Research using machine learning to solve scientific problems. Covers AI quality control, sea-surface temperature estimation, label-free cell analysis, and cyber defense.
A team at Stanford University developed a machine learning-based method that accelerates battery development for electric vehicles, reducing testing times from almost two years to 16 days. The approach optimizes the charging process, finding better protocols to test and predicting battery performance based on only a few charging cycles.
Army researchers developed a new algorithm that enables collaborative and communication-efficient deep learning, reducing the need for centralized data pooling. The algorithm decreases communication overhead by up to 70% without sacrificing performance accuracy or learning rate.
A new study uses artificial intelligence to identify groups of disease-related genes from huge amounts of gene expression data. The researchers found that the AI model discovered relevant patterns that agree well with biological mechanisms in the body, suggesting potential applications in precision medicine and individualized treatment.
NYU assistant professors Anna Choromanska, Christine Constantinople, and Daniele Panozzo have been awarded Sloan Fellowships for their innovative research in machine learning, brain science, and partial differential equations. The fellowships provide $75,000 over two years to support their research.
Researchers at Texas A&M University developed a tool to identify the source of errors caused by software updates using deep learning. The algorithm, which analyzes performance counters, can find bugs in a matter of hours instead of days.
Researchers are exploring a human-centric approach for 6G communications, emphasizing the need for secure, affordable, and accessible networks that protect users' mental and physical health. The technology will also require innovative solutions such as decentralized blockchain networks and artificial intelligence to enhance performance.
Researchers found that novelty activates dopamine neurons, promoting associative learning in animals and humans. This discovery has implications for improving learning strategies and designing more efficient machine learning algorithms.
A study published in Perspectives on Behavior Science found that AI models can accurately interpret behavioral data, outperforming a popular visual-aid tool. This could lead to better decision-making and tailored interventions for individuals with developmental disabilities, mental health issues or learning difficulties.
A research team from HKU developed a novel deep learning approach to predict disease-associated mutations in metal-binding sites. The approach uses spatial features and physicochemical sequential features to train a model, achieving an AUC of 0.90 and accuracy of 0.82.
A new study by Oxford University Press USA reveals that machine learning can predict long-term risks of heart attack and cardiac death. The research used machine learning to assess cardiovascular risk factors in subjects, aligning accurately with actual events over a 15-year period.
Researchers harness cyber security techniques to give control to those targeted, without resorting to censorship. Algorithms can identify potential hate speech and provide a score for its likelihood, allowing users to view or delete unseen content.
The study found that Norwegian, Swedish, and Danish languages use topicalization to move sentences elements to the front, making them stand out from other languages. This feature allows speakers to emphasize certain words without changing the overall meaning of the sentence.
Researchers are developing a framework to assess the 'intelligence' of AI systems by grading them on problem-solving skills and adaptability. The AIQ test will evaluate systems based on accuracy, time taken, and data requirements.
Researchers have developed a new approach called MACH, which reduces the training resources required for large-scale machine learning models. By dividing data into smaller buckets and using compressed sensing, the system can process 70 million queries and 49 million products in minutes, compared to hours or days with traditional methods.
The article highlights the need for regulators to prioritize continuous monitoring and risk assessment in managing AI/ML-based medical technology. The authors suggest that less emphasis should be placed on planning for future algorithm changes, and instead focus on developing new processes to identify and manage associated risks.
Researchers created an artificial intelligence tool to identify neutrophils primed for NETosis, a process where white blood cells expel inflammatory DNA into circulation. The new technology allows scientists to measure NETosis in different diseases and test drugs that may inhibit or promote the process.
Scientists at the University of Tsukuba created an AI program called MC-SleepNet to automatically classify mouse sleep stages, achieving 96.6% accuracy and high robustness against noise in biological signals. This system can significantly assist researchers by automating data annotation, accelerating research on sleep patterns.
Researchers employ neural networks to predict molecular bond energies, reducing computational cost and improving accuracy. The combination of AI and quantum chemistry calculations provides an efficient tool for quickly predicting molecular bond energies in complex systems.
Argonne researchers used a machine learning algorithm to relate known molecular structures to larger data sets, reducing computational costs while maintaining precision. The approach improved the accuracy of predictions about battery electrolyte candidates, enabling scientists to identify potential materials for next-generation batteries.
