Articles tagged with Natural Language Processing
Researchers found that AI chatbots give users a uniform range of information, similar to a conventional web search, but with a narrower scope. The study's authors warn of 'knowledge collapse' as language models become increasingly trained on AI-generated text, potentially reducing diversity and nuance.
A new memory technology tames heat at the nanoscale, enabling rapid switching and reducing energy consumption. By stacking alternating layers of conductive and insulating materials, the researchers achieve a 76% reduction in reset energy demand and a 30-fold decrease in data drift.
Researchers found that hedge funds that effectively bet against public sentiment outperformed those that rode sentiment, earning a 0.4% monthly premium. This result persists even after controlling for fund characteristics and economic risks.
UniSpec delivers lossless LLM acceleration without retraining while adapting automatically to different hardware platforms and multilingual workloads. The framework achieves up to 2.6× faster inference than existing methods across multiple models, hardware, and languages.
A study by Bar-Ilan University researchers found that learning is driven primarily by changes in the strength of existing neural connections. The models became significantly better at learning as the amount of training data increased, but the proportion of lost connections remained roughly the same.
The review explores how integrating Federated Learning (FL), Reinforcement Learning (RL), and Natural Language Processing (NLP) can overcome modern NLP system limitations, such as protecting user privacy and adapting to changing environments. The study presents a unified framework that combines FL, RL, and NLP as three co-equal pillars.
A new study from UT San Antonio finds that emergency call takers' word choice can predict depression and anxiety symptoms, while positive expressions are often suppressed in high-stress professions. The research uses natural language processing to evaluate the psychological well-being of a critical but under-researched workforce.
A study found that forecast error types influence public emotion during disasters, with anxiety and worry being the most common emotions. The researchers suggest that communicating forecast uncertainty effectively could improve public trust and reduce emotional distress during future extreme weather events.
Researchers from Drexel University developed BioCoach, a program using AI and computer vision to analyze video and provide form coaching in real time. The system analyzes visual appearance and motion patterns, as well as 3D skeletal movements and body shape, to deliver detailed biomechanics-based feedback.
A new study maps how 29 factors influence one another in young adults aged 18-40, revealing self-reinforcing loops that worsen depressive symptoms and disrupt sleep. The model highlights the complexity of these interactions and suggests a nuanced approach to breaking the cycles.
Researchers analyzed a large dataset of Japanese Instagram ads to identify linguistic patterns linked to click-through rates. The study found that word choices for supplement and cosmetic ads differed significantly, with risk-related words boosting CTR in supplement ads, while motion-related language increased CTR in cosmetic ads.
A new study from Sultan Qaboos University demonstrates how AI can analyze Oman's Labour Law of 2023, revealing complex interdependencies between its articles. The research identifies influential 'hubs' within the law that may be triggered by changes, enabling policymakers to anticipate broader impact.
AI language models can extract and integrate information from vast amounts of unstructured environmental data, identifying pollutants and their toxic effects. While still in its early stages, the application of LLMs in aquatic risk assessment has the potential to support more dynamic and data-driven risk management strategies.
A new study analyzed over 460,000 scientific abstracts to identify key themes, trends, and research gaps in aging research. The study found a growing separation between basic biological studies and clinical research, highlighting the need for integration to translate laboratory discoveries into medical applications.
Researchers have launched a new multimodal medical dataset, Observer, capturing anonymized, real-time interactions between patients and clinicians. The dataset links video, audio, transcripts, and electronic health records to study subtleties like body language and environmental factors affecting care.
A study published by Michigan Medicine suggests that AI can help increase the chance of people with risky drinking patterns or signs of alcohol use disorder getting outreach and help. The NLP tool analyzed full text of health records and identified over 47,500 patients with some sign of risky drinking.
Researchers developed CellWhisperer, an AI method and software tool that links gene expression with descriptive text across millions of biological samples. It provides a virtual AI-based colleague to support biologists in their research, making biomedical data exploration easier and more exciting.
The study analyzed eye movement data to predict processing effort and found that word and sentence length are effective predictors of text readability among English learners. This approach provides valuable insights for selecting reading materials and developing reading tests.
Researchers at the University of Missouri have developed an AI-powered method to detect hidden hardware trojans in chip designs, offering a 97% accurate solution. The approach leverages large language models to scan for suspicious code and provides explanations for detected threats.
