Articles tagged with Artificial Neural Networks
Researchers at WVU are developing software for robots to learn and adapt in real-time, inspired by the neural networks of electric fish. The goal is to enable robots to navigate different terrains autonomously without human supervision.
Researchers from University of Warsaw create spiking neuron using photons to mimic biological brain's behavior. This achievement paves the way for photonic neural networks that process information faster and more efficiently than conventional systems.
Researchers have developed a deep learning algorithm that can accurately assess the stage of head and neck cancer using standard CT scans, outperforming expert radiologists. The algorithm demonstrated superior accuracy in measuring the extent of cancer spread, especially for patients with high-risk disease.
Researchers at MIT have developed a new method that uses optics to accelerate machine-learning computations on low-power devices. By encoding model components onto light waves, data can be transmitted rapidly and computations performed quickly, leading to over a hundredfold improvement in energy efficiency.
Researchers develop mechanical neural networks (MNNs) with tunable beams that can learn behaviors and adapt to external forces. The MNNs, composed of a triangular lattice pattern, exhibit smart properties through machine learning algorithms. Early prototypes overcame lag issues and achieved accurate performance in various applications.
A new CNN framework, PE-Net, is proposed for predicting machine remaining useful life (RUL) accurately. The framework uses a novel architecture with small-sized one-dimensional convolution kernels and deep networks to learn features from input time series signals.
Scientists have developed a solution to communication challenges in neuromorphic chips using superconducting devices. This allows artificial neural systems to operate 100,000 times faster than the human brain, with potential applications in industrial control and human conversations.
A team of researchers at Harvard University has developed an ionic circuit that performs analog matrix multiplication, a key operation in neural networks, using ions in liquid. The breakthrough uses a pH-gated ionic transistor and expands to a 16x16 array for more complex computations.
A recent grant will fund a project developing new hardware for machine learning, aiming to curb unsustainable energy use in AI systems. The new algorithms being developed are made available to the research community and compatible with an openly shared computing platform.
Current AI models are restricted by a lack of experience in real-world environments, despite achieving significant advancements in virtual settings. Researchers are now exploring ways to bridge this gap with foundation models that can operate in physical spaces.
A recent study published in Aging-US found that feeling lonely, unhappy, or hopeless increases one's biological age more than smoking. The research used digital models of aging to analyze the effects of various factors on aging rates, revealing a significant correlation between mental health and accelerated aging.
City digital twin technology is used to create synthetic training data for deep learning models, which are then trained on a combination of real and synthetic data. This approach yields promising results for architectural segmentation tasks, particularly for modern building styles.
A team led by York University has developed a new technique to keep drinking water safe in refugee settlements using machine learning and ensemble forecasting systems. The approach can predict the probability of residual chlorine remaining in stored water, providing critical information for aid workers to ensure safe drinking water.
Rice University's ROBE Array algorithm slashes the size of DLRM memory structures, allowing training on 100 megabytes of memory and a single GPU. The method matches state-of-the-art DLRM training methods with improved inference efficiency.
Researchers at MIT developed an AI model that can detect Parkinson's disease from breathing patterns, using a neural network to assess the presence and severity of the condition. The device is non-invasive and can be used in patients' homes without any bodily contact.
The NeuRRAM chip demonstrates wide range of AI applications with equivalent accuracy while reducing energy consumption by up to 70% compared to traditional compute platforms. It also supports various neural network models and architectures, enabling diverse AI applications on edge devices.
The study demonstrates the creation of physical reservoirs using chaotic dynamics, enabling alternative approach to AI-based pattern detection. The researchers exploited emergence and pattern formation phenomena under incomplete synchronization in chaotic dynamics, revealing a rich variety of ways in which the network synchronizes.
Researchers developed a Flashover Prediction Neural Network (FlashNet) model to forecast deadly fire events, beating other AI-based tools with up to 92.1% accuracy across various building floorplans. The model's performance improved when given real-world data, highlighting its potential for saving firefighter lives.