Researchers at MIT developed a model that learns a compact state representation for soft robots, optimizing movement control and material design parameters. This enables 2D and 3D soft robots to complete tasks quickly and accurately in simulations.
Researchers have developed a deep machine learning algorithm that can predict the quantum states of molecules, enabling faster design of drug molecules and new materials. The algorithm can process complex quantum chemical data in seconds on a laptop or mobile phone, revolutionizing computational chemistry and molecular physics.
Researchers studied mouse brain activity while learning tasks, finding neural networks become more focused and selective over time. The team developed computational models to inform decision-making neuroscience, revealing the role of inhibitory neurons in cognition.
Researchers used physics-informed generative adversarial networks (GANs) to model subsurface flow in the Hanford Site, achieving exaflop performance. The approach enabled estimation of hydraulic conductivity and hydraulic head with high accuracy, overcoming the limitations of traditional methods.
A machine learning model identified patients at risk of requiring kidney replacement therapy, leading to a significant improvement in dialysis initiation rates. The system calculates weekly risk scores and alerts clinicians to optimize treatment decisions, resulting in better patient outcomes.
A novel attention-based deep learning method automatically learns clinically important regions on whole-slide images to classify them. The new approach outperformed the current state-of-the-art approach that requires detailed annotations for its training.
An international team of researchers has been awarded a $10 million European Research Council Synergy Grant to develop machine learning algorithms for enhancing Earth observation datasets. They will also develop machine-learning-based parametrizations for clouds and land-surface processes to improve climate modeling.
Researchers develop LOGAN, a deep neural network that can transform shapes between unpaired domains, enabling automatic translation of objects like chairs to tables. The method learns unique features and preserves key characteristics during transformations.
The company has proposed a new family of prior distributions: TRIP, which improves Fréchet Inception Distance for GANs and Evidence Lower Bound for VAEs. The model was experimentally validated in cells and animals, demonstrating its potential for accelerating drug discovery.
Researchers at Princeton University explore adversarial tactics applied to artificial intelligence, which can trick systems into causing gridlock or revealing sensitive information. Machine learning systems are vulnerable to data poisoning and evasion attacks, which can compromise their performance and safety.
A new algorithm combines the capabilities of two spacecraft instruments, enabling lower-cost and higher-efficiency space missions. The virtual super instrument uses deep learning to analyze ultraviolet images and synthesize useful scientific data.
Researchers at Mainz University explore fundamental aspects of artificial intelligence using machine learning techniques and interdisciplinary approaches combining physics, biology, and materials sciences. The goal is to understand why modern systems are successful and develop better machine learning methods.
Researchers developed a machine learning method to enhance optoacoustic imaging quality without sacrificing it. The approach uses sparse data, allowing for reduced sensor numbers and improved diagnosis accuracy, facilitating clinical decision-making.
Researchers are developing a form of cybersecurity inspired by human biological systems, detecting and addressing threats in their earliest stages. The team is also offering training and research opportunities to students from underrepresented backgrounds.
Automated machine learning techniques improve efficiency and accuracy in analyzing heart function on cardiac MRI scans. The study, conducted in the UK, found that AI can analyze a scan in approximately four seconds with similar precision to experts.
A recent systematic review and meta-analysis suggests that artificial intelligence can detect diseases from medical imaging with similar accuracy to health-care professionals. However, the true power of AI remains uncertain due to limited high-quality studies, and researchers call for higher standards of research and reporting.
ORNL's labwide AI Initiative applies machine learning and deep learning to tackle complex problems in materials science, disease diagnosis, and cybersecurity. The lab's powerful computing resources and expertise enable researchers to develop new technologies and extract insights from massive datasets.
A team of researchers, led by Hagit Shatkay, is developing computational methods to accelerate discovery in astroparticle physics, a crucial step towards understanding dark matter. By analyzing noisy sensor data from an underground experiment, the team aims to detect and identify dark-matter particles.
A team of scientists from Skoltech and Moscow Institute of Physics and Technology studied the movements of 19 esports players, including professionals and amateurs. The results show that machine learning methods can accurately predict a player's skill level in 77% of cases, with professional players moving more than beginners.
Researchers use deep neural networks to simulate light-induced molecular reactions on long time scales, accelerating computation by up to 19 years. This method enables better understanding of biological processes like carcinogenesis and ageing, with potential applications in material ageing and photosensitive drugs.
