Articles tagged with Neural Networks
Researchers created an abstract language that describes protein molecules' shapes and structures, enabling predictions of their dynamics. This method uses machine learning algorithms to analyze molecular movements and provides insights into disease causes and targeted drug therapies.
Researchers are developing a novel deep learning technique to identify relationships between brain networks and Alzheimer's disease using algorithms mimicking neural networks. The goal is to pinpoint specific areas in the brain to slow and treat disease progression.
Scientists have created a new nanodevice that acts like a brain cell and can be joined to form networks that solve problems in a brain-like manner. These systems can identify possible mutations in a virus, relevant for ensuring vaccine efficacy.
A new data processing module called attentive normalization improves the performance of deep neural networks by combining feature normalization and feature attention. The hybrid module significantly increases accuracy while using negligible extra computational power, and facilitates better transfer learning between different domains.
A team of engineers and computer scientists are developing a theory of deep learning based on rigorous mathematical principles to improve reliability and predictability in AI systems. They will use three perspectives: local to global understanding, statistical analysis, and formal verification.
Researchers have found a way to improve the accuracy of brain-inspired computing systems using memristors, which are at least 1,000 times more energy-efficient than conventional transistor-based AI hardware. This could lead to a significant reduction in carbon emissions from training one AI model.
Researchers developed a neural network to assess stored blood quality, achieving 76.7% agreement with experts in identifying damaged red blood cells. The network outperformed expert predictions when trained using only storage duration.
Researchers identified a new role for bi-directional connections in accelerating communication between brain regions. By creating loops, these connections can establish resonance and amplify signals, reducing the need for synchronization and increasing network efficiency.
Researchers create novel method using artificial intelligence to connect static and dynamic calculations, enhancing system security and safety requirements. The approach enables operators to anticipate disruptions and optimize resource allocation for a more resilient power grid.
Researchers found CAMSAP1 plays a crucial role in regulating axon/dendrite differentiation by creating an unbalanced distribution of microtubules among processes. The study resolves a long-standing question in neuroscience about the decisive factor for neuronal polarity establishment.
Researchers are using multidisciplinary approaches, cutting-edge imaging technologies, and cyber resources to study synaptic weight and its effects on the brain. The team aims to determine what factors shape synaptic structures and function, shedding light on basic understanding of the brain.
Researchers at Tohoku University and the University of Gothenburg developed a novel voltage-controlled spintronic oscillator capable of closely imitating non-linear oscillatory neural networks. The technology allows for strong tuning with negligible energy consumption, enabling efficient training of large neural networks.
Scientists from Nanyang Technological University (NTU Singapore) have developed an AI system that recognizes human hand gestures with high precision. The system combines skin-like electronics with computer vision and achieves accuracy even in poor environmental conditions.
A new study by MIT neuroscientists provides a mathematical model showing how the brain overcomes unpredictable disturbances to produce reliable computations. The model describes an inclination toward robust stability built into neural circuits due to connections between neurons.
Researchers at the University of Illinois trained light-sensitive neurons using timed pulses of light during early cell development, leading to improved connections, responsivity, and gene expression. The early training resulted in long-lasting improvements, whereas cells trained later had transient responses.
In a breakthrough study, scientists successfully implanted highly specialized neural stem cell grafts directly into mouse spinal cord injuries, showing they integrated with host networks and behaved like neurons. The grafts displayed spontaneous activity, responded to sensory stimuli, and formed functional connections with host neurons.
A researcher at Sandia National Laboratories has won an Early Career Research Program award to develop methods for applying physics laws to observe large-scale physical events. The project aims to achieve a millionfold change in scale, from meter- to microscale features.
Researchers at Washington University in St. Louis developed a new algorithm called Parallel Residual Projection (PRP) to solve linear inverse problems by breaking them down into smaller tasks that can be solved in parallel on standard computers.
A team led by University of Pittsburgh's Jingtong Hu is working on a project to train algorithms that can accurately diagnose pneumonia caused by COVID-19 using CT scans. The goal is to create a mobile scanning device that can quickly screen for signs of the disease in crowded places.
A study by Heidelberg University and Max-Planck-Institute found that the distance to criticality can be adjusted in a brain-inspired chip, but only complex tasks benefit from it. Optimal network dynamics can be tuned using homeostatic plasticity by adapting mean input strength.
