Articles tagged with Artificial Neural Networks
MetaSeg achieves the same segmentation performance as U-Nets but requires 90% fewer parameters, making medical image segmentation more cost-effective. The new approach leverages implicit neural representations to quickly adjust to new images and decode accurate labels.
A team of researchers developed a computational method that can design intrinsically disordered proteins with desired properties. The work uses automatic differentiation to optimize protein sequences and leverages molecular dynamics simulations for precision. This breakthrough has the potential to reveal new insights into diseases like...
Researchers at Institute of Science Tokyo developed a new framework for generative diffusion models by reinterpreting Schrödinger bridge models as variational autoencoders. This approach reduces computational costs and prevents overfitting, enabling more efficient generative AI models with broad applicability.
Researchers assessed the effectiveness of a single session with 'Amanda,' a ChatGPT-4o-based chatbot, versus a brief journaling task in addressing non-abusive relationship conflict. The study found that both interventions improved participants' specific relationship problems, overall relationships, and well-being.
Scientists at Seoul National University have developed a framework to manipulate emergent behavior in animal groups and robot swarms. The approach uses physics-informed AI to learn local interaction rules, enabling the control of collective patterns such as rings, clumps, and flocks.
Researchers at Politecnico di Milano developed photonic chips for training physical neural networks, eliminating digitisation requirements. This allows for faster, more robust, and efficient network training using light signals.
Researchers have developed a new light-based chip that cuts power consumption for image recognition tasks by up to 100 times, using lasers and microscopic lenses fabricated onto circuit boards. This breakthrough enables faster performance and potentially strain-free AI systems.
Researchers have developed a silicon chip that uses light to perform convolution operations for AI, reducing energy consumption and increasing speed. The chip achieves near zero energy performance, leap forward for future AI systems.
The Wits MIND Institute has received a $1 million boost from Google.org, enabling it to drive next-generation breakthroughs in natural and artificial intelligence. The partnership aims to advance the scientific understanding of both natural and artificial intelligence, foster breakthrough research and technological innovation.
A global study surveyed 14,000 patients across 43 countries, finding that those in poorer health were more likely to reject AI. Patients preferred explainable AI and wanted clinicians to make final decisions.
Researchers developed AI models that can identify signs of heart failure in patients from Appalachia using low-tech electrocardiogram results. The models achieved high accuracy and could potentially provide clinicians with an edge in protecting patients' cardiac health.
A former diplomat warns that algorithms lack empathy and intuition, which are essential for successful negotiations. However, AI can streamline diplomacy and amplify human aspirations when used carefully. Diplomats need training in AI ethics and global cooperation to ensure equal access and deployment.
Researchers at the University of Vaasa developed smart packaging that can detect subtle color changes in printed packages, enabling cost-effective solutions for industries like food and beverage, healthcare, and logistics. This technology provides a human-eye accurate and environmentally friendly alternative to electronic sensors, pavi...
Researchers at the University of Rochester are developing biologically inspired predictive coding networks for digital image recognition using analog circuits, which could lead to more efficient drones. The team aims to approach the performance of existing digital approaches and translate it to complex perception tasks needed by self-d...
A team of computer scientists created 2,300 original sudoku puzzles and asked AI tools like OpenAI's ChatGPT to solve them. The results showed that while some AI models could solve easy sudokus, most struggled to provide accurate explanations, raising questions about the trustworthiness of AI-generated information.
Researchers highlight the potential of solvent-based recycling and AI-assisted sorting to recycle complex plastics. However, the study emphasizes that replacing fossil-based plastics with biobased alternatives poses significant challenges, requiring comprehensive approaches and life cycle assessments.
A new study by Cornell University reveals Amazon's AI shopping assistant Rufus gives vague or incorrect responses to users writing in some English dialects like AAE. The researchers propose a framework for evaluating chatbots that can better serve users from diverse linguistic backgrounds.
Researchers found that LLMs are better at representing non-sensorimotor concepts but struggle with sensory and motor concepts. Incorporating sensory input improves LLM performance and leads to more human-like representation, consistent with previous human studies on multimodal learning.
A novel method predicts the working status of high-formwork support systems using a combination of finite element model simulations, deep learning, and large language models. The framework achieves superior performance over existing methods and demonstrates potential applications in complex structures.