Researchers from Skoltech found that professional players move around more often and intensely than amateurs, while sitting still during game events. Machine learning methods correctly predicted a player's skill level in 77% of cases.
Danish researchers at Aarhus University are developing an AI system to detect market manipulation and fraud in global stock exchanges. The project, called DISPA, aims to replace manual sampling with automated analysis of trading activity.
Researchers use TDA to inject knowledge of real world into neural networks, reducing training time and increasing intelligibility. This approach enables machines to focus on meaningful features and improve performance in tasks like face recognition.
Researchers developed a deep learning model that extracts patterns from gene locations and functions to identify disease associations. The KAUST model achieves better accuracy than state-of-the-art methods by combining multiple datasets and incorporating graph convolutional networks.
Researchers trained an AI model using human-generated clickbait data, resulting in improved performance compared to other systems. The study found differences in headline creation between humans and machines, highlighting the need for high-quality training data to improve machine learning models.
Researchers developed a neural network model using machine learning to predict Universe structure formation. The new model is more accurate than existing analytic methods and efficient enough for large-scale simulations.
Researchers believe that studying animal brains can improve AI's ability to tackle complex tasks like dish-washing. By understanding how biological neural networks work, AI systems may be able to overcome barriers and achieve superhuman performance.
A team of scientists at Bar-Ilan University has developed a new type of ultrafast artificial intelligence algorithm based on the slow dynamics of brain function. This breakthrough outperforms traditional machine learning algorithms in various fields.
A new study from McGill University uses machine learning-guided virtual reality simulators to accurately assess the capabilities of neurosurgeons. The researchers found that these AI-powered tools can predict the level of expertise with 90% accuracy, enabling more efficient and effective mentorship.
Researchers at Cincinnati Children's Hospital Medical Center designed an AI-powered system to streamline clinical trial recruitment. The Automated Clinical Trial Eligibility Screener (ACTES) reduces patient screening time by 34 percent and improves enrollment by 11.1 percent compared to manual screening.
The KDD Cup 2019 competition featured three tracks tackling societal challenges such as transportation and malaria. Teams won prizes of up to $15,000 by applying machine learning tools to complex problems.
Researchers used machine learning to design novel polymers with superior heat transfer properties. The method achieved outstanding prediction performance even with limited data sets, leading to the identification of promising 'virtual' polymers.
A team of UCI researchers developed a deep reinforcement learning algorithm called DeepCubeA, which can solve the Rubik's Cube in under 20 moves, outperforming human solvers. The algorithm works on other combinatorial games and demonstrates symbolic, mathematical, and abstract thinking capabilities.
Researchers developed Deep-CEE, a deep learning technique to speed up finding galaxy clusters. The novel approach uses AI models trained on images to identify galaxy clusters, replacing manual analysis by astronomer George Abell.
A new machine learning approach enables researchers to encode quantum mechanical laws into neural nets, simulating molecular motion billions of times faster than conventional methods. This breakthrough advances research in fields like drug development, protein simulations, and reactive chemistry.
Artificial intelligence is being incorporated into a first-year mass communications class at Lehigh University, equipping students with skills to adapt and shape AI. The new approach aims to prepare students to report on AI's impact on journalism and help shape its future.
A Rutgers University study uses artificial intelligence to control a robotic arm that efficiently packs boxes, saving businesses time and money. The system develops software and algorithms for robust motion and real-time monitoring to detect failures.
A new AI model, D3M, generates complex 3D simulations of the universe in milliseconds, achieving accuracy comparable to high-accuracy models. The breakthrough enables researchers to explore various cosmic scenarios without sacrificing accuracy.
Dartmouth researchers developed an algorithm to measure brain activity patterns and assess conceptual understanding in students. The method produced neural scores that significantly predicted individual differences in performance on concept knowledge tests, highlighting the brain's role in processing complex information.
A study by Tokyo Institute of Technology researchers explores the connection between biological evolutionary open-endedness and recent studies in machine learning. They propose combining neural networks with artificial life ideas to create autonomous systems that invent or discover new things.
Three young scientists, Murad Mamedov, et al., receive $150,000 for their novel methods in understanding the human immune system. The Michelson Prizes recognize groundbreaking research using genomics, AI, and machine learning to transform human health.