A groundbreaking study using MRI scans of 130 mammalian brains found that brain connectivity levels are equal in all species, including humans. The research revealed a universal law: Conservation of Brain Connectivity, which suggests that the efficiency of information transfer in the brain's neural network is the same across mammals.
Researchers at Graz University of Technology developed a new machine learning algorithm called e-prop, which significantly expands the possible applications of AI. This novel approach uses spikes to enable more efficient information processing and reduces energy consumption.
Zhao and Cheng are working on a project to develop new gradient-free methods for training various types of deep neural networks. They aim to create an algorithmic and theoretical framework for model parallelization based on gradient-free optimization, as well as efficient distributed workflow systems.
Researchers at Medical University of South Carolina found that the brain uses similar visual areas for mental imagery and vision, but with less precision. This study has potential applications for understanding PTSD and other mental health disorders affecting mental imagery.
Researchers analyzed hundreds of thousands of Instagram posts about vaping and found that 40% were promoting flavored e-liquids to young audiences. The study highlights the need for stricter laws and regulations on social media advertising targeting younger users.
Researchers from North Carolina State University discovered that incorporating Hamiltonian function into neural networks enables them to better predict and respond to chaos. This innovation has significant implications for improved artificial intelligence applications.
Researchers from Lobachevsky University and international colleagues have developed a model that demonstrates the existence of concept cells in the brain, which can process and learn abstract concepts. The study suggests that individual neurons, rather than large neuronal complexes, are responsible for complex tasks performed by humans.
Feng Xiong is developing a two-dimensional synaptic array to enable computers to process vast datasets with less power and greater speed. This technology aims to mimic the brain's efficient learning process, allowing for more precise adjustments between states.
Using simulated silicon neurons, researchers found that energy constraints can lead to a dynamic, at-a-distance communication protocol more robust and energy-efficient than traditional computer processors. This protocol enables computing on a secondary network of spikes, allowing for efficient communication and processing.
A team of scientists proposes a memristive neurohybrid chip to create compact biosensors and neuroprostheses with high adaptability. The system combines neural cellular and microfluidic technologies for real-time registration, processing, and stimulation of bioelectrical activity.
An international team developed artificial neurons that can precisely target specific brain cells using optogenetics and light patterns. This technology has the potential to replace damaged brain circuits and restore communication between brain regions.
Researchers developed Early Bird, an energy-efficient method for training deep neural networks, which can use 10.7 times less energy than traditional methods to achieve the same level of accuracy. This breakthrough could lead to significant cost savings and a reduction in greenhouse gas emissions.
Research reveals that early visual experience drives precise alignment of cortical networks to unite inputs from both eyes, enabling unified binocular representation. This process occurs within the first week after eye opening and refines neuron response properties.
A new standard for machine learning in telecommunications networks has been approved, enabling faster data transmission rates of up to 20 Gbps and reducing latency to less than 5ms. This breakthrough is made possible by the application of deep learning techniques, allowing for complex pattern recognition and network load management.
New research reveals that certain internal clock neurons in fruit flies, previously thought to send time-keeping cues to the brain, actually receive cues from the external environment. This finding has significant implications for understanding circadian rhythm disruptions and their associated health problems.
Researchers at MIT developed a new automated AI system that reduces the energy required for training and running neural networks. The system, called a 'once-for-all' network, trains one large neural network comprising many pretrained subnetworks, reducing carbon emissions by low triple digits.
Researchers found that rapid-acting antidepressants share the ability to regulate both synaptic potentiation and reciprocal homeostatic mechanisms, which weaken synaptic strength during sleep. This suggests that slow-wave responses could be a useful measure for determining treatment efficacy and developing novel treatments.
Researchers propose a new framework to quantify the predictability of temporal networks, which encodes the ordering and causality of interactions between nodes. The study found that the contributions of topology and temporality to network predictability vary significantly across different types of real networks.
Researchers at the University of Texas at Austin developed a method to make big data processing more energy efficient using magnetic components. By leveraging lateral inhibition in artificial neurons, they achieved an energy reduction of 20-30 times compared to standard back-propagation algorithms.
Researchers at MIT used machine learning to streamline the discovery process for new materials, narrowing down 3 million candidates to eight promising options in just five weeks. The neural network was able to predict properties and optimize criteria, improving upon conventional analytical methods.
Research in mice reveals that lactation temporarily changes how a mother's TIDA neurons regulate prolactin secretion, causing them to fire more frequently and out of rhythm. However, these changes are fully reversible after weaning, suggesting a unique adaptation to motherhood.
