the benefits of promoting healthspan and lifespan. The course provides a comprehensive introduction to theoretical and practical basics of longevity medicine, including molecular mechanisms, biomarkers of aging, and geroprotector regimens.
A team of RIT researchers is working on developing an artificial intelligence system that can learn over time and play the popular video game Starcraft II. This project has the potential to advance practical solutions such as self-driving cars, service robots, and other real-world applications.
Researchers at UOC used AI to analyze urban scenes and identify patterns that may lead to accidents, suggesting that complexity and layout are key factors. The technology aims to aid traffic authorities in reducing the likelihood of accidents by providing real-time hazard warnings and optimizing traffic flow.
Researchers have successfully demonstrated a significant speed-up in robot learning time using quantum physics, enabling machines to learn faster and make better decisions. This breakthrough has promising implications for the development of autonomous systems.
Researchers develop compact optical machine-learning decryptors that process information at the speed of light without consuming power. These devices can be integrated on CMOS chips and have a neuron density of over 500 million neurons per square centimeter.
Researchers at The University of Tokyo used artificial intelligence to show that T helpers in the adaptive immune system act like a neural network, optimizing responses to pathogens. The study may lead to improved vaccine development and stronger immune responses.
Researchers at SUTD developed a new connection between exploration-exploitation trade-off in multi-agent AI systems and Catastrophe Theory. This discovery aims to improve the performance of AI systems, such as robotic space missions and healthcare management.
A new deep-learning algorithm, CARRL, is designed to help machines build a healthy skepticism of their measurements and inputs. By combining reinforcement-learning algorithms with deep neural networks, researchers created an approach that outperformed standard machine-learning techniques in scenarios with uncertain and adversarial inputs.
Professor Sun's research focuses on deep learning and meta-learning for recognizing images and videos. Her team is working on a food app that uses AI to track nutrition and achieve a healthy diet, but faces challenges due to cultural diversity.
The study uses natural language processing to extract adverse effects from Spanish health records, highlighting a major health problem and improving clinical decision making. The system learns more effectively with larger corpora, contributing to closing the gap in clinical text mining across languages.
Artificial intelligence has already shown promise in pathology, including image classification and diagnosis of diabetic retinopathy. Future developments aim to further enhance diagnostic accuracy and improve patient outcomes through augmented intelligence. The authors emphasize the need for careful validation, performance monitoring, ...
A new AI tool developed at the University of Gothenburg uses deep learning to analyse microscope images, extracting more details and information than traditional methods. The tool, called Deep Track 2.0, simplifies data generation and allows for real-time analysis and customised information retrieval.
A new bioinformatics tool called PlasmidHawk has been developed by Rice University researchers to track the origin of synthetic genetic code. The tool uses a sequence alignment-based approach and was found to outperform recent deep learning approaches in lab-of-origin prediction, achieving 76% accuracy.
Researchers found that exposure to new experiences dampens established brain connections, allowing for flexible strategy encoding. Novelty triggers neural mechanisms that facilitate learning from new tasks and rules.
A machine learning algorithm helps differentiate between benign and premalignant polyps in CT colonography data, with a sensitivity of 82% and specificity of 85%. The findings suggest a role for machine learning-derived algorithms in boosting the effectiveness of CT colonography as a screening tool for colorectal cancer.
Yingyan Lin, an assistant professor at Rice University, has received a $400,000 NSF CAREER Award to develop more efficient deep learning hardware accelerators. Her goal is to push forward ubiquitous intelligent devices and green artificial intelligence, addressing the gap between complex algorithms and limited resources.
Sun and Tong's project aims to make AI workflows more shareable and replicable. The researchers will further develop the open-source GeoWeaver system into a stable operational platform for NASA's EOSDIS archive.
Researchers developed a deep learning model that accurately predicts lung cancer patients' survival expectancy, significantly better than traditional machine learning models. The model can analyze large amounts of data to understand how various factors affect lung cancer survival periods.
A novel machine learning algorithm developed by Princeton physicist Hong Qin accurately predicts planetary orbits without using traditional physics laws. The technology has potential applications in predicting plasma behavior in fusion facilities, challenging the fundamental role of theories in science.
Researchers at Wyss Institute and Google Research used machine learning to design highly diverse AAV capsid variants that can evade neutralizing antibodies. The approach produced over 57,000 variants with improved functional diversity, potentially leading to improved gene therapies and reduced immunogenicity.
Researchers used metabolomics and machine learning to identify biomarkers for COVID-19 diagnosis and risk assessment. The study found 26 biomarkers that differed between mild and severe illnesses, potentially revealing new clues on how SARS-CoV-2 affects the body.
Researchers at Geisinger Health System developed an AI algorithm using echocardiogram videos to predict mortality within a year. The model outperformed other clinically used predictors and improved cardiologists' prediction accuracy by 13 percent.
Researchers showed that deepfake detectors can be defeated by inserting inputs called adversarial examples into every video frame, which cause AI systems to make mistakes. The attack still works after videos are compressed. Key findings include high success rates of over 99% for uncompressed and 84.96% for compressed videos.
A new study published in the Proceedings of the National Academy of Sciences presents a computationally-based modeling approach that simulates infant language learning. The researchers found that infants do not learn consonant- and vowel-like phonetic categories, but rather learn to distinguish between speech sounds in a more nuanced way.
Naked mole-rats form distinctive, colony-specific chirps that convey individual's social membership and are culturally transmitted across generations. These dialects change when a queen dies and young pups learn the dialect of their adoptive groups.
A new machine learning model, integrated with CT scans of the lungs, surpasses benchmarks in predicting disease severity and supports hospital resource management. Additionally, Owkin develops models to discover coronavirus epitopes that may improve future vaccine efficacy.
A new study published in Nature Communications found that deep learning models surpass standard machine learning approaches in analyzing brain imaging data, generating more accurate representations of the human brain. This is particularly beneficial for complex problems requiring large datasets and advanced analysis.
Researchers developed a multi-fidelity graph network approach to predict material properties with improved accuracy, enabling predictions for disordered materials. The new method reduced mean absolute errors by 22-45% compared to traditional approaches.
Researchers at UCLA have developed Diffractive Deep Neural Networks (D2NNs) for all-optical object classification, achieving higher accuracy than individual constituent D2NNs and digital AI models. The success of the ensemble learning approach demonstrates the power of combining multiple predictions to obtain a more accurate prediction.
New research identifies a risk of collusive pricing behavior when AI algorithms set prices without observing competitor prices. This can lead to monopolistic price effects and supracompetitive market outcomes. The study suggests that independent AI algorithms can result in these negative consequences.