Research using machine learning to solve scientific problems. Covers AI quality control, sea-surface temperature estimation, label-free cell analysis, and cyber defense.
Nanoengineers developed new graph network-based models that accurately predict material properties, outperforming existing AI technology in complex tasks. The MEGNet models can learn relationships between elements and overcome data limitations in materials science, enabling rapid discovery of transformative materials.
A machine learning platform called AirSurf-Lettuce uses computer vision and deep learning to categorize lettuce crops in fields, measuring quantity, size, and location. This technology can help reduce yield loss up to 30% by providing precise harvest times and improving crop management decisions.
A new machine learning approach for low-dose CT imaging has been shown to perform as well as, or better than, traditional iterative techniques in an overwhelming majority of cases. The method allows radiologists to fine-tune images according to clinical requirements, enabling faster and more accurate scans.
Researchers investigating the effects of Internet-based learning on university students, finding mixed results in acquiring domain-specific knowledge and difficulties with critical thinking. The study also highlights the role of algorithms in shaping online learning environments.
Scientists from the University of Bristol and ETH Zurich have developed an interactive VR software framework that enables humans to train machine-learning algorithms using 'on-the-fly' quantum mechanics calculations. This allows for high-quality training data generation, improving machine learning models and accelerating scientific dis...
A multi-institution research team, led by Worcester Polytechnic Institute's Eric Young, is developing a biosecurity tool to identify genetically engineered organisms in the environment. The tool uses unique DNA signatures to distinguish between engineered and naturally occurring microorganisms.
A new framework called Learn to Grow has been developed to enable artificial intelligence systems to better learn new tasks while forgetting less of what they have learned regarding previous tasks. This framework allows AI systems to retain previous skills and apply them to new tasks, improving performance and efficiency.
Researchers at Osaka University developed an automatic diagnosis system using deep learning and MEG, achieving high accuracy in classifying patients with neurological diseases. The system outperformed conventional methods using waveforms, offering a promising approach for clinical practice.
Researchers have developed a new framework that enables deep neural networks to learn new tasks while minimizing the loss of previously learned information. The Learn to Grow framework demonstrates improved performance in both new and old tasks, with backward transfer occurring when learning a new task enhances previous task accuracy.
Researchers at the University of Texas at Austin developed an AI agent that can gather visual information and reconstruct a full 360-degree image of its surroundings. The agent uses deep learning to choose the most informative shots, similar to how humans would take pictures in different directions based on prior experience.
A study by the European Society of Cardiology found that machine learning algorithms can accurately predict heart attacks and deaths with over 90% accuracy. By analyzing 85 variables from imaging data, the algorithm identified patterns correlating to death and heart attack, surpassing human performance.
A team of researchers at Virginia Tech has created a new system to efficiently distribute data processing tasks across thousands of servers in supercomputers, achieving balanced loads and improved performance. The novel technique uses machine learning to predict task types and amounts, allowing for optimized load balancing.
Researchers have developed a brain-machine interface that can generate synthetic speech by controlling a virtual vocal tract based on brain activity. The technology has the potential to restore fluent communication in individuals with severe speech disabilities, including those with paralysis and neurological diseases.
The US Department of Energy has announced $20 million in funding for artificial intelligence research, with a focus on improving grid operation and management. The projects aim to develop faster grid analytics, better asset management, and sub-second automatic control actions to reduce costs and avoid grid outages.
A workshop published a roadmap for AI in medical imaging, highlighting key research themes and prioritizing foundational machine learning research. The report emphasizes the need for collaboration among professionals, funding agencies, and institutions to develop innovative imaging technologies.
Dr. Blake Richards has made significant contributions to mathematical models of learning and memory in the brain, providing insights into the neurobiological basis of animal and human intelligence. His work explores the neural basis of deep learning and its potential to revolutionize our understanding of the brain.
Researchers at Virginia Tech use drones and AI to complement human searchers, analyzing historical data from over 50,000 lost person scenarios. The system aims to provide large-scale data for better decision-making, addressing niche problems in the search process.
A machine learning model can reproduce the swarming behavior of locusts by integrating methods from philosophical action theory and quantum optics. The 'Projective Simulation' learning model was successfully applied to a locust's specific swarming behavior, demonstrating its potential for realistic application to biological systems.
