Articles tagged with Neural Networks
Researchers have developed a method using nanomagnets to perform artificial intelligence, slashing energy costs and offering huge efficiency gains. The technology uses 'nanomagnetic states' to process and store data, cutting out the need for software simulation.
Researchers at Caltech developed Neural-Fly, a deep-learning method that enables drones to adapt to wind conditions in real-time. The method achieved significant improvements in drone performance compared to existing adaptive control algorithms.
Researchers mapped neural networks in marmoset and macaque monkeys to find that multiple areas in the frontal lobe control vocalization, contradicting a long-held view. This discovery could lead to a better understanding of speech disorders such as stuttering and apraxia.
A new case study reveals significant differences between human and AI perception in breast-cancer screenings. Researchers found that AI systems consider tiny details in mammograms that are irrelevant to radiologists, highlighting the need for understanding and correcting AI decision-making before trusting it for life-critical medical d...
Researchers developed a machine-learning method that allows robots to pick up and place never-before-seen objects in random poses, requiring only 10 human demonstrations. The system uses a neural network specifically designed to reconstruct 3D shapes, enabling the robot to generalize to new object orientations.
Researchers at Salk Institute discover that brain parses information through interactions of waves of neural activity, changing how data is processed and affecting attention and focus.
A new deep learning method, PDD-Net, uses 3D deep convolutional neural networks to extract deep metabolic imaging indices from PET scans for differential diagnosis of parkinsonian diseases. The method achieved high sensitivity and specificity rates for Parkinson's disease and other parkinsonian syndromes.
Researchers developed an AI algorithm to model first impressions and accurately predict how people will be perceived based on a photograph of their face. The algorithm's findings align with common intuitions or cultural assumptions, such as people who smile being seen as more trustworthy.
A study at the University of Helsinki found that in utero exposure to mother's antiepileptic or antidepressant medication can lead to widespread changes in cortical networks, affecting local and global brain function. This may have implications for infants' neuropsychological development and future research on environmental factors.
A new control allocation method using a neural network improves the performance of quadrotor controllers by considering aerodynamic effects. This approach reduces errors in command generation and delivers better thrust and torque signals.
A new study found that microglia regulate neuronal subtypes differently in response to bacteria, affecting intrinsic excitability. Pyramidal cells exhibited lower excitability, while Purkinje cells showed higher excitability when modulated by microglia.
A study by Tokyo University of Science researchers has demonstrated that a computationally-light model can simulate complex brain cell responses, including periodic and quasi-periodic responses. The Izhikevich neuron model was found to be capable of reproducing both types of responses at lower computational cost.
A project aims to develop software toolkits to assess neural network robustness and potential security vulnerabilities. The goal is to create a framework for building secure AI systems, emphasizing human expertise in data collection and testing.
The UNC Charlotte team developed a universal AI algorithm called AutoClass to clean noisy single-cell RNA sequencing (scRNA-Seq) data. The algorithm effectively removes noise and enhances downstream analysis in multiple aspects, demonstrating its robustness and scalability.
Researchers have identified the complete series of 10 factors that regulate the development of brain cell types in the visual system of fruit flies. This discovery opens new avenues of research to understand how brain development evolved in different animals and holds clues for regenerative medicine.
A novel 'rational' neural network reveals underlying mathematical equations through Green's functions, enabling humans to understand machine-generated findings. This breakthrough in partial differential equation learning holds promise for advancing scientific exploration of weather systems, climate change, and more.
A new study suggests that a high-salt diet can lead to the hyperactivity of brain cells, resulting in increased constriction of blood vessels and worsening of cardiometabolic diseases. The research also found that excessive salt consumption can trigger an unusual response in which neurons become more active despite reduced blood flow.
Researchers from the University of Cambridge and Oslo identify a century-old mathematical paradox as the Achilles' heel of modern AI. The paradox limits the existence of stable and accurate neural networks, making many AI systems untrustworthy in high-risk areas.
Researchers reexamined hundreds of experiments on neural activity and consciousness, finding that experiment parameters determine results. The study used artificial intelligence to predict which theory would be supported by each experiment with 80% success.
Researchers at Duke University found a collection of coordinated brain regions that predict and direct social behavior in mice. By analyzing the electrical activity of these regions, they identified how social or solitary an individual mouse is and were able to prompt them to be more gregarious. This study may lead to better diagnostic...