A new study reveals that AI systems transition from relying on word positions to meaning-based understanding as they receive enough data for training. The transition occurs abruptly, similar to a phase transition in physical systems, and is driven by the amount of data available.
Researchers developed an AI-powered microscope system to measure soil fungi presence and quantity, providing insights into soil health and fertility. The low-cost optical microscopy with machine learning technology can be used by farmers and land managers worldwide.
A new study in Nature Communications found that AI models exhibit a geometric property called convexity, which helps humans form and share concepts. Convexity is also linked to the performance of AI models on specific tasks.
Researchers pioneer a new AI method to uncover cognitive strategies in decision-making, revealing suboptimal behavioral patterns. The study uses small neural networks to predict animal choices with accuracy comparable to larger neural networks.
Researchers developed a simplified model to explain visual processing in the primary visual cortex, achieving 75% accuracy with fewer layers. The 'minimodels' for individual neurons are just as powerful as large models, providing an accurate and interpretable way to study visual computation.
Researchers used machine learning to simulate galaxy evolution and supernova explosions, achieving speeds four times faster than supercomputers. This breakthrough enables the study of galaxy origins, including the creation of the Milky Way's elements essential for life.
An interdisciplinary team at TU Wien has developed a method that allows for the exact calculation of how reliably a neural network operates within a defined input domain. This enables mathematical guarantees for the safe use of AI in sensitive applications.
A new diagnostic tool uses AI to analyze handwriting signals, detecting subtle motor symptoms associated with Parkinson's disease. The device has shown an average accuracy of 96.22% in distinguishing patients from healthy individuals.
The AI for Good Global Summit 2025 will showcase AI innovations delivering better healthcare and education, reducing disaster risks, ensuring water and food security, and bolstering economic resilience. The event, organized by the International Telecommunication Union (ITU), features talks from AI leaders and 100+ demos.
Computer scientist Zeynep Akata has developed innovative methods to combine visual, linguistic, and conceptual elements in AI to increase user trust. Her research on explainable AI aims to make image classification decisions more transparent.
Researchers developed an AI tool called AAnet to characterize cancer cell diversity, identifying five distinct cell groups with different gene expression profiles. This could lead to more targeted therapies and improved patient outcomes.
An AI model developed by Ehsan Ghane at the University of Gothenburg can predict the durability and strength of woven composite materials, reducing development time. The model integrates material laws to make extrapolations outside training data, enabling better understanding of material behavior.
Researchers found that brain's dopamine neurons encode a map of possible future rewards across time and magnitude, guiding adaptive behavior in uncertain environments. This biological insight aligns with recent advances in AI, particularly distributional RL algorithms, which learn from reward distributions rather than averages.
Dr. Deanna Kaplan's innovative voice-capture app, Fabla, captures unstructured voice narratives to study how clinical interventions influence daily life. The platform has found applications across diverse health domains, including veteran experiences and healthcare provider burnout.
A new study by UChicago scientists found that AI-powered weather prediction models are remarkable but not magical, struggling to predict unprecedented weather events. The model can achieve impressive accuracy for short-term forecasts but fails to extrapolate beyond existing training data, leading to false negatives and potential mispre...
A team of researchers developed a machine learning model called Aurora that accurately forecasts various Earth systems, including air quality and tropical cyclone tracks. The model outperforms traditional systems at a fraction of the cost, enabling better preparedness for extreme weather events.
Researchers found that artificial intelligence tools can accurately predict disease for patients with typical symptoms but struggle with those exhibiting atypical symptoms. Human oversight is necessary for high-quality patient-centered care when using AI as an assistive tool.
Training neural networks on simple cognitive tasks improves their ability to handle more complex ones. By adopting the principles of early childhood education, researchers found that recurrent neural networks can be trained faster and with better results.
Researchers at Duke University have developed a new framework called HUMAC that enables robots to collaborate like humans by teaching them Theory of Mind. After just 40 minutes of guidance, robot teams exhibited strong collaborative behaviors and achieved high success rates in simulations and physical tests.
Researchers discovered similarities between AI and human brains with aphasia, offering new insights into diagnosis and improving AI's fluency. The study suggests that understanding internal patterns in AI models may lead to smarter and more trustworthy AI.
Researchers propose Input-Driven Plasticity model, which integrates past and new information to guide memory retrieval. The model is robust to noise and uses it as a means to filter out less stable memories.
