Researchers use artificial intelligence to ID mosquitos
A Convoluted Neural Network was trained on a library of mosquito images to classify species, sex, and strain. The system achieved 99.96% prediction accuracy for class identification.
Artificial Neural Networks
Articles tagged with Artificial Neural Networks
AI that detects post-stroke depression type can help stroke survivors get right treatment
Researchers developed an AI that can detect the type of post-stroke depression (PSD) in patients, which is crucial for providing the right treatment. The study used a probabilistic artificial neural network to analyze data from 274 patients and found that early detection is key to improving functional recovery.
Misinformation or artifact: a new way to think about machine learning
Researchers exploring the nature of AI failures reveal 'adversarial examples' may not be intentional mistakes. Instead, they might be 'artifacts' created by interactions between network and data patterns. This rethink suggests that misfires could offer useful information if interpreted correctly.
Do neural networks dream visual illusions?
Researchers studied how convolutional neural networks respond to brightness and color visual illusions, finding that they are similarly deceived as humans. The study highlights the limitations of CNNs in mimicking human vision, revealing both similarities and differences between the two.
New data processing module makes deep neural networks smarter
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.
How to make AI trustworthy
A new tool, DeepTrust, generated automatic indicators of data and prediction trustworthiness in neural networks, addressing the need for trust in AI. The researchers used subjective logic to assess neural network architectures, providing insights into testing reliability and maximizing accuracy.
SourceUniversity of Southern California·JournalFrontiers in Artificial Intelligence·DateAug 27, 2020
New neural network differentiates Middle and Late Stone Age toolkits
Researchers developed a neural network to distinguish between Middle and Late Stone Age assemblages by analyzing frequent tool combinations. The study found that the combined occurrence of backed pieces, blade technologies, and absence of core tools reliably identifies Late Stone Age assemblages.
A leap forward for biomaterials design using AI
A team of researchers at Tokyo Tech successfully used machine learning with an artificial neural network model to predict two key properties of self-assembled monolayers, enabling advanced material screening and design. This approach opens up new possibilities for the development of biomaterials with desired functions.
Argonne scientists use artificial intelligence to strengthen power grid resiliency
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.
Teaching physics to neural networks removes 'chaos blindness'
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.
Artificial brains may need sleep too
Artificial neural networks became unstable after continuous unsupervised learning, but exposure to Gaussian noise mimics slow-wave sleep stabilized them. This finding has implications for the development of biologically realistic AI systems.
Get excited by neural networks
Scientists at UTokyo-IIS developed a machine learning algorithm to infer excited states from ground states of materials. The algorithm used artificial neural networks to analyze data from core-electron absorption spectroscopy, revealing new insights into chemical reactivity and material function.
SourceInstitute of Industrial Science, The University of Tokyo·Journalnpj Computational Materials·DateJun 3, 2020
AI stock trading experiment beats market in simulation
Researchers developed a novel AI-managed trading strategy that outperforms traditional methods, achieving greater gains and fewer losses. The proposed system utilizes convolutional neural networks to analyze layered images of current and past market data, leading to more accurate predictions and reduced randomness.
SourceChinese Association of Automation·JournalIEEE/CAA Journal of Automatica Sinica·DateJun 1, 2020
Early Bird uses 10 times less energy to train deep neural networks
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.
Neural hardware for image recognition in nanoseconds
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.
Quantifying objects: bees recognize that six is more than four
Insects can perform numerical estimation and compare sets of objects, recognizing larger quantities. Researchers found a simple model for neural networks to learn numerical cognition tasks using a single neuron.
New study allows brain and artificial neurons to link up over the web
A novel nanoelectronics device has enabled brain neurons and artificial neurons to communicate with each other over the internet. This breakthrough study shows how three key emerging technologies can work together: brain-computer interfaces, artificial neural networks and advanced memory technologies.