Artificial intelligence can amplify human capabilities, reducing systemic glitches and errors in medical decision-making. Machine learning models can analyze vast amounts of data to identify patterns predictive of outcomes and help diagnose diseases.
Researchers developed a system to automatically identify and classify violin bow gestures, providing real-time feedback for students. The system achieved 94% accuracy in identifying bowing techniques, enabling practical learning scenarios for music education.
A team of healthcare data scientists and doctors developed an AI system using machine learning algorithms to predict the risk of early death due to chronic disease. The system performed better than current standard approaches and showed promise in improving preventative healthcare.
Researchers found that children and teens learn about 2 bits per minute to acquire linguistic knowledge, filling a 1.5 MB floppy disk. This challenges assumptions that human language acquisition happens effortlessly.
Researchers used machine learning techniques on MRI brain scans to predict cognitive development in children aged 2. The study found that white matter connections at birth are highly predictive of future cognitive outcomes.
AI systems often employ 'Clever Hans' strategies that are not meaningful from a human perspective, but still achieve success. Researchers have developed explainable AI technology to identify these flawed strategies and identify more intelligent problem-solving approaches.
A new study from Carnegie Mellon University reveals that learning scientific information results in changes in the actual structure of memory-related areas of the brain. The researchers found evidence of microstructural, informational, and network change in the left hippocampus during the learning of organic compounds.
A recent NSF CAREER Award recipient aims to develop trustworthy machine learning from untrusted models, with a focus on safeguarding complex systems that piggyback on less-than-trustworthy technology. The grant will support work in creating tools to verify model safety and detect abnormal phenomena throughout the system's lifecycle.
Researchers develop AI algorithm to optimize traffic management in optical telecommunications networks, increasing efficiency by 30%. The new approach uses deep reinforcement learning and can learn autonomously without prior knowledge.
A study by German scientists uses AI to enhance climate and Earth system models, improving predictions for extreme events and seasonal changes. By combining physical modeling with machine learning techniques, researchers aim to create more accurate models that capture complex dynamic processes.
The Rhine-Main Universities Initiative Funding for Research supports the DeCoDeML network, combining expertise in deep continuous-discrete machine learning to tackle unresolved issues. Researchers from Mainz, Darmstadt, and Frankfurt will examine how machine learning can be made comprehensible to human understanding.
Researchers found that machine learning can identify customer needs from user-generated content (UGC) more accurately and efficiently than traditional methods. The approach uses machine learning to analyze UGC data and remove redundancies, followed by human analysis to formulate customer needs.
Game behavior can be analyzed to predict a person's personality features, including gender. Researchers used machine learning on large amounts of game data from the Steam gaming platform to make accurate predictions. The study shows that even limited information about gameplay and achievements can provide good predictive values.
A new machine learning methodology was developed by researchers at NCCR MARVEL to capture chemical intuition from partially failed trials. This approach helps chemists improve their synthesis conditions and create novel materials with higher surface areas.
Columbia engineers create a robot that learns what it is from scratch with zero prior knowledge of physics or motor dynamics. The robot uses deep learning to create a self-model, allowing it to adapt and learn from its own experiences.
The MIT robot uses a soft-pronged gripper, force-sensing wrist cuff, and external camera to see and feel the tower and its individual blocks. It learns from visual and tactile feedback, and adapts its moves in real-time to avoid toppling the tower.
A study published in PLOS ONE suggests that machine learning can provide an equally accurate and reliable prognosis for patients with cardiovascular disease, compared to traditional methods. The use of genetic programming reduces bias and human error, allowing complex associations to be made transparent to clinicians.
Researchers challenged a popular idea about how machine learning algorithms think by applying information theory to classification problems. They found that classifiers with many layers do not necessarily trade off between prediction and compression as previously thought.
A recent study from the University of Waterloo found that measuring AI's ability to learn is challenging due to the complexity of tasks. The researchers discovered that no mathematical method can determine whether an AI-based tool can handle a task or not, even with precise task descriptions.
A new study combines two forecasting methods with machine learning to estimate local flu activity, producing the most accurate estimates available, a week ahead of traditional healthcare reports. The approach, called ARGONet, outperforms earlier methods in more than 75% of states studied.
Researchers used artificial neural networks to learn atomic interactions from quantum mechanics, bypassing complex calculations. The ANN was used as a surrogate model to account for errors and improve predictions.