A University at Buffalo-led study proposes using AI-powered handwriting analysis to identify spelling issues, poor letter formation, and other indicators of dyslexia and dysgraphia. The work aims to augment current screening tools and provide an early detection tool for these neurodevelopmental disorders.
A study suggests that groups of artificial intelligence language models can self-organise into societies, reaching consensus on linguistic norms, and are prone to tipping points in social convention. Collective biases emerge between agents through interactions, a blind spot in most current AI safety work.
Researchers have designed a headphone system that translates several speakers simultaneously, preserving voice direction and qualities. The Spatial Speech Translation system uses off-the-shelf noise-cancelling headphones fitted with microphones to separate out different speakers in a space and translate their speech.
The Global Confidence Degree-based Graph Neural Network (GCD-GNN) framework improves financial fraud detection by integrating global confidence metrics with advanced graph learning techniques. It achieves record-breaking accuracy on real-world datasets, including a 97.26% AUC on T-Finance.
A new review advocates for building confidence in AI applications by implementing robust data governance frameworks, enhancing transparency, and involving stakeholders. The authors emphasize the importance of addressing ethical implications and ensuring equitable access to AI-driven innovations in clinical oncology.
Researchers at the University of California San Diego have made a groundbreaking discovery about how our brains learn new information. Using sophisticated imaging techniques, they found that individual neurons follow multiple rules during learning, rather than one set of uniform rules as previously thought. This new understanding has s...
Researchers have created a breakthrough photonic chip that can train nonlinear neural networks using light, accelerating AI training while reducing energy use. The chip uses a special semiconductor material to reshape how light behaves, enabling reconfigurable systems with wide mathematical function expression.
Scientists have developed an all-optical activation function based on sound waves for photonic computing, enabling the creation of energy-efficient artificial intelligence systems. This breakthrough could potentially facilitate the scaling up of physical computing systems and pave the way for more efficient optical neural networks.
Scientists have built a digital twin of the mouse brain's visual cortex using AI, predicting neural activity and anatomical features. The model can generalize to new visual inputs and data, speeding up brain research and understanding intelligence.
The conference gathered international researchers to discuss AI's role in drug discovery and development, including generative AI strategies for designing chemical compounds. The speakers emphasized the significance of personalized medicine, where therapies will be tailored to each patient's unique molecular profile.
Artificial neural networks trained on spontaneous retinal activity patterns show improved motion prediction in natural scenes. The approach also enhances performance when combined with naturalistic movie data.
Researchers developed new AI models, InstaNovo and InstaNovo+, to vastly improve accuracy and discovery in protein science. These models excel in tasks such as de novo peptide sequencing, identifying microorganisms, and discovering novel peptides, with implications for personalized medicine, cancer immunology, and beyond.
Göttingen research team develops infomorphic neurons that learn independently and self-organize among neighboring neurons. This allows the smallest unit in the network to control its own learning, enabling novel machine learning approaches and a deeper understanding of brain function.
Researchers at National University of Singapore invent new computing cell that can mimic electronic neurons and synapses, reducing size by a factor of 18 and energy consumption. The discovery enables AI systems to process more information while using less energy.
A new brain-like computer uses analog computing to process and store information in the same location as biological neurons, reducing power consumption by 0.25%. The device, called a memristor network, is more efficient than conventional transistor-based computers and has implications for autonomous vehicles and drones.
Researchers developed an AI model that classifies variable stars from light curves with high accuracy, outperforming traditional approaches. The StarWhisper LightCurve series achieves near 90% accuracy with minimal manual intervention, paving the way for parallel data analysis and multi-modal AI applications in astronomy.
The collaboration aims to accelerate the development and commercialization of inait's innovative AI technology, using its unique digital brain AI platform. It will focus on joint product development, go-to-market strategies, and co-selling initiatives, initially targeting the finance and robotics sectors.
Researchers at Saarland University are developing leaner, customized AI models and techniques like knowledge distillation to reduce energy consumption. These smaller models enable small and medium-sized businesses to access powerful AI technology without a large technical infrastructure.
A new AI tool, NicheCompass, visualizes a cell's social network to help treat cancer. By analyzing millions of cells from patient samples, the tool predicts molecular changes and identifies potential targets for personalized treatments.
Researchers at Technical University of Munich developed a new AI training method that significantly reduces energy consumption. The approach uses probabilities to determine parameters, making the training process 100 times faster while maintaining accuracy comparable to existing procedures.