A study published in Frontiers in Microbiology has found that machine learning analysis of microscopy images can be used to identify bacteria resistant to antibiotics. Researchers discovered that shape changes in bacterial cells can predict drug resistance, suggesting a new approach for detecting and predicting drug resistance.
Researchers at Kaunas University of Technology improved an algorithm to detect Alzheimer's disease from MRI images, achieving over 98% accuracy. The new model uses a modified neural network and adapts to variations in data, such as differences in hospital equipment and patient positions.
A study found that intact astrocyte networks are essential for neural homeostasis, synaptic plasticity, and spatial cognitive abilities in adult mice. Disrupting these networks impairs spatial learning and memory due to altered neuronal excitability and compromised synaptic transmission.
A new MRI probe can monitor individual populations of neurons and reveal how they interact with each other. The technique uses genetically targeted probes to detect neural activity and provide a more precise picture of brain function.
A team of scientists developed a soft haptic sensor that can accurately estimate contact points and forces using computer vision and deep neural networks. The sensor is sensitive enough to detect even tiny forces and detailed object shapes.
A deep learning model has been developed to infer horizontal motion on the Sun's surface using temperature and vertical motion data. The technique shows promise for future high-resolution solar observations and laboratory plasmas.
KAUST researchers develop an artificial electronic retina that mimics human vision and recognizes handwritten numbers with high accuracy. The retina uses perovskite nanocrystals to detect light intensity via capacitive change, offering a more energy-efficient alternative to existing systems.
Research reveals that the brain's electrical rhythms fluctuate between high precision and low precision states several times per second, affecting how relevant information is transmitted. Cross-frequency coupling enables selective attention by modulating the strength of different frequencies, while distinguishing between different type...
The University of Essex team has devised a new approach to training neural networks called Target Space, which stabilizes the learning process by tweaking neuron firing strengths. This method enables deeper neural networks with fewer training examples and computing resources, accelerating AI breakthroughs.
Researchers studied how diverse neural network training datasets impact generalization. They found that data diversity is key to overcoming bias, but also degrade performance when neural networks are trained for multiple tasks simultaneously. The study highlights the importance of designing diverse and controlled datasets in machine le...
Researchers at Tokyo Institute of Technology have developed a new AI processor called Hiddenite, which achieves state-of-the-art accuracy in sparse neural networks with lower computational burdens. The chip drastically reduces external memory access for enhanced computational efficiency.
A team of researchers from Skoltech and universities developed a neural network-based solution for automated recognition of chemical formulas on research paper scans. The algorithm combines molecules, functional groups, fonts, styles, and printing defects to mimic existing molecular template depiction styles.
A new machine learning process has been developed to identify and classify hip fractures from X-rays, achieving 92% accuracy and confidence. This approach aims to improve patient outcomes, reduce care costs, and alleviate the bottleneck of unreported radiographs in the UK.
Neuroscientists have designed neural organoids with both mature neurons and astrocytic glial cells to study interactions between brain cells. The new technology enables the emulation of brain activity during healthy and disease states, opening doors to rapid drug screening for neurological diseases.
Researchers propose a new neural network-based method to visualize chemical reactions in a 2D plane, grouping similar reactions together. The visualization helps chemists understand the global chemical reaction space and identify underused or unused reaction types.
Researchers discovered increased activity in the hippocampus during anesthesia and sleep, preceding Alzheimer's symptoms by years. This abnormality may enable early diagnosis and treatment of the disease.
Researchers at KTH Royal Institute of Technology and Stanford University have developed a material that enables the commercial viability of neuromorphic computers mimicking the human brain. The material, MXene, combines high speed, temperature stability, and integration compatibility in a single device.
The MIT team developed a computer model that can perform sound localization tasks as well as humans, and adapts to real-world environments. The model uses convolutional neural networks and was trained on over 400 sounds, including human voices and animal sounds.
Cornell researchers have successfully trained various physical systems, including mechanical, optical, and electrical systems, to perform machine learning tasks. The developed training algorithm enables diverse systems to be chained together for efficient processing.
Scientists at Vienna University of Technology have developed a new type of neural network that can accurately simulate the quark-gluon plasma, a state of matter present in the early universe. The networks use gauge invariant convolutional neural networks to recognize patterns and predict properties of the plasma.
Researchers found that REM sleep has specific properties allowing rapid arousal in response to predatory stimuli. The medial subthalamic nucleus (mSTN) plays a key role in this process, producing a lowered arousal threshold during REM sleep for detecting predator threats.
