Articles tagged with Artificial Intelligence
Researchers developed OptoGPT, an algorithm that designs optical multilayer film structures for various applications. It produces designs in 0.1 seconds and contains six fewer layers on average compared to previous models.
Researchers analyzed reflections of light on human eyeballs in real and AI-generated images, finding that fake images often lack consistency between left and right eye reflections. The study employed methods typically used in astronomy to quantify the reflections and checked for consistency.
A new system for evaluating evidence in craniofacial superimposition uses likelihood ratios to provide quantitative support for forensic experts. The proposed method excels in calibration and discriminating power, offering a tool for integrating and evaluating evidence objectively.
Researchers have developed PrISMa, a novel platform that seamlessly connects materials science, process design, techno-economics, and life-cycle assessment to identify effective and sustainable carbon capture solutions. The platform has been tested on over 60 real-world case studies, providing valuable insights for stakeholders.
Large language models (LMs) exhibit human-like reasoning flaws, particularly in tasks requiring semantic content and believability. LMs struggle to distinguish between valid and invalid arguments, mirroring human errors, especially when semantic content is sensical and believable.
A new study is testing whether digital tools can help predict which patients with depression will benefit from specific treatments and deliver them on demand. The researchers hope to use these tools to broaden access to psychiatric care and realize the promise of precision psychiatry.
Scientists have developed a new technique that leverages X-ray photon correlation spectroscopy, artificial intelligence, and machine learning to create unique 'fingerprints' of materials. These fingerprints can be analyzed by neural networks to yield new information about material behavior under stress and relaxation.
Researchers developed a method to assess the reliability of foundation models, enabling users to choose the best model for their task without testing it on real-world data. The approach measures consensus among multiple models and aligns representations to compare consistency.
Researchers caution against relying solely on LLMs for human participants due to cross-cultural variation and limited variability in responses. While LLMs can analyze text, they struggle with implicit, emotional, or context-dependent tasks, and their black-box nature hinders reproducibility.
A study found that over 60% of participants preferred AI decision-making over humans in redistributing earnings, despite potential discrimination concerns. However, participants were less satisfied with the fairness and accuracy of AI decisions compared to human-made choices.
A new study by NYU Langone Health shows that an AI tool can draft responses to patients' EHR queries as accurately as human healthcare professionals, with greater perceived empathy. The tool has the potential to dramatically reduce physicians' In Basket burden while improving their communication with patients.
A study published in Nature Medicine found that off-the-shelf wearable trackers can monitor the response to two treatments for atrial fibrillation and heart failure. The devices provided clinically useful information similar to in-person hospital assessments, with a neural network helping to analyze missing data.
A new study at Hebrew University examined the role of AI in mental health therapy, focusing on empathy. The researchers propose a hybrid model where AI supports therapeutic processes without replacing human therapists.
A new camera system called PrivacyLens can replace people in images with generic stick figures, protecting their identities and reducing unnecessary surveillance. This technology could prevent embarrassing photos from being shared online and make patients more comfortable using cameras for chronic health monitoring.
A new study found that AI enhances creativity by boosting novelty and usefulness of stories, making them more enjoyable, well-written, and engaging. However, this comes at the expense of less varied content, as AI-assisted stories contain more similarities to each other.
Researchers developed an AI model that accurately predicts Alzheimer's disease progression in 82% of cases, outperforming clinical tests. The model uses cognitive tests and MRI scans to identify individuals at risk, allowing for targeted treatment and closer monitoring.
UCF's STRONG-AI initiative aims to uplift bright, low-income undergraduate students in pursuing well-rounded AI education through faculty and peer mentorship and scholarship. The program has received over 150 applications and will select 10-15 students annually based on financial aid eligibility and academic success.
Researchers developed an AI model to analyze heart MRI scans, which can save NHS time and resources. The model provides a complete analysis of the entire heart using a view that shows all four chambers, leading to faster and more accurate diagnosis of heart failure and other cardiac conditions.
The WISDOM project aims to integrate healthcare and research data to combat CIMDs through AI-driven models. The initiative hopes to improve disease diagnosis, treatment, and patient outcomes by sharing data securely across borders.
Scientists developed customized AI tools, including ChatGPT, to provide accurate responses on digital pathology. The tools help pathologists without extensive coding experience analyze tissue samples, bridging the gap between pathology and digital pathology skills.