Researchers at Argonne National Laboratory are using machine learning algorithms to optimize engine simulations, significantly reducing design time and increasing accuracy. The project aims to create a more efficient and emissions-free combustion process, with potential applications in the automotive industry.
A research team led by Heng Huang aims to create a framework for secure and efficient multi-site collaborative big brain data mining. The project addresses computational challenges in analyzing complex brain disorders and genomics data.
Jakoah Brgoch aims to improve energy efficiency in LED lighting by developing new materials and algorithms. He will use machine learning to identify new phosphors and predict material behavior under temperature changes.
Researchers at PNNL developed a deep learning model that surpasses current cloud detection algorithms by nearly doubling precision and reducing processing time. The model's performance is promising for global forecasting applications.
Researchers developed a novel classification method combining brain imaging data from MRI and fMRI to recognize patients with benign epilepsy with centrotemporal spikes. This approach improved diagnosis accuracy, enabling early detection and treatment, which leads to better health outcomes for children. The study used a dataset of 40 B...
Japanese researchers have developed a technique for automatic anime colorization using deep learning, which is based on recent advances in machine learning approaches. The technique uses a combination of image segmentation and voting techniques to refine the colorization result.
Researchers developed a new machine learning method that allows AI to make classifications without negative data, a crucial component in traditional classification technology. This breakthrough enables AI systems to function effectively even when limited by data regulation or business constraints.
A new study from The George Institute for Global Health at the University of Oxford found that machine learning models can accurately predict the risk of emergency hospital admissions. By analyzing electronic health records, these models identified factors such as age, sex, and socioeconomic status to provide a more robust prediction t...
A team of computer scientists developed a novel deep learning-based approach to automatically generate realistic caricatures from portraits. Their method, called CariGANs, successfully generated clearer and more accurate caricature depictions that resemble hand-drawn artwork by professionals.
AI systems have a 'near unlimited capacity' for data processing and learning, potentially making diagnoses and surgical interventions more accurate. However, doctors will remain essential for holistic patient care, addressing individual preferences and social circumstances.
Researchers have discovered three types of attacks that can be launched using a computer's graphics processing unit (GPU) to spy on web activity, steal passwords, and break into cloud-based applications. The attacks work by exploiting vulnerabilities in the GPU's memory utilization and performance counters.
Researchers developed a new approach to interpret machine learning algorithms, revealing how they can produce nonsensical answers even when given meaningful inputs. By reducing inputs to the bare minimum required for correct answers, the method provides insights into algorithmic limitations and potential solutions.
Researchers at NASA's Goddard Space Flight Center are developing machine learning algorithms to analyze petabytes of data, extract new patterns and correlations, and improve science discoveries. The technology also helps monitor spacecraft health, prevent catastrophic failures, and assist engineers in designing new missions.
Researchers from Florida Atlantic University developed a machine learning model that detects Medicare fraud with high accuracy. By analyzing large datasets from Medicare Part B, they identified a unique process to map fraudulent providers, achieving a statistically significant detection rate.
A neural network mimics the fruit fly's visual system and distinguishes between individuals based on sight alone. The system achieved an F1 score of 0.75, surpassing human biologists' performance in a similar task.
Researchers will work together to develop tools and methods to manage machine learning risks, including methods to defend against adversarial inputs and develop robust training methods. The center aims to foster a long-term research community in trustworthy machine learning.
A proposed taxonomy categorizes conscience protections based on state or private actors, as well as provider conduct. Bioethicists must shape AI and machine learning to prevent moral pitfalls, including exacerbating human biases and weakening patient confidentiality.
Researchers are using real-time learning to create a unified theory on how to optimize microgrid capacity through DC distribution. The team aims to create optimal power profiles for power electronic converters in DC distribution networks.
The University of Pittsburgh has received a $1.2 million NSF grant to analyze electronic anesthesia records and prevent postoperative complications and death using machine learning and Big Data analysis. Dr. Heng Huang will develop a new deep learning algorithm to predict surgical outcomes based on historical patient data.
The special issue brings together contributions from top academics and industry experts, highlighting the use of generative chemistry and GANs for de-novo molecular design. Experimental validation of generated molecules demonstrates high activity and selectivity in a novel inhibitor of Janus Kinase 3.
A novel machine learning framework developed by Virginia Tech researchers accelerates the discovery of new materials through computer simulations. The framework, which trains on the fly, enables faster development of accurate computational models of materials with potential biomedicine and energy applications.