Researchers analyzed eye-tracking data to identify key factors influencing word processing during English reading. Word length was found to be the most critical factor in determining skipped words and total reading time, while word frequency and predictability played a significant role in initial processing and rereading.
Bezzo's project enables embodied cognition capabilities in large language models, equipping robots with human-like understanding of human intent. The work has wide-ranging potential for assistive technologies, collaborative robots, and defense missions.
Researchers developed a platform called CRESt that incorporates insights from literature, chemical compositions, and imaging to optimize materials recipes. CRESt uses robotic equipment for high-throughput testing and large multimodal models to further optimize materials recipes.
Hollings researchers use natural language processing to identify primary cancer types in medical notes, improving treatment accuracy and personalized care. The AI model achieves high accuracy rates, outperforming standard medical codes in identifying primary cancer diagnoses.
Researchers developed CoSyn, a new approach to train open-source models using AI-generated scientific figures and charts. The resulting dataset, CoSyn-400K, includes over 400,000 synthetic images and 2.7 million sets of corresponding instructions. CoSyn-trained models match or outperform proprietary peers in various benchmark tests.
Two Bar-Ilan University researchers have won ERC grants to advance cancer treatments using targeted protein degradation and improve regulatory effectiveness through machine learning. The projects aim to pioneer new therapeutics and design more effective interventions based on empirical evidence.
A new study demonstrates that retrieval-augmented generation can eliminate hallucinations in clinical LLMs while protecting patient privacy. The RAG-enhanced model responds significantly faster than cloud-based systems, delivering safer results without compromising data security.
Paulina DeVito, a Ph.D. candidate in computer science at Florida Atlantic University, has been awarded the NSF Graduate Research Fellowship to pursue research on developing large language model-based approaches for analyzing public discourse on social media. Her work aims to understand how people discuss emerging technologies and infor...
A team of researchers combined artificial intelligence and statistical modeling to analyze language patterns in three major sections of the Bible. They distinguished between three distinct scribal traditions spanning the first nine books of the Hebrew Bible, known as the Enneateuch. The model also determined the most likely authorship ...
Researchers used large language models to predict conversational derailment, achieving comparable accuracy to deep learning models. This approach supports healthier online communities at reduced cost by leveraging general-purpose LLMs.
University of Missouri researchers create digital sentiment map using AI to analyze public Instagram posts, linking emotional tone to real-life features. The tool aims to improve city services, identify areas of concern, and inform emergency response decisions.
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.
Capsule eliminates I/O overhead and optimizes performance by designing a subgraph loading mechanism and pipelined parallel strategy for large-scale graph neural network training, providing 22.24% memory usage and up to 12.02x performance improvement over existing systems
The Deep Computational Text Analyser (DECOTA) is an open-access AI tool that transforms open-ended survey responses into clear themes in minutes, not months. Developed by the University of Bath, DECOTA delivers insights around 380 times faster and over 1,900 times cheaper than human analysis.
Researchers used a large language model to analyze clinical reports of autism patients and found that repetitive behaviors and special interests are most indicative of an autism diagnosis. The study aims to improve diagnostic guidelines by reducing the focus on social factors.
Researchers leveraged a large language model to classify artwork types without pre-prepared training data, achieving high accuracy. This approach enables performance comparable to conventional machine learning methods while reducing human effort and time required for data organization.
A new AI-based tool can translate a person's thoughts into continuous text without requiring language comprehension, and it can be trained in under an hour. The system was developed by adapting a previous brain decoder to a new person using short, silent videos.
A new study assesses the historical knowledge of AI chatbots like ChatGPT-4 and finds they struggle with nuanced, PhD-level inquiry. The models performed best on legal systems and social complexity but struggled with topics such as discrimination and social mobility.
Researchers developed a new method, k* distribution method, to visualize and assess how well deep neural networks categorize related items together. The model reveals clustered, fractured or overlapping arrangements of data points, indicating accuracy and reliability issues.
The HumanTech Summit 2024 conference brings together experts in technological and social innovation, discussing key topics such as data security and interpersonal relationships. With a proven legacy of impact, the event has attracted over 1,400 participants and featured renowned keynote speakers from top universities.
A novel no-code prototype, Auto-DSM, generates design structure matrices (DSMs) using large language models. It improves productivity and accuracy compared to traditional methods, with speeds of up to four minutes per DSM generation.