Neuroscience opens the black box of artificial intelligence
Researchers at Otto-von-Guericke-Universität Magdeburg are using brain research methods to analyze artificial neural networks and improve explainable AI. The Cognitive neuroscience inspired techniques project aims to understand the internal processes of ANNs and identify malfunctions.
New artificial neural network model bests MaxEnt in inverse problem example
A new artificial neural network model has been developed to solve inverse problems, demonstrating accuracy comparable to the maximum entropy (MaxEnt) approach. The model's versatility and robustness against noisy data have been showcased in various tests, including recovering electron single-particle spectral densities.
SourceNational Centre of Competence in Research (NCCR) MARVEL·JournalPhysical Review Letters·DateFeb 17, 2020
Artificial intelligence is becoming sustainable!
Researchers at Politecnico di Milano developed a novel circuit that can execute advanced AI operations in one operation, reducing energy consumption and paving the way for more sustainable AI computing accelerators. This breakthrough enables faster and more efficient training of neural networks, crucial for applications like facial rec...
Finally, machine learning interprets gene regulation clearly
Researchers have developed a custom artificial neural network that can analyze molecular signals controlling gene function, enabling biologists to understand complex mechanisms of gene regulation. This breakthrough enables the creation of machine learning algorithms that reflect common concepts in biology.
Synthesizing an artificial synapse for artificial intelligence
Researchers at the University of Pittsburgh have developed an artificial synapse that mimics the human brain's ability to create neuronal connections. This breakthrough technology could revolutionize AI and cognitive computing, enabling faster and more efficient processing of complex tasks.
Deep neural networks uncover what the brain likes to see
Researchers developed a novel computational approach using deep artificial neural networks to predict neural responses to images. The study found that certain stimuli, such as checkerboards or sharp corners, elicit strong responses from neurons, contradicting current dogma in the field.
Agriculture of the future: Neural networks have learned to predict plant growth
Researchers trained neural networks to predict plant growth patterns using computer vision algorithms and efficient graphics processing units. The system uses Raspberry Pi with Intel Movidius graphics card to calculate and predict the optimal ratio of nutrients, enabling continuous monitoring and prediction in artificial growing systems.
SourceSkolkovo Institute of Science and Technology (Skoltech)·JournalIEEE Transactions on Instrumentation and Measurement·DateOct 31, 2019
Artificial networks shed light on human face recognition
A new study reveals that human brains process faces in a similar way to artificial intelligence systems, with unique activation patterns playing a key role in recognition. The researchers found parallels between the human visual system and deep neural networks, which can improve face recognition capabilities.
Researchers make neural networks successfully detect DNA damage caused by UV radiation
Artificial neural networks successfully identify minor changes in DNA structure caused by UV radiation, enabling early detection of potential cancer risks. The technique uses surface-enhanced Raman spectroscopy and has the potential to be used for medical diagnostics.
Expanding the use of AI through the Internet of Things
Researchers at the University of Delaware are developing new memory devices that can support neural networks in low-power embedded systems. These advancements aim to improve the lifetime and reliability of IoT devices, which currently struggle with battery power and memory constraints.
From years to days: Artificial Intelligence speeds up photodynamics simulations
Researchers use deep neural networks to simulate light-induced molecular reactions on long time scales, accelerating computation by up to 19 years. This method enables better understanding of biological processes like carcinogenesis and ageing, with potential applications in material ageing and photosensitive drugs.
Deep learning pioneer to give Turing Lecture at Heidelberg Laureate Forum
Bengio will discuss recent successes and limitations of deep learning, as well as research directions for building human-level AI. The event is part of the Heidelberg Laureate Forum, a networking event bringing together young researchers and award recipients.
Researchers demonstrate all-optical neural network for deep learning
A two-layer all-optical artificial neural network has been successfully demonstrated for complex classification tasks, outperforming computer-based neural networks. The researchers plan to expand this approach to large-scale optical deep neural networks for specific practical applications.
Understanding the animal brain could help robots wash your dishes
Researchers believe that studying animal brains can improve AI's ability to tackle complex tasks like dish-washing. By understanding how biological neural networks work, AI systems may be able to overcome barriers and achieve superhuman performance.