The ZEN-MRI project, funded by the German Ministry of Education and Research, aims to achieve a congenial co-existence between human beings and robots in public areas. Researchers will investigate how to design robots that are perceived as unthreatening and sympathetic.
Children and young people are generally positive about AI's role in healthcare, with many believing it can improve their care and outcomes. However, they also emphasize the importance of human oversight, particularly when it comes to empathy and ethical decision-making.
This review explores the progress of AIGC visual content generation and traceability, highlighting advancements in image quality, controllable image generation, and watermark-related technologies. Watermarking techniques are categorized into different methods to authenticate and verify AI-generated images.
Researchers at Kaunas University of Technology developed an end-to-end system to eliminate visual distractions caused by overlapping camera angles in live football broadcasts. The system uses YOLOv8 object detection and video inpainting technology to remove unwanted objects, such as cameramen, from frames.
Researchers at the University of Cambridge have developed an atlas of proteins that reveals how they behave inside human cells. The tool allows for the identification of new proteins involved in important bodily functions, including fat distribution and protein creation.
GenSQL integrates a tabular dataset and a generative probabilistic AI model to analyze complex tabular data. It can detect anomalies, predict outcomes, and generate synthetic data with just a few keystrokes.
Researchers sequenced DNA of 87 structurally colored bacteria to identify genes responsible for their glittering structures. The findings could lead to environmentally friendly dyes and materials.
Researchers from SLAC, Stanford and other institutions have developed a technique to improve time resolution for the MeV-UED electron camera and trained an artificial intelligence model to tune the beam for various experimental needs. This enables unprecedented precision in exploring novel effects in materials and chemistry.
Researchers developed an AI model that can estimate lung function from chest radiographs with high accuracy, potentially expanding options for pulmonary function assessment in patients who have difficulty performing spirometry. The study found a remarkably high agreement rate between the AI model's estimates and actual spirometric data.
A recent study by Cornell University found that employees who were monitored by AI reported feeling less autonomy and had worse performance. In contrast, those who were monitored by humans perceived the monitoring as more supportive of their development.
Scientists have developed a machine learning program that can identify blobs of plasma in outer space known as plasmoids. The program will analyze data from NASA's Magnetospheric Multiscale (MMS) mission to better understand magnetic reconnection and its effects on the electrical grid.
A study published in PLOS ONE found that nearly 70% of participants rated ChatGPT-generated jokes as funnier than those created by people. The researchers also tested the AI against professional humor writers, with results showing that ChatGPT's headlines were just as funny as those from The Onion.
Researchers develop a method that fuses AlphaFold's strengths with computer simulations based on physics laws to predict protein structures, enabling faster drug development. The approach filters down initial hypotheses to a more manageable set of structures, increasing the effectiveness of pharmaceuticals.
A new technique, METAPHOR, visualizes embryo metabolism to predict implantation success and full-term birth, increasing the probability of success in assisted reproduction processes.
The article reviews recent advances in applying artificial intelligence to oncology, showcasing promising improvements in cancer care. The authors emphasize the need for interdisciplinary collaboration, rigorous validation, and ethical principles to harness AI's potential.
Researchers found that AI models that analyze medical images can predict patient demographics with high accuracy but struggle to diagnose patients from diverse backgrounds. The models use demographic shortcuts, leading to incorrect results for women, Black people, and other groups.
A newly developed deep learning algorithm has outperformed existing methods in predicting osteoporosis risk, potentially leading to earlier diagnoses and better outcomes. The model identified key factors such as weight, age, and grip strength as significant contributors to osteoporosis risk.
Researchers developed an AI model that accurately predicts metal yield strength by combining physical theory with machine learning. The model outperforms traditional methods, which often rely on extensive experimentation.
Researchers at U of T have developed a deep-learning model called PepFlow that can predict the full range of conformations for peptides, which are shorter than proteins but perform similar biological functions. The model combines machine learning and physics to capture precise and accurate conformations within minutes.
Researchers found that people are more likely to accuse others of lying when AI makes an accusation, highlighting the need for caution in implementing similar technologies. The study suggests that over-reliance on AI lie-detection tools may lead to careless accusations and reinforce biases.
A study found that large language models (LLMs) like ChatGPT underperform state-of-the-art detectors but can explain their analysis in plain language. LLMs' semantic knowledge makes them well-suited for detecting deepfakes, providing a common sense understanding of reality.
