Articles tagged with Neural Networks
Researchers used deep learning to analyze patterns of taxi demand and predict demand significantly better than current technology. This approach could help lessen idle time for taxis, making cities cleaner and improving safety in congested areas.
Researchers recorded electrical activity of hundreds of neurons in a mouse model for up to half an hour, discovering competing neural networks that operate at different timescales. The findings show that certain networks can synchronize their activity, while others slow down or speed up in a coordinated manner.
A novel 'memtransistor' device developed by Northwestern University's Mark C. Hersam can process information and store memory like the human brain, potentially revolutionizing computing. The memtransistor combines characteristics of a memristor and transistor, operating with multiple terminals similar to neural networks.
Lobachevsky University scientists discovered that GDNF protects cultures from cell death and maintains network activity during hypoxia. The neurotrophic factor partially negates the consequences of hypoxia by influencing synaptic plasticity.
MIT researchers developed a special-purpose chip that increases the speed of neural-network computations while reducing power consumption. The chip can calculate dot products for multiple nodes in a single step, improving efficiency and making neural networks more practical for handheld devices.
Researchers created a computer model using neural networks to predict areas prone to corruption, finding that longer government terms and certain economic variables increase the likelihood. The study aims to contribute to anti-corruption efforts by targeting high-risk regions.
Researchers studying nervous system adaptation to ischemic damage hope to develop effective therapeutic strategies by understanding how neural networks function under stress. They have developed methods for modeling different phases of ischemia and studied the features of neural network operation under such effects.
Researchers developed codes MENNDL and RAVENNA to efficiently design and train neural networks, generating and training up to 18,600 networks simultaneously. This enables the training of highly accurate networks in a fraction of the time, with applications in self-driving cars, intelligent robots, and scientific experiments.
A new noninvasive approach to treat tinnitus has shown promising results in a double-blind study, alleviating symptoms in 20% of participants. The therapy involves alternating audio and somatosensory stimulation, delivered through headphones and mild pulses on the neck or cheek.
Researchers from HSE and University of Valladolid created a neural network prediction model to detect corruption cases in Spanish regions. The model uses macroeconomic and political determinants to estimate the probability of corrupt cases emerging over three years, providing valuable insights for anti-corruption measures.
Researchers at the University of Michigan have created a new type of neural network made with memristors that can dramatically improve the efficiency of teaching machines to think like humans. The system, called reservoir computing, uses fewer nodes and requires less training time than traditional neural networks.
A new study shows that very low levels of electrical stimulation can instruct an appropriate response or action in the brain, bypassing damaged senses. The findings have significant implications for the development of neuro-prosthetics and brain-computer interfaces.
A study published in eLife reveals that certain mammalian neurons have shapes and electrical properties well-suited for deep learning. The algorithm simulates how these neurons collaborate to achieve deep learning, offering a more biologically realistic approach.
Researchers have developed a new writing method to create any desired magnetic pattern on nanowires, mimicking brain information processing. This breakthrough could lead to the creation of hardware neural networks, which may surpass software-based approaches in efficiency.
Researchers at Lobachevsky University are developing a neural network prototype based on memristors that can analyze and classify living cell culture dynamics. The project aims to create compact electronic devices that function as part of bio-like neural networks in conjunction with living biological cultures.
Researchers efficiently used Stampede2's 1024 Skylake processors to complete a 100-epoch ImageNet training with AlexNet in 11 minutes, setting the fastest time recorded to date. The Layer-Wise Adaptive Rate Scaling (LARS) algorithm enabled this breakthrough, allowing for larger-than-ever batch sizes and adaptive learning rate adjustments.
Researchers at Osaka University designed a novel gait recognition method that can overcome intra-subject variations by view differences. The proposed architectures outperformed state-of-the-art benchmarks in accordance with their suitable situations of verification/identification tasks and view differences.
Researchers at Northeastern University's Center for Complex Network Research have identified fundamental rules for how the brain controls movement in nematode worms. The study provides unprecedented detail on how individual neurons control specific types of locomotion, paving the way for future research into human brain function and ne...
A new approach brings transparency to self-driving cars and other self-taught systems by automatically error-checking neural networks. Researchers found thousands of bugs missed by previous techniques, activating up to 100% of network neurons and improving accuracy up to 99%.