Neural networks will help manufacture carbon nanotubes
Researchers have developed an effective method to monitor carbon nanotube films using artificial neural networks (ANN). The technique can help predict the efficiency of single-walled carbon nanotubes synthesis and improve the overall production framework, leading to new horizons for real-life applications.
Possibilities of the biosimilar principle of learning are shown for a memristor-based neural network
Scientists from Russia and Greece have successfully implemented a spiking neural network based on memristors, demonstrating the feasibility of local learning rules. The research enables autonomous unsupervised learning of complex neural networks, paving the way for new applications in AI.
Which is the perfect quantum theory?
Researchers employed machine learning to analyze images of quantum systems and identify the most predictive theory. The study used artificial neural networks to distinguish between competing theories, selecting the one that best described observed phenomena in high-temperature superconductors.
Machine learning reveals how strongly interacting electrons behave at atomic level
Scientists have made a breakthrough in understanding the behavior of strongly interacting electrons using machine learning techniques, discovering a new state called Vestigial Nematic State. The technique uses artificial neural networks to recognize different forms of electronic matter and reveals symmetries of complex image-arrays fro...
Controlling primate neural activity using artwork from artificial neural networks
Researchers controlled specific neurons in macaques' brains with images generated by artificial neural networks, demonstrating a new tool for neuroscientists to design experiments. This breakthrough uses current computer vision applications to predict and control visually evoked neural responses in primate brains.
Can science writing be automated?
A team of scientists at MIT developed a neural network that can read scientific papers and generate a plain-English summary. The system, called RUM, uses vectors rotating in multidimensional space to represent words and improve memory and recall capabilities.
SourceMassachusetts Institute of Technology·JournalTransactions of the Association for Computational Linguistics·DateApr 18, 2019
Deep learning merges advantages of holography and bright-field microscopy for 3D imaging
Researchers developed Bright-field Holography to overcome limitations of holographic 3D imaging. The method combines the image contrast advantage of bright-field microscopy with the snapshot volumetric imaging capability of holography, allowing for rapid creation of images equivalent to those from a bright-field microscope.
Researchers use artificial neural networks to streamline materials testing
A team at NYU Tandon School of Engineering has designed an artificial neural network approach that can predict the elastic modulus of graphene-enhanced composites from just one sample, streamlining materials testing. This reduces the need for extensive experimentation, lowering costs and accelerating product development.
Machine learning and quantum mechanics team up to understand water at the atomic level
Researchers used artificial neural networks to learn atomic interactions from quantum mechanics, bypassing complex calculations. The ANN was used as a surrogate model to account for errors and improve predictions.
SourceNational Centre of Competence in Research (NCCR) MARVEL·JournalProceedings of the National Academy of Sciences·DateJan 7, 2019
Hardware-software co-design approach could make neural networks less power hungry
A team of researchers developed a neuroinspired hardware-software co-design approach that can make neural network training more energy-efficient and faster. The approach uses a type of energy-efficient neural network called spiking neural networks, combined with the soft-pruning algorithm to minimize computing power and time.
NYU researchers pioneer machine learning to speed chemical discoveries, reduce waste
Researchers at NYU Tandon School of Engineering have developed a machine learning system that pairs artificial neural networks with infrared imaging to control and interpret small-scale chemical reactions. This technique can reduce the decision-making process from one year to weeks, saving tons of chemical waste and energy.
Artificial intelligence aids automatic monitoring of single molecules in cells
Researchers at Osaka University have developed an AI-based system to track individual fluorescently labeled molecules in living cells. The system can analyze hundreds of thousands of molecules in a short period, providing reliable data on molecule status and dynamics.
Artificial intelligence helps reveal how people process abstract thought
A study by Cameron Buckner using deep neural networks suggests that human knowledge stems from sensory experience, a school of thought known as empiricism. The networks demonstrate how abstract knowledge is acquired and can be used to understand complex tasks in neuroscience and psychology.
