Figures Abstract Graph Neural Network (GNN) faces limitations in few-shot image classification due to insufficient adaptive feature extraction and limited long-range dependency modeling. To address these challenges, this study proposes an Improved Graph Neural Network (IGNN) integrating two key innovations. Firstly, we design an Attention-Enhanced Feature Extraction module, which combines Efficient Channel Attention (ECA) and self-attention mechanisms, enabling the model to dynamically focus on discriminative intra-image details and inter-image contextual relationships, thereby improving feature representation robustness. Secondly, we introduce a gated recurrent unit (GRU)-based Pre-message-passing mechanism, which establishes cross-sample associations between support and query sets before message propagation, effectively capturing long-range dependencies and mitigating information smoothing. The experimental results of three public datasets demonstrate that our proposed framework outperforms the existing methods and shows significant potential. It offers a pragmatic tool for applications requiring rapid adaptation to limited data, such as remote sensing and medical image analysis. Citation: Chen J, Fu B, Zou L (2026) IGNN: An improved graph neufral network with integrated attention and pre-message-passing for few-shot image classification. PLoS One 21(4): e0348057. https://doi.org/10.1371/journal.pone.0348057 Editor: Nagaraju Y, Dayananda Sagar College of Engineering, INDIA Received: June 14, 2025; Accepted: April 5, 2026; Published: April 28, 2026 Copyright: © 2026 Chen et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability: The authors used the publicly available datasets Omniglot dataset, MiniImageNet dataset, and CUB-200-2011 for the experiments. The Omniglot dataset can be accessed at https://github.com/brendenlake/omniglot. The MiniImageNet dataset can be accessed at https://image-net.org/update-mar-11-2021.php. The CUB-200-2011 dataset can be accessed at https://www.vision.caltech.edu/datasets/cub_200_2011/. Funding: This research was funded by the Sichuan Science and Technology Program, grant number 2025YFHZ0007, 2024JDHJ0015 and the Fundamental Research Funds for
Apr 28, 2026 · via journals.plos.org
Abstract Chronological age predicts cancer survival but does not capture differences in biological aging rates. We apply FaceAge, an artificial intelligence algorithm that predicts biological age from a facial photograph, to serial clinical facial photographs to calculate the Face Aging Rate (FAR; change in FaceAge divided by the time between photographs). We analyze data from 2276 cancer patients receiving radiation therapy, using photographs captured during routine care. Higher FAR is associated with worse overall survival in stratified analyses of cohorts with the following intervals between photographs: short 10-365 days (adjusted hazard ratio [aHR] and 95% confidence interval: 1.25 [1.03-1.51]), mid 366–730 days (aHR: 1.37 [1.00-1.86]), and long 731-1,460 days (aHR: 1.65 [1.22-2.22]) after adjustment for time between photographs, sex, race, and diagnosis. FAR provides additional prognostic information beyond single time-point measures of FaceAge. FAR is a non-invasive prognostic biomarker that captures dynamic changes in biological aging. Similar content being viewed by others Introduction Aging is a multifaceted biological process characterized by the decline of physiological functions and increased vulnerability to disease and death1. It affects nearly all living organisms and is intricately linked with various diseases, particularly cancer, as both result from the accumulation of cellular damage over time2,3,4,5. Chronological age has long been recognized as a predictor of survival; this relationship is particularly evident in cancer patients6,7,8,9,10,11,12. However, it treats all individuals within an age group identically, disregarding variations in biological aging rates13. Recognizing these limitations, it becomes essential to explore biological age indicators that can be readily implemented into clinical practice, especially those that quantify the rate of aging and are easily accessible, for personalized risk assessment. Recent longitudinal evidence demonstrates that significant variation in biological aging trajectories can already be quantified in young adulthood, underscoring the potential for early interventions before disease manifestation14. This study aims to
Apr 28, 2026 · via nature.com
