Add BrightSurf on Google Email

Light meets deep learning: computing fast enough for next-gen AI

Researchers developed a novel design for the chip using a crossbar layout, outperforming state-of-the-art photonic counterparts in terms of scalability and technical versatility. The synergy of powerful photonics with the novel crossbar architecture enables next generation neuromorphic computing engines.

SourceInstitute of Electrical and Electronics Engineers·JournalIEEE Journal of Selected Topics in Quantum Electronics·TypeLiterature review·DateMar 22, 2023

Harnessing incoherence to make sense of real-world networks

A new approach to describing network connections can help predict system strong and weak points, crucial for understanding disease spread and communication networks. Researchers found that mapping hierarchies and incoherence within a system enables prediction of strong and weak connections.

SourceUniversity of Birmingham·JournalProceedings of the National Academy of Sciences·TypeComputational simulation/modeling·DateMar 20, 2023

Characterizing abnormal neural networks in dogs with anxiety

Researchers found that dogs with anxiety have altered brain connectivity, particularly between the amygdala and hippocampus. The study used fMRI to characterize abnormal neural networks in anxious dogs, providing insight into anxiety disorders in both animals and humans.

SourcePLOS·JournalPLOS ONE·TypeObservational study·DateMar 15, 2023

Mapping unknown territory

Scientists at Max-Planck-Gesellschaft created an interactive atlas of gene expression in the zebrafish brain, revealing hundreds of genes with single-cell resolution. The new map integrates seamlessly with existing data, providing new insights into neural structure and function.

SourceMax-Planck-Gesellschaft·JournalScience Advances·TypeObservational study·DateMar 1, 2023

AI analyses cell movement under the microscope

Researchers at University of Gothenburg developed AI method using graph theory and neural networks to analyze cell movement, enabling better understanding of biological processes and development of new medical technologies. The method can reconstruct cell paths and test medication effectiveness as potential cancer treatments.

SourceUniversity of Gothenburg·JournalNature Machine Intelligence·TypeExperimental study·DateFeb 16, 2023

Charting a course in the brainy frontier

Kyoto University researchers have created a map comparing circuit structure with neural activity in mammals, revealing a new mechanism behind visual cortex activities. This discovery sheds light on the hidden connections between neurons and could provide directions for constructing power-efficient deep neural networks.

SourceKyoto University·TypeExperimental study·DateFeb 16, 2023

Novel optical and fMRI platform identifies brain regions that control large-scale brain network

Researchers have developed a novel experimental platform that combines optical and fMRI techniques to study large-scale brain networks. The platform identified the anterior insular cortex as a key player in controlling the Default Mode Brain Network, which is active during daydreaming, memory retrieval, and envisioning the future.

SourceUniversity of North Carolina Health Care·JournalScience Advances·TypeExperimental study·DateFeb 15, 2023

Scientific AI’s ‘black box’ is no match for 200-year-old method

A new study uses Fourier analysis to understand how deep neural networks learn complex physics. By analyzing the equation of a fully trained model, researchers were able to identify crucial information about how the network learns and generalizes. This breakthrough could accelerate the use of scientific deep learning in climate science.

SourceRice University·JournalPNAS Nexus·TypeComputational simulation/modeling·DateFeb 13, 2023

Helpful disturbance: How non-linear dynamics can augment edge sensor time series

Engineers at Tokyo Institute of Technology have developed a technique to support the classification performance of neural networks operating on sensor time series by feeding recorded signals into elementary non-linear dynamical systems. This approach increases the classification performance by augmenting the data through additional tim...

SourceTokyo Institute of Technology·JournalChaos Solitons & Fractals·TypeExperimental study·DateJan 25, 2023

Computers reimagined

Prof. Shahar Kvatinsky's neuromorphic chip integrates storage and processing functionalities, achieving 97% handwritten letter recognition accuracy with low energy consumption. The chip's design enables potential applications in camera sensors, eliminating the need for digital image enhancement.

SourceTechnion-Israel Institute of Technology·JournalNature Electronics·TypeExperimental study·DateJan 15, 2023

Artificial Intelligence searches an early sign of osteoarthritis from an x-ray image – might save from unnecessary treatments and examination

Researchers developed an AI-based neural network to detect early knee osteoarthritis from x-ray images, matching doctors' diagnoses in 87% of cases. This method could help reduce unnecessary examinations, treatments, and even knee joint replacement surgery.

SourceUniversity of Jyväskylä - Jyväskylän yliopisto·JournalDiagnostics·TypeObservational study·DateDec 15, 2022

USTC make breakthroughs in atomistic neural network representations for chemical dynamics simulations

A USTC team has made significant advancements in atomistic neural network (AtNN) representations for chemical dynamics simulations. By decomposing system properties into atomic contributions, AtNN can satisfy different symmetries and periodicities, achieving accurate and efficient molecular dynamics simulations of complex systems.