Researchers successfully taught microrobots to swim via deep reinforcement learning, allowing them to adapt to changing conditions and perform complex maneuvers. The AI-powered swimmers can navigate toward any target location on their own, showcasing their robust performance in fluid flows and uncontrolled environments.
Researchers at MIT have developed a machine-learning system that uses computer vision to monitor the 3D printing process and correct errors in real-time. The system successfully printed objects more accurately than other 3D printing controllers, enabling engineers to incorporate novel materials into their prints with ease.
The new AI system uses associative learning to detect similarities in datasets, reducing processing time and computational cost. By leveraging optical parallel processing and light signals, the system can identify patterns and associations more efficiently than conventional machine learning algorithms.
Artificial neural networks use white matter cracks and cerebral cortex furrows to estimate biological age, which can indicate possible disease or injury. The 'black box' problem has been solved, revealing the algorithm's decision-making process.
Researchers at UVA have created a new AI tool called SITHCon that can decode speech in different speeds, enabling machines to process information more efficiently. This breakthrough technology has the potential to alter how artificial neural networks 'think' and reduce AI's massive carbon footprint.
Researchers at NIST have developed a new type of hardware for AI that uses magnetic tunnel junctions, which are less energy-intensive than traditional silicon chips. The new technology has already passed a virtual wine-tasting test and shows promise for reducing energy use in AI systems.
Researchers from Osaka University found that facial similarity plays a crucial role in ratings of trustworthiness for observers of the same sex, but not for observers of opposite sex. The study suggests that facial similarity is an important factor affecting social judgments for same-sex interactions.
Researchers developed a novel convolutional neural network for facial expression recognition, outperforming conventional models while being computationally less expensive. The new model achieved an accuracy of 72.4% using only 58,000 parameters.
A team of researchers led by Danilo Vasconcellos Vargas has developed a new method called 'Raw Zero-Shot' to evaluate the robustness of artificial neural networks in image recognition. The study found that Capsule Networks produced the densest clusters, indicating improved transferability and potential solutions for improving AI robust...
Researchers from the University of Tsukuba develop a model that combines multiple theories to simulate motor learning in humans. The study found that larger amounts of motor exploration aid in learning sensitivity derivatives and transforming errors into motor corrections.
A new AI system uses artificial neural networks to recognize objects more accurately and stably, despite changing visual inputs. The system mimics human eye movements to improve machine vision capabilities, reducing errors in self-driving cars and other applications.
A new AI-powered mental health application, FuturSelf, uses machine learning to identify the shortest path to mental stability. The system offers personalized recommendations for improving long-term well-being.
Researchers use machine learning to automatically analyze Reflection High-Energy Electron Diffraction (RHEED) data, enabling faster and more efficient discovery of new materials. The study focused on surface superstructures in thin-film silicon surfaces and identified optimal synthesis conditions using non-negative matrix factorization.
Researchers have developed an AI-powered approach to calculate molecular spectra using Graph Neural Networks (GNNs), significantly reducing computation time and improving accuracy. The SchNet model achieved a 20% increase in accuracy while reducing computational time, enabling the analysis of complex molecules like quantum dots.
The Department of Energy has awarded Early Career Research Program funding to three Oak Ridge National Laboratory scientists. The awardees will receive $500,000 annually for five years to support their research in fusion energy, advanced scientific computing, and biogeochemical controls on phosphorus cycling.
A new training algorithm for deep spiking neural networks (SNNs) uses biologically plausible spatiotemporal adjustment to improve performance and reduce energy consumption. This approach achieves competitive classification accuracy with only 3% of the energy used by traditional artificial neural networks.
A team from Nagoya University created an artificial neural network model that performed the delayed matching-to-sample task and analyzed its behavior. The model was able to evolve to exhibit human-like metamemory, adapting to its environment by learning and evolving. This breakthrough aims to create machines with memories like humans.
A study by HBP scientists found that wakefulness, non-REM sleep, and REM sleep have complementary functions for learning: experiencing stimuli, solidifying experiences, and discovering semantic concepts. This research suggests that unusual dreams, simulated using Generative Adversarial Networks, can improve brain learning by introducin...