Fans erupt as Disneyland rolls out facial recognition technology across park entrances The Happiest Place on Earth is rolling out facial recognition technology to expedite park entrance and catch out fraudsters. Guests “may choose to use entrance lanes equipped with facial recognition technology” at both Disneyland Park and Disney California Adventure, where cameras capture an image and convert it into “unique numerical values” to verify identity, according to the company. The system compares those values for a match and, in most cases, deletes them within 30 days. Disneyland officials say the optional technology is designed to streamline entry while preventing fraud. The entertainment giant says participation is voluntary, noting that traditional entry lanes without biometric scanning remain available. In those lines, cast members manually verify tickets instead of relying on facial recognition. The company also emphasized that “the security, integrity and confidentiality” of guest data is a priority, though it acknowledged that “no security measures are perfect or impenetrable.” At the parks, reactions from visitors were mixed. “Pretty much every other place is doing the same thing, John LeSchofs, 73, a frequent parkgoer who comes every six weeks, told the Los Angeles Times. “The police, the government, they’re all using facial recognition. I don’t think it’s going to stop.” Others expressed hesitation, particularly around transparency and consent. Robert Howell, 30, visiting from Virginia, said he was unaware of the technology until arriving at the park. “It’s a little scary because it’s not clear how it’s going to be used,” Howell said. “With TSA I know that’s an option that you can opt out, but I didn’t realize you could here so I just did it.” For some families, concerns center on how the technology affects children. Sandra Contreras said she felt uneasy when it came to her young daughter. “When
Apr 28, 2026 · via nypost.com
FORM Earns Two Golds at 2026 American Business Awards, Further Proving Industry Leadership in Mobile Task Management Best Mobile Operations Management Solution and Best Use of Augmented Reality awards underscore AI and AR innovation that is delivering real results for retailers and CPG brands BOSTON, April 28, 2026 /PRNewswire/ -- FORM, a leading provider of mobile task management and retail execution solutions, today announced it won two Gold Stevies® in the 2026 American Business Awards® for its GoSpotCheck image recognition platform. The awards recognize FORM's unparalleled technological leadership and the company's continued success in delivering measurable results on behalf of retail, grocery, distributor and CPG companies everywhere. FORM won Gold Stevies for its Mobile Operations Business Management Solution, GoSpotCheck, and the solution's augmented reality capabilities that help field teams improve on-shelf availability, ensure planogram compliance, and close the gap between corporate strategy and in-store execution. The awards come on the heels of another major recognition FORM received in March when GoSpotCheck earned a 2026 Artificial Intelligence Excellence Award from The Business Intelligence Group, one of the world's foremost independent AI recognition programs. "It's not only our customers noticing our innovation; the broader industry is paying attention to FORM's market dominance and technological leadership," said Matt Collins, Chief Marketing Officer, FORM. "These awards are gratifying because they prove the mettle of this overall team, which is working tirelessly to deliver real value to customers. And we're only getting better with our recent merger, which is leading to a bigger scale and reach than we've ever had." The American Business Awards are the U.S.A's premier business awards program. All organizations operating in the U.S.A. are eligible to submit nominations – public and private, for-profit and non-profit, large and small. More than 3,600 nominations from organizations of all sizes and in virtually every
Apr 28, 2026 · via prnewswire.com
Belt scale provides performance across applications Tecweigh’s HDS (heavy-duty S-type load cell) belt scale system is designed for high-capacity, high-impact applications. Tecweigh says the scale frame is built to withstand harsh operating conditions, while its load cell design ensures consistent readings across a range of belt speeds and material flows. The HDS is available in configurations ranging from one to four idlers and comes standard with onboard static calibration weights that can be raised and lowered during conveyor operation. The load cell can be installed on new conveyors or retrofitted onto existing systems. When paired with the Tecweigh WP50 processor, the company says operators gain real-time insights into throughput, totals and system performance. Scale tech allows for autonomous material recognition Wingfield Scale Co. launched iD Point image recognition technology for use on its WingScan Loadmaster. According to Wingfield Scale, iD Point uses advanced machine learning and AI to identify what material is being moved on trucks and railcars and attach a material code to each measurement. Users upload images of any material, then run a “learning” step so the system can recognize it in real time. Wingfield Scale says iD Point makes WingScan a fully autonomous, multidimensional measurement system. iD Point is currently in prerelease as a WingScan add-on feature.