MIT researchers develop a method to test feature-attribution methods for machine-learning models. They find that even the most popular methods often miss important features in an image and some perform as poorly as a random baseline. This has major implications for high-stakes situations like medical diagnoses.
A team of scientists from Gwangju Institute of Science and Technology developed a deep learning-based approach to predict SC2 battle outcomes by considering army composition and terrain type. The proposed model leveraged parameter sharing, enabling it to analyze complex factors accurately and make predictions.
A new project aims to improve the performance of Graph Neural Networks (GNNs) by leveraging weak supervision and additional information. The research has potential applications in fraud detection, agriculture, and cancer diagnosis.
Researchers at RIKEN CBS demonstrate that neural networks minimize energy cost and solve mazes efficiently, pointing to a set of universal mathematical rules. The findings will aid in analyzing impaired brain function and generating optimized neural networks for artificial intelligences.
A two-year study aims to restore functional vision among veterans and service members with head trauma-related visual dysfunction using immersive virtual reality technology. Researchers will explore the feasibility and benefits of applying new technologies to rehabilitative treatment activities.
A team of researchers used neural network data to study grid cell activity in the brain, finding that collective neural activity is shaped like a torus, or doughnut. The study provides new insights into how large networks of neurons produce properties that cannot be inferred from individual cells.
Researchers have discovered a new biomarker, microstate D, associated with increased sleep disturbance and inattention symptoms in adults with ADHD. The study, published in Biological Psychiatry: Cognitive Neuroscience and Neuroimaging, provides evidence for a more precise diagnosis of the disorder.
A transdisciplinary research team at Göttingen Campus has found a new perspective on the rhythmic processes in the brain. They discovered that adapting interneurons can switch between very slow rhythms and fast rhythms, challenging previous assumptions about their function.
A Salk Institute team has uncovered a neural network in the brain that connects breathing rhythm with feelings of pain and fear. This discovery could lead to the development of an analgesic that prevents opioid-induced respiratory depression, a major cause of overdose deaths.
A team of researchers from the University of Groningen developed an AI-based system that can identify individual Holstein cows in a milking station based on their coat pattern. The system achieved a recognition rate of 99.7% and has several advantages, including non-invasiveness, cost-effectiveness, and scalability.
Researchers at the University of Göttingen studied how blocking certain enzymes affects brain adaptability in healthy and diseased mice. In healthy mice, inhibiting these enzymes blocked neuronal plasticity, while in stroke-affected mice, it restored lost plasticity.
Washington University researchers have designed a new processing-in-memory (PIM) circuit that can increase PIM computing's performance by orders of magnitude. The circuit uses resistive random-access memory PIM, allowing for analog computations and eliminating the need for digital conversions.
Researchers aim to develop AI agents that reuse information, adapt quickly to new conditions and collaborate by sharing experiences. The goal is to enable machines to continually learn from their collective experiences and improve performance on novel and previous tasks.
A Michigan Tech-developed machine learning model uses probability to classify breast cancer shown in histopathology images and evaluate the uncertainty of its predictions. The model outperforms similar models and can measure uncertainty, promising time savings and referrals to human experts.
A new strategy for ultrafast and energy-efficient all-optical computing is proposed based on convolutional neural networks. The approach uses cascaded silicon waveguides to control light amplitude and phase, achieving ultrafast computing times of several picoseconds with low energy consumption.
Researchers at the University of Gothenburg have successfully combined a memory function with a calculation function in the same component, enabling more efficient technologies like mobile phones and self-driving cars. The discovery opens the way for brain-like computers that can perform tasks effectively and energy efficiently.
Researchers at Tohoku University and the University of Gothenburg have developed a new spintronic technology that integrates a memristor-controlled oscillator array, allowing for efficient brain-inspired computing. This breakthrough enables sophisticated cognitive tasks like image recognition with reduced energy consumption.
A KAUST team developed an improved method for detecting malicious intrusions using deep learning, achieving accuracy rates of up to 99% in simulations of different kinds of attacks. This stacked deep learning approach promises an effective defense against cyberattacks and could prevent outages in critical infrastructure.
Researchers discovered the retrosplenial cortex as the site of value decision-making in the brain. Persistency allows value signals to be effectively represented across different brain areas, especially the RSC. Artificial intelligence networks mimicking mouse decisions showed remarkably similar results.