Scientists use artificial neural networks to predict new stable materials
Researchers use artificial neural networks to predict crystal stability in garnets and perovskites, achieving accuracy up to 10 times that of previous models. The team's web application allows for fast computation of material properties on various devices.
Helping computers fill in the gaps between video frames
Researchers at MIT develop Temporal Relation Network (TRN) module to help CNNs recognize activities by observing key frames. The module achieves top accuracy of 95% in activity recognition on Jester dataset, outperforming existing models.
Attacking aftershocks
Using deep learning algorithms, researchers have developed a system that forecasts aftershocks significantly better than random assignment. By analyzing earthquake data and physics-based models, they identified the second invariant of the deviatoric stress tensor as an important factor in predicting aftershock locations.
Machine learning technique reconstructs images passing through a multimode fiber
Researchers use deep neural networks to recognize images transmitted over optical fibers, achieving high accuracy despite distortions caused by environmental factors. The technique has potential for improving endoscopic imaging in medical diagnosis and increasing the information-carrying capacity of fiber-optic telecommunication networks.
UCLA-developed artificial intelligence device identifies objects at the speed of light
The UCLA-developed artificial neural network device can analyze large volumes of data and identify objects at the speed of light. It uses diffraction of light to process images without advanced computing programs or energy consumption.
The ultimate combination: A 3D-printed optical deep learning network
A new 3D-printed optical deep learning network called Diffractive Deep Neural Network (D2NN) has been developed by Xing Lin and colleagues. This system processes information through layers of optically diffractive surfaces that work together to recognize handwritten digits with high accuracy.
If only A.I. had a brain
Researchers developed an artificial synapse inspired by the human brain, which efficiently processes information and demonstrates excellent energy efficiency. This breakthrough could lead to the development of energy-efficient neuromorphic computing, revolutionizing AI devices and transforming industries.
Researchers move closer to completely optical artificial neural network
Artificial neural networks can now be trained directly on an optical chip, paving the way for less expensive, faster, and more energy-efficient AI. This breakthrough enables complex tasks like speech or image recognition to be performed more efficiently.
Training artificial intelligence with artificial X-rays
A new approach uses machine learning to generate computer-generated X-rays to supplement real images, increasing the size of training sets for AI systems. This method improves classification accuracy for common and rare conditions by up to 40%, overcoming a challenge in applying artificial intelligence to medicine.
Test tube artificial neural network recognizes 'molecular handwriting'
Researchers at Caltech developed an artificial neural network made of DNA that can accurately identify handwritten numbers. The network, designed by Kevin Cherry, uses a 'winner take all' competitive strategy and undergoes complex reactions to classify molecular information.
Using artificial intelligence to understand volcanic eruptions from tiny ash
Scientists developed an AI program to categorize volcanic ash particle shapes, helping provide information on eruptions and aid hazard mitigation. The program achieved a success rate of 92% in classifying basal shapes with probability ratios.
Artificial intelligence helps to predict likelihood of life on other worlds
A team at Plymouth University used artificial neural networks to classify five planets as potentially habitable, estimating a probability of life in each case. The technique may aid in selecting targets for future space missions with improved spectral detail.
How a yeast cell helps crack open the 'black box' behind artificial intelligence
Researchers developed a visible neural network, DCell, that uses real-world cellular behaviors and constraints to predict cellular growth. The system can simulate cellular growth nearly as accurately as a real cell grown in a laboratory.
Engineers design artificial synapse for 'brain-on-a-chip' hardware
Researchers at MIT have designed an artificial synapse that can precisely control the strength of an electric current flowing across it, similar to the way ions flow between neurons. The team found that their chip and its synapses could recognize samples of handwriting with 95% accuracy.
Memristors power quick-learning neural network
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.
New algorithm repairs corrupted digital images in one step
A new algorithm, developed by a team at the University of Maryland, uses artificial neural networks to address multiple flaws in a single image. The algorithm can be trained on high-quality images and then applied to any image with imperfections.