Apr 28, 2026 · via pitandquarry.com
A woman strolls into a grocery store, thinking about grabbing some apples. Before she even reaches the produce aisle, a security camera has scanned her face. Whether the system is checking for shoplifters or simply logging her arrival, her face has joined a digital ledger, a trace she can’t easily erase. Retailers, banks, airports, stadiums and office buildings are doing the same. kAmqFE H92E :7 E96 H@>2?’D 724:2= :?7@C>2E:@? :D DE@=6? @C >:DFD65n x7 2 4J36C4C:>:?2= DE62=D 96C A2DDH@C5[ D96 42? 492?86 :E] x7 E96J 24BF:C6 96C 4C65:E 42C5 ?F>36C[ D96 42? 42?46= E96 42C5] qFE D96 42?’E C6D6E @C C6G@<6 E96 2AA62C2?46 @7 96C 4966<3@?6D]k^Am kAmu24:2= C64@8?:E:@? DJDE6>D 5@?’E <66A 24EF2= :>286D] %96J 4@?G6CE 2 7246 :?E@ 2 k2 9C67lQ9EEADi^^HHH]5:8:E2=D6?D6]2:^3=@8^9@H\5@6D\724:2=\C64@8?:E:@?\H@C<Qm>2E96>2E:42= E6>A=2E6k^2m E92E >2AD E96 A@D:E:@?D 2?5 AC@A@CE:@?D @7 E96 7246’D 762EFC6D] (96? 2?@E96C 42>6C2 D42?D 2 A6CD@? =2E6C[ E96 DJDE6> 4964<D E96:C =:G6 7246 282:?DE E96D6 E6>A=2E6D E@ 4@?7:C> 2? :56?E:EJ]k^Am kAmx? >J H@C< 2D 2 k2 9C67lQ9EEADi^^HHH]C:E]65F^5:C64E@CJ^;DH:4D\;@?2E92?\H6:DD>2?Qm4J36CD64FC:EJ AC@76DD@Ck^2m 2E #@496DE6C x?DE:EFE6 @7 %649?@=@8J[ x 92G6 7@F?5 E92E 6G6? E9@F89 E6>A=2E6D 2C6 >@C6 D64FC6 E92? A9@E@D – H9:49 2?J@?6 @?=:?6 42? 42AEFC6 2?5 >2?:AF=2E6 – E6>A=2E6D[ E@@[ 42? 36 DE@=6?] ~?46 E92E 92AA6?D[ E96D6 5:8:E2= <6JD 4C62E6 2 =:76=@?8 GF=?6C23:=:EJ] x7 2 724:2= C64@8?:E:@? 52E232D6 :D 3C624965[ E96 “=@4<D” E92E 2 E6>A=2E6 @A6?D – 2446DD:?8 2 32?< 2AA[ 86EE:?8 E9C@F89 D64FC:EJ 2E 2? 2:CA@CE[ 6?E6C:?8 2? @77:46 3F:=5:?8 – 42?’E 36 C6D6E] p A6CD@?’D 7246 :D A6C>2?6?E[ 2?5 D@ :D E96 E9C62E]k^Am kAm%96 E9C62E :D?’E E96@C6E:42=] q:@>6EC:4 52E2 92D 366? DE@=6? :? 52E2 3C62496D] x? a_ac[ 3:@>6EC:4 52E2 7C@> 2 724:2= C64@8?:E:@? DJDE6> FD65 2E 32CD 2?5 4=F3D :? pFDEC2=:2 k2 9C67lQ9EEADi^^HHH]H:C65]4@>^DE@CJ^@FE23@I\724:2=\C64@8?:E:@?\3C6249^QmH2D 924<65k^2m] p?5 :? a_`h[ 3:@>6EC:4 52E2 7C@> 2 A:=@E 724:2= C64@8?:E:@? DJDE6> D6E FA 3J &]$] rFDE@>D 2?5 q@C56C !C@E64E:@? k2 9C67lQ9EEADi^^HHH]@:8]59D]8@G^C6A@CED^a_a_^C6G:6H\43AD\>2;@C\4J36CD64FC:EJ\:?4:56?E\5FC:?8\a_`h\3:@>6EC:4\A:=@E^@:8\a_\f`\D6Aa_QmH2D 3C624965k^2m :? 2? 2EE24< @?