The university's new Robotics and Autonomous Systems Teaching and Innovation Center (RASTIC) will provide students with hands-on experience in robotics, autonomous systems, and self-driving technology. The lab aims to boost Massachusetts' competitiveness in the tech sector by supporting innovative projects and startups.
Artificial Intelligence can now identify legendary batting techniques used by Sir Donald Bradman and modern players. Researchers developed a deep learning computer vision AI model to detect lateral backlift batters from straight ones.
Researchers at the University of Missouri are applying AI to analyze protein dynamics, identifying potential target sites for new drug therapies. The approach can simulate protein changes related to conditions like cancer, enhancing the chances of successful therapies.
MIT researchers develop ExSum, a framework to formalize explanations of machine-learning models into quantifiable rules. This allows for testing assumptions about model behavior and reveals unexpected insights, such as negative words having sharper contributions to model decisions.
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.
Scientists at the University of Oxford have developed an 'optomemristor' device that facilitates three-factor learning and emulation of biological computations, making it possible to perform complex machine learning tasks. The device uses both light and electrical signals to interact and consume very little energy.
A study found that trainee teachers who received AI-generated feedback improved their diagnostic reasoning, identifying potential learning difficulties in pupils more accurately. The AI system analyzed the trainees' work and provided clear, adaptive feedback.
A new AI-based approach can predict cardiac arrest with significant accuracy, identifying patients at risk and predicting the likelihood of sudden cardiac death. The technology stands to transform clinical decision-making and increase survival rates from lethal arrhythmias.
Researchers have developed a new method called Shared Interest that enables users to aggregate, sort, and rank individual explanations of a machine-learning model's reasoning. This technique uses quantifiable metrics to compare how well the model's reasoning matches human thinking, helping to uncover concerning trends in decision-making.
Researchers found that AI-enhanced diagnosis helps doctors accurately detect fetal congenital heart disease, with fellows making the most accurate diagnoses. The new system uses graphical charts to represent the AI's analysis of ultrasound videos, improving accuracy and trust among medical professionals.
Researchers at MIT developed a framework for robotic manipulation systems that can perform complex tasks using a two-stage learning process. This allows robots to learn abstract ideas about manipulating deformable objects, such as pizza dough, and execute skills to complete tasks.
Adversarially robust models capture aspects of human peripheral processing, with results showing similarity in image transformations and perception alignment. The study's findings shed light on the goals of peripheral processing in humans and could help improve machine learning models.
A new framework for portfolio management uses deep reinforcement learning to predict price trends and make strategic decisions, overcoming limitations of existing systems. The system consists of evolving agent modules and strategic agent modules, allowing for modular design and scalability.
GIST researchers propose a new strategy for crime prevention using artificial intelligence, trained on a large-scale dataset of deviant incident reports and corresponding images. The model, called DevianceNet, can accurately classify and detect deviant places, making it a useful tool in urban safety development.
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.
Anastasios Kyrillidis has won a National Science Foundation CAREER Award to explore the theory and design of non-convex optimization algorithms. His research aims to devise algorithmic foundations and theory that will accelerate problem-solving in machine learning, information processing, and optimization.
Researchers at Purdue University have created a device that can dynamically rewire itself to adapt to new data, enabling artificial intelligence to learn and remember information like the human brain. This breakthrough could lead to more efficient AI systems for tasks such as image recognition and decision-making.
A team of researchers from the Institute of Industrial Science, The University of Tokyo, used a mathematical model to examine the implications of intergenerational learning. They found that learning accelerated the evolutionary process, which may assist in designing more efficient hybrid algorithms.
A new technique uses compression to reduce data transmission size, allowing for efficient federated learning on wireless devices. The approach has been shown to condense data packets by up to 99%, making it suitable for areas with limited bandwidth.
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.
A team of scientists developed an AI-based model to predict personal thermal comfort based on spatial parameters, achieving exceptional accuracy. The study highlights the importance of incorporating architectural features in models to reduce energy consumption.
A team at the University of Washington has created an optical computing system that not only reduces noise but also utilizes it to improve creative output. The system uses a Generative Adversarial Network and demonstrates the viability of this technology at a large scale.