SourceUniversity of Science and Technology of China·JournalWiley Interdisciplinary Reviews Computational Molecular Science·DateDec 14, 2022

Effects of antidepressants taken during pregnancy are poorly understood, scientists note

A review of over 100 scientific articles suggests that the safety of antidepressants during pregnancy is endorsed by science, but their effects on fetal neurodevelopment are poorly understood. Brazilian researchers propose using lab-grown mini-brains to investigate this impact and identify potential alterations.

SourceFundação de Amparo à Pesquisa do Estado de São Paulo·JournalSeminars in Cell and Developmental Biology·TypeSystematic review·DateDec 13, 2022

Bolstering the safety of self-driving cars with a deep learning-based object detection system

Researchers at Incheon National University have developed an IoT-enabled, real-time object detection system for autonomous vehicles. The YOLOv3-based model achieved high accuracy (>96%) in detecting 2D and 3D objects, outperforming other state-of-the-art detection models.

SourceIncheon National University·JournalIEEE Transactions on Intelligent Transportation Systems·TypeComputational simulation/modeling·DateDec 12, 2022

Brain aging is not one-size-fits-all: Chronological age, biological age and gender affect the shrinkage of different brain areas

Research reveals that brain function networks are affected differently by aging, gender, and blood immune factors. The study found correlations between cytokine clock, brain shrinkage, and gender, with females having a faster ticking cytokine clock.

SourceBuck Institute for Research on Aging·JournalProceedings of the National Academy of Sciences·TypeData/statistical analysis·DateDec 2, 2022

Nanoengineers develop a predictive database for materials

The researchers have developed an AI algorithm called M3GNet that can predict the structure and dynamic properties of any material. The algorithm was used to create a database of over 31 million yet-to-be-synthesized materials with predicted properties, facilitating the discovery of new technological materials.

SourceUniversity of California - San Diego·JournalNature Computational Science·TypeComputational simulation/modeling·DateNov 28, 2022

Chung-Ang University researchers develop algorithm for optimal decision making under heavy-tailed noisy rewards

Chung-Ang University researchers propose a new algorithm, MR-UCB and MR-APE, to tackle stochastic multi-armed bandit problems with heavy-tailed noise distributions. The methods guarantee minimal loss for worst-case scenarios with minimal prior information.

SourceChung Ang University·JournalIEEE Transactions on Neural Networks and Learning Systems·TypeComputational simulation/modeling·DateNov 22, 2022

A 5G-enabled AI-based malware classification system for the next generation of cybersecurity

A novel AI-based malware detection and classification system has been developed for 5G-enabled Industrial Internet of Things (IIoT) systems. The system achieved an accuracy rate of 97% on benchmark datasets, enabling the secure connection of applications such as smart cities and autonomous vehicles.

SourceIncheon National University·JournalIEEE Transactions on Industrial Informatics·TypeSystematic review·DateNov 8, 2022

CABBI team adds powerful new dimension to phenotyping next-gen bioenergy crop

Researchers at CABBI used unmanned aerial vehicles with machine learning methods to select the best candidate genotypes in miscanthus breeding programs. The new method leverages high-resolution aerial imagery and three-dimensional neural networks to estimate crop traits such as flowering time, height, and biomass production.

During sleep, one brain region teaches another, converting novel data into enduring memories

Researchers used a neural network model to study how brain regions interact during sleep. They found that the hippocampus and neocortex work together to convert fleeting information into long-term memory. The findings suggest that alternating between REM and slow-wave sleep stages is crucial for strong memory formation.

SourceUniversity of Pennsylvania·JournalProceedings of the National Academy of Sciences·DateOct 24, 2022

Huge unveiling of schizophrenia brain cells show new treatment targets

Researchers at the University of Copenhagen have made a breakthrough in understanding schizophrenia by analyzing individual brain cells. The study identified specific neurons and networks affected by the disease, suggesting that targeting these areas could lead to new treatment options.

SourceUniversity of Copenhagen - The Faculty of Health and Medical Sciences·JournalScience Advances·TypeExperimental study·DateOct 24, 2022

Deep learning with light

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.

Engineers record neurons to pinpoint synaptic links

Researchers created a 3D electrode array that maps the locations and activity of up to 1 million potential synaptic links in living brains. The system uses recordings of millisecond-scale evolution of electrical pulses in tens of thousands of neurons, allowing for dense and accurate mapping of brain circuits.

SourceRice University·JournalNature Biomedical Engineering·TypeExperimental study·DateOct 19, 2022