Apr 28, 2026 · via caledonianrecord.com
Japan's APPI amendment bill would open narrow lane for some AI uses, tighten rules elsewhere A bill to amend Japan's APPI would codify a new concept of "statistical processing," while putting sharper boundaries around minors' data, facial feature data, opt-out sharing and enforcement. Contributors: Takashi Nakazaki Partner Anderson Mori & Tomotsune Japan's Personal Information Protection Commission announced 7 April that the Cabinet of Japan approved a bill to amend the Act on the Protection of Personal Information. The PPC frames the package as accomplishing two things at once: facilitating data use, including artificial intelligence-related data use, while strengthening protection and enforcement where data handling creates greater risks for individuals. The bill, which does move in both directions simultaneously, is not a wholesale rewrite of Japan's privacy law, enacted in 2003. It is a targeted reform package, and a meaningful one. Unlike the EU General Data Protection Regulation, which generally begins by asking what legal ground supports data processing, Japan's amendment bill does not try to import a new across-the-board lawful-basis model into the APPI. Instead, it works more surgically, by changing specific consent-sensitive parts of the APPI and building new rules around higher risk practices. Under Japan's private-sector framework, the more important questions are usually about purpose limitation, use beyond the stated purpose and rules on third-party sharing and overseas transfers. So, for most privacy teams, the more useful question is not whether Japan is moving closer to Europe. It is which APPI rules would be loosened, and which would get tougher under the amendment bill. A new statutory concept, not a free pass The most significant change is the introduction of a new statutory concept, "statistical processing," defined as the creation of statistics and similar analytical acts that derive trend- or characteristic-level information from large volumes of information, excluding
Apr 28, 2026 · via iapp.org
(The Conversation is an independent and nonprofit source of news, analysis and commentary from academic experts.) (THE CONVERSATION) A woman strolls into a grocery store, thinking about grabbing some apples. Before she even reaches the produce aisle, a security camera has scanned her face. Whether the system is checking for shoplifters or simply logging her arrival, her face has joined a digital ledger, a trace she can’t easily erase. Retailers, banks, airports, stadiums and office buildings are doing the same. But what if the woman’s facial information is stolen or misused? If a cybercriminal steals her password, she can change it. If they acquire her credit card number, she can cancel the card. But she can’t reset or revoke the appearance of her cheekbones. Facial recognition systems don’t keep actual images. They convert a face into a mathematical template that maps the positions and proportions of the face’s features. When another camera scans a person later, the system checks their live face against these templates to confirm an identity. In my work as a cybersecurity professor at Rochester Institute of Technology, I have found that even though templates are more secure than photos – which anyone online can capture and manipulate – templates, too, can be stolen. Once that happens, these digital keys create a lifelong vulnerability. If a facial recognition database is breached, the “locks” that a template opens – accessing a bank app, getting through security at an airport, entering an office building – can’t be reset. A person’s face is permanent, and so is the threat. The threat isn’t theoretical. Biometric data has been stolen in data breaches. In 2024, biometric data from a facial recognition system used at bars and clubs in Australia was hacked. And in 2019, biometric data from a pilot facial recognition system
Apr 28, 2026 · via bozemandailychronicle.com
A recent Pentagon contract just revealed the smartest artificial intelligence strategy in business. While some builders and investors chase fully autonomous everything – self-driving cars, pilotless airplanes, human-less factories – the real winners are asking: How much autonomy should we actually give our AI systems? The answer is reshaping how companies deploy AI across every industry. It's not what the hype machine is selling. Beacon AI, a California-based aviation software company, recently signed a four-year contract with US Special Operations Command (USSOCOM) worth up to $49.5 million. The deal is not for a fully autonomous