A new study from the University of Copenhagen reveals that AI has helped detect significantly more cases of breast cancer and reduce radiologist workloads. The AI system has been used to analyze tens of thousands of X-rays every year, resulting in a 12% increase in detected cases, including more small tumours of one centimetre or less.
AlphaFold's groundbreaking ability to predict protein structures is set to transform predictive medicine, enabling the development of personalized vaccines and adaptive clinical trials. However, the review also highlights crucial challenges and ethical considerations surrounding AI integration with clinical data.
The special issue explores challenges and opportunities in managing synthetic genomics risks, introducing a common global baseline for nucleic acid synthesis screening. Review articles provide insights into enhancing gene synthesis security and biosecurity practices of synthetic DNA providers.
Researchers found that AI-generated submissions went undetected in a UK university's examinations system, earning higher grades than real students' answers. The study suggests that AI tools could potentially allow students to cheat and achieve better grades than their peers.
A recent study published in PLOS ONE found that AI-generated exam answers were undetectable in 94% of cases, outperforming human submissions. The University of Reading's blind test challenged educators to detect AI-generated content, highlighting the need for global education sector policies and guidance on AI use.
Researchers at Boston University developed an AI model that analyzes speech patterns to predict the likelihood of Alzheimer's disease in patients with mild cognitive impairment. The model achieved an accuracy rate of 78.5% and could potentially revolutionize dementia screening, making it more accessible and efficient.
A new AI model has been developed to improve clinical trial recruitment for eye disease, specifically Geographic Atrophy. The system identified almost twice as many candidates and with higher precision compared to conventional approaches, showing promise for overcoming a major obstacle in GA clinical trials.
Researchers investigated the efficiency of modern neural network-based generative models, comparing them to traditional sampling techniques. The study found that modern diffusion-based methods may face challenges due to a first-order phase transition, but also exhibit superior efficiency in certain cases.
Researchers found that humans distinguish between voices based on emotional cues, with happy voices being perceived as more human-like. However, the brain's response differs, with AI voices eliciting stronger responses in error detection and attention regulation areas.
A new study found that animals use a wide range of strategies to accomplish tasks, many of which are just as effective as the optimal solution but require less brain power. The research provides a theoretical framework for understanding these 'good enough' strategies and their potential applications in animal behavior.
A new model developed by Flatiron Institute researchers proposes that individual neurons exert more control over their surroundings, which could be replicated in artificial neural networks. This updated model treats neurons as tiny 'controllers' and may lead to better AI performance and efficiency.
A new study developed an AI-based approach, DiffPALM, to predict protein interactions with high accuracy, outperforming traditional methods. This advancement has significant implications for drug development and disease treatment, and the researchers have made it freely available for further research.
Researchers found ChatGPT consistently ranks resumes with disability-related honors and credentials lower than the same resumes without those honors. However, when trained to prioritize disability justice and DEI principles, the system improved its rankings for most disabilities, except for autism and depression.
Scientists at UVA and Toyota Research Institute create language representations of driving behavior to enable robots to associate words with environmental interactions. This allows cars to provide guidance and adjust speed in challenging situations, improving safety and usability.
TopoFormer uses a transformer model trained on tens of thousands of protein-drug interactions to recreate 3D structures as one-dimensional information that current models can understand. This enables more accurate predictions of new drug interactions and potentially reduces development time and costs.
A new system named SQUID, a computational tool created by Cold Spring Harbor Laboratory scientists, helps interpret how AI models analyze the genome. It reduces background noise and leads to more accurate predictions about genetic mutations.
Research suggests that large-language models could play a role in managing the energy grid, particularly in emergency response, crew assignments, and wildfire preparedness. However, significant challenges remain, including data availability, safety guardrails, and reliability, which must be addressed to ensure safe deployment.
Researchers at Cold Spring Harbor Laboratory designed a new way for AI algorithms to move and process data more efficiently, inspired by the human brain. This design allows individual AI neurons to receive feedback and adjust on the fly, processing data in real-time.
Deep learning enhances STS diagnosis and treatment by automating contouring for radiation therapy and predicting treatment responses. The review highlights key areas where deep learning is making an impact, including data acquisition and processing, algorithm development, clinical applications, and pathological diagnosis.