A new study by UCLA Health reveals that standard medical record surveillance methods miss youth with suicidal thoughts and behaviors in children, boys, and Black and Hispanic youths. Machine learning algorithms improved detection rates when incorporating additional data from visit notes.
Researchers analyzed Twitter posts and Google searches from 2016-2020 to identify seasonal allergy patterns across the US. They found a strong national pulse of allergy symptoms in March-May, with significant regional variations.
A new training algorithm called ternarized gradient BNN (TGBNN) enables learning capabilities for binarized neural networks (BNNs) on IoT edge devices. The proposed MRAM-based CiM architecture achieves faster convergence and matching accuracy with regular BNNs.
A new study by Zhejiang University highlights the disproportionate health challenges faced by sexual and gender-diverse individuals during the COVID-19 pandemic. SGD individuals experienced higher rates of COVID-19 symptoms and mental health issues compared to non-SGD users, according to a large-scale social media analysis.
WorldScribe, a new software, uses generative AI to provide real-time text and audio descriptions of surroundings for people who are blind or have low vision. The tool can adjust the level of detail based on user commands or camera frame time.
Researchers aim to advance training methodologies for superintelligent systems, ensuring they learn from imperfect and evolving data. The project seeks to develop new techniques to improve AI reliability, reduce biases, and increase accuracy.
Researchers found that large language models (LLMs) can improve systems that detect bots by outperforming state-of-the-art systems by 9%. However, LLMs can also reduce the performance of existing detectors by up to 30% when used to evade automated detection.
Researchers at EPFL developed a next-generation miniaturized brain-machine interface capable of direct brain-to-text communication on tiny silicon chips. The MiBMI system can decode neural signals generated when a person imagines writing letters or words with high accuracy and low power consumption.
A recent study found that 20% of patients with chronic cough received an opioid prescription, with older patients and those with Medicaid insurance more likely to be prescribed these drugs. The research team hopes to explore alternative treatment options to reduce reliance on opioids for chronic cough care.
A study of Reddit comments on ChatGPT's launch reveals the tech community is more strongly divided in their opinions about AI. While non-tech discussions focus on social issues like job replacement, tech-centric subreddits have more focused and opinionated debates.
A recent study analyzing 300,000 X posts found that 48 US states have a more positive than negative tone towards nuclear energy, with a national average at 54% positive. Concerns about waste, cost, and safety dominate negative sentiment, while technology themes fuel positive sentiments highlighting innovations and job creation.
MOSS, an open-sourced conversational large language model, demonstrates unprecedented capabilities through cross-lingual pre-training, preference-aware training, and tool augmentation. The model learns general concepts and handles diverse user intents, making it a versatile AI assistant.
Researchers found low rates of metastasis and lymph node involvement in post-menopausal patients with early-stage ER+ breast cancer. This suggests that clinical guidelines for de-escalating surgery may be safely extended to younger patients.
A new AI method developed by Swedish researchers can identify toxic substances based on their chemical structure, potentially replacing animal testing. The method has been shown to be more accurate and broadly applicable than existing computational tools, offering a promising alternative for environmental research and authorities.
The study reveals that experienced programmers exhibit brain responses similar to those of fluent readers processing sentences, indicating a strong resemblance between learning computer and natural languages. The research suggests that understanding how people learn to code can be informed by expertise in second-language learning.
Researchers used an AI-assisted application to help people write cartoon captions for The New Yorker Cartoon Caption Contest. The tool analyzed incongruity and generated suggestions, resulting in jokes rated 30% funnier than those written without assistance.
Researchers at WVU are developing an AI tool to reduce medication errors that lead to hospital readmissions, aiming to improve patient safety and reduce healthcare costs. The tool will analyze patient records and identify high-risk patients, alerting pharmacists to potential issues.
The JGU Center for Lifelong Learning is developing personalized AI-based learning experiences for adult learners, aiming to improve motivation and learning outcomes. The project will also discuss the challenges of AI technologies in teaching modern foreign languages.
Researchers developed a predictive model to detect users and content related to Islamic State extremists on social media, identifying potential propaganda messages and their characteristics. The study's findings can help social media companies and law enforcement agencies track and prevent the spread of extremist propaganda.
Research by Clotilde Napp found that gender biases in language are stronger in more economically developed and individualistic countries. The study used natural language processing to analyze text corpora from over 70 countries, revealing a phenomenon known as the gender equality paradox.