Researchers from Lehigh University and Columbia University have developed a new testing approach for deep learning platforms used in self-driving cars, malware-detection, and other systems. Their method, called DeepXplore, exposes thousands of unique incorrect corner-case behaviors, enabling faster identification and fixing of errors.
Researchers have developed an AI-powered technology that can decode what the human brain is seeing by analyzing fMRI scans from people watching videos. The breakthrough could lead to new insights into brain function and improve artificial intelligence.
Researchers developed an algorithm based on ant trail networks, which adapts to changes in the environment and avoids taking the shortest path. The algorithm is inspired by how ants navigate through complex vegetation and repair broken trails using a 'greedy search' method.
Nerve cell networks reorganize themselves during periods of inactivity, becoming hypersensitive and prone to overreaction when signals are reinstated. Researchers developed a high-speed microscopy process to visualize communication networks of living neurons, shedding light on the effects of blocking neural pathways.
A neuroimaging study reveals that blind individuals perform better on a touch discrimination task when their hands are crossed due to stronger frontal-parietal connectivity. In contrast, sighted individuals show greater activity in parietal and premotor areas with uncrossed hands.
The study seeks to discover calcium sources in synapses that prolong neurotransmitter release, a process more expressed in neurodegenerative diseases. Researchers will investigate key players in this process using asynchronous release of neurotransmitters.
Researchers develop a general-purpose technique to analyze neural networks trained for natural-language-processing tasks. The method applies to any black-box text-processing system, revealing idiosyncrasies in human translators' work and identifying gender biases in machine translation systems.
Researchers discovered a new learning mechanism that spans seconds, allowing for the storage of entire sequences of events, including places traversed. This finding challenges the widely accepted Hebbian learning rule, suggesting no causal relationship between interconnected neurons is required to form long-lasting associations.
Researchers from SLAC and Stanford used neural networks to analyze images of strong gravitational lensing, performing complex analyses in a fraction of a second. The technique has the potential to transform astrophysics by analyzing vast amounts of data quickly and automatically.
The Journal of Applied Remote Sensing has awarded three exceptional articles for their outstanding contributions to remote sensing research and applications. The winning articles focus on ice cloud measurement, a neural-network architecture for scene classification, and through-wall imaging.
Researchers developed neural networks to evaluate short narratives, improving predictions over a baseline system. The AIs classified texts into popular and non-popular categories, highlighting the importance of understanding story structures in narrative evaluation.
Machine learning scientists at Disney Research developed a dynamic word embeddings model that uncovers how the meanings of words change over time. The model, which integrates neural networks and statistics used in rocket control systems, detects semantic change throughout history by analyzing semantic vector spaces.
Researchers have developed a method for designing energy-efficient neural networks, reducing power consumption by up to 73% compared to standard implementations. The new approach uses an analytic tool to evaluate and prune low-weight connections, resulting in more efficient networks with fewer connections.
Researchers developed a new algorithm that can turn audio clips into highly-realistic videos of people speaking, using available public domain video footage. This technology has potential applications in improving video conferencing and creating realistic virtual reality experiences.
The Wyss Center is working on a $19M project to develop a high-resolution, implantable neural interface network that can record and stimulate neural activity. The system, dubbed 'neurograins', aims to provide new treatments for sensory deficits and monitor physiological parameters in real-time.
Researchers develop fully automated method to analyze neural networks trained on visual data, shedding light on node firing patterns and emphasis on different visual properties. The approach provides specific insights into the organization of human brain and computer vision algorithms.
Researchers used mathematics and MRI to better understand how neurological disorders affect brain connections. They discovered sub-networks called eigenmodes, which communicate information between brain regions.
A study from UT Southwestern Medical Center reveals a network of neurons vital for learning vocalizations in songbirds, which may hold clues to addressing speech disorders in humans. The discovery complements ongoing research into the brain's role in vocal learning and its potential applications for treating neurodevelopmental conditions.
Researchers at Cleveland Clinic conducted the first randomized controlled trial of DBS for neuropathic pain, targeting brain structures related to emotion and behavior. The study showed significant improvements in indices of depression, anxiety, and quality of life, suggesting a potential shift away from analgesia-based treatments.
Researchers at UC Berkeley created a robot called DexNet 2.0 that can pick up and move unfamiliar objects with high accuracy, using deep learning to learn grasps for 6.7 million 3D shapes.