The Learning to Synthesize (LS-DNN) approach splits input signals into low and high spatial frequency bands, enabling deep neural networks to process and synthesize them. The algorithm is robust in handling noisy intensity signals, making it suitable for applications like x-rays and sonograms.
Researchers developed a computer program to identify each nerve cell in fluorescent microscope images of living worms, overcoming previous challenges by creating unique genetic modifications. The program uses a mathematical algorithm to analyze images and assign neuron identities based on position variations between individual animals.
Researchers developed a platform to coculture neurons and muscle cells, capturing the emergence of neuromuscular junctions and synchronized bursting patterns. The study provides new insights into biohybrid machines and their potential applications in fields like intelligent drug delivery and environment sensing.
A team from Deakin University in Australia developed an improved sight-correcting system for self-driving vehicles. By watching human operators complete tasks, the vehicles can learn to make decisions based on visual information, reducing the need for extensive training data.
A new chip has been developed at TU Wien that can recognize certain objects within nanoseconds, leveraging artificial intelligence and a special material. The chip integrates the neural network with its AI directly into the image sensor, making object recognition faster by many orders of magnitude.
Researchers found that brain cells generate a 'new song' with the same beat for each breath, adapting to changing rhythms throughout the day. The discovery could lead to new approaches to treating breathing disorders and may even help combat opioid-related deaths.
Researchers successfully separate and observe single-molecule magnets (SMMs) on a magnetically neutral silica substrate using transmission electron microscopy. This breakthrough enables the development of auto-associative memories and multi-criterion optimization systems, mirroring the human brain.
Researchers at MIT have developed a machine learning method to fill in the missing low-frequency seismic waves in human-generated seismic data, allowing for more accurate mapping of underground structures. The technique was trained on simulated earthquakes and used to infer missing frequencies from new input data.
A multidisciplinary study led by UB researchers has developed a new experimental tool to study how neuronal networks recover their function after neuron loss. The study shows that the network quickly activates self-regulation mechanisms that reinforce existing connections and restore circuit functionality.
Researchers at Mayo Clinic have created an artificial intelligence (AI) algorithm that can detect unseen characteristics of hypertrophic cardiomyopathy using standard EKGs. The AI's ability to diagnose the disease was found to be highly accurate, with an area under the curve of 0.96, outperforming traditional tests.
Researchers created a digital amplifier model using a deep neural network that can accurately simulate the sound of various guitar amplifiers, including popular brands like Marshall and Orange. The study uses black-box modelling to replicate the observed input-output mapping of analogue circuitry.
Researchers created a neural network that autonomously finds solutions well-adapted to quantum advantage demonstrations, aiding in developing new efficient quantum computers. This breakthrough enables the prediction of quantum advantages in complex networks, which is crucial for creating cost-effective and reliable quantum devices.
Scientists have identified recurring patterns in brain neurons that can be used to explain their behavior and function, paving the way for creating artificial intelligence that mimics the human brain. By understanding these patterns, researchers aim to develop new treatments for neurological disorders and improve current technology.
A deep learning method using a convolutional neural network (CNN) accurately differentiates between malignant and benign solid masses in small renal masses on contrast-enhanced CT scans. The corticomedullary phase showed the highest AUC value, indicating its effectiveness in malignancy prediction.
Researchers develop 'flash and freeze' method to study structure and function of synapses in intact neural circuits. The method allows for simultaneous observation of structural changes during signaling, revealing a near-identity between structurally and functionally defined vesicle pools.
Researchers developed a system to accurately detect space debris in Earth's orbit using laser ranging telescopes and neural networks. The new algorithm significantly improves the success rate of space debris detection, allowing for safer spacecraft maneuvers.
Scientists at Tokyo Institute of Technology found that overly strong connections can invert the effect of connectivity on complex activity, leading to more regular patterns. This phenomenon is observed in various natural and engineered systems, including neurons, coupled oscillators, and wireless terminals.
Researchers are testing whether changing brain's fuel source from glucose to ketones could potentially save neurons and neural networks over time. The study, funded by a $2.5 million grant, aims to understand how ketones affect brain cells and connectivity in the face of insulin resistance.
Researchers developed an innovative method to measure the complexity of image representations in deep neural networks, shedding light on their processing stages. The study found that classification accuracy depends on the network's ability to simplify information, with more accurate results from simplified representations.