Researchers at Rice University have developed a new technique that reduces computational overhead for deep learning by up to 95% using hashing, a tried-and-true data-indexing method. The technique blends locality-sensitive hashing and sparse backpropagation to achieve significant savings in energy and time.
A team of Kyoto University researchers has successfully used neural network-based artificial intelligence to decode and predict visual content in the human brain. The technology, known as Deep Neural Network (DNN), shows promise for improving brain-machine interfaces and potentially even understanding consciousness.
A recent study from the Center for BrainHealth found that strategic brain training positively affects neural connectivity for individuals with TBI. The research challenges the widely held belief that recovery from a TBI is limited to two years after an injury.
Researchers at the University of Michigan have developed a new memristor chip that can process complex data, such as images, much faster and with less power than traditional systems. Inspired by how mammals see, the chip uses pattern recognition to shortcut energy-intensive processes.
Researchers have discovered how the enteric nervous system, a complex network of nerve cells in the gut, is formed during mouse development. The study reveals that individual progenitor cells produce specific types of cells, which form overlapping columns and exhibit synchronized activity.
A neural explanation for 'monkey see, monkey do' involves a specialized circuit in primates analyzing social interactions like grooming and fighting. The study found that brain networks associated with visual features are highly active when observing these interactions, similar to the human brain's social interaction systems.
Researchers developed a neural network-based method to stylize photos without losing original image details. The technique uses deep machine learning to preserve boundaries and edges while transferring styles.
A NJIT graduate student has won an IBM fellowship to develop computer systems that mimic the human brain's architecture. The goal is to raise computers' learning and inference capabilities while lowering energy consumption.
The article discusses the challenges of copyright protection in the era of automated art, particularly with the rise of Deep Neural Networks. While DNN creations can generate original works, the issue of attribution and originality remains complex due to varying national laws and human inputs.
Scientists at The Scripps Research Institute found that neurons can form networks independently of synaptic activity, suggesting genetic programs control neural circuit assembly. These findings were confirmed by a complementary study at the Max Planck Institute for Experimental Medicine.
Researchers developed a new method to measure brain health by analyzing neural networks' responses to artificial touch experiences provided by a finger prosthesis. The technique offers precise insights into the cooperation between neurons and can reflect the entire brain's health, providing potential breakthroughs in neurological disea...
Researchers developed high-speed computational software 'Parallel STEPS' to model neuronal interactions and functions. The new approach achieved significantly improved performance, enabling faster simulation of complex models and revealing new insights into individual neuron behavior.
Researchers observe how large groups of neurons learn and unlearn a new association in the brain, blurring lines between stimuli. The findings have implications for studying emotional memory disorders like PTSD.
Researchers from the University of Granada found a close association between magnetic systems and certain brain activity states, including 'spin-glass' states that can lead to frozen neuronal activity. This study provides a novel theoretical framework for understanding biological mechanisms behind destabilization of these states.
New insights reveal that inhibitory neurons contribute to finely-tuned networks in the cortex, linking together neurons with similar functional properties. This discovery raises important questions about how these connections are formed during development.
Swiss researchers use neural networks to challenge the resolution limit of telescopes, recovering features that were previously invisible. The technique, inspired by a generative adversarial network, achieves better results than previous methods, such as deconvolution, and has vast potential for future astronomical observations.
The proposed SA-RBF-TSE algorithm corrects TDOA measurements using a Self-Adaptive RBF neural network and improves position estimation. It adapts the number of RBF nodes and center vector based on error distribution, resulting in high accuracy and reliability compared to other algorithms.
Researchers Juan Cassasquilla and Roger Melko repurpose Google's TensorFlow algorithm to distinguish phases of a simple magnet and find the boundary between phases. The successful results open up new opportunities for research and potential real-world applications in condensed matter physics.
Researchers at University of Bristol and UCLan uncover new type of LTP controlled by kainate receptors, promising therapeutic strategies for dementia and epilepsy. The study's findings have far-reaching implications for understanding memory and neurodegenerative disorders.
A new chip designed by MIT researchers has the potential to make voice control ubiquitous in electronics, offering significant power savings. The chip's ability to minimize memory bandwidth and compress weights associated with each node enables efficient speech recognition, making it practical for relatively simple electronic devices.
Researchers developed a novel neural network method that can categorize complex datasets without prior knowledge. By using an 'act as if' principle, they trained networks to mimic human intuition, allowing them to identify boundaries in data. This method has potential applications in physics analysis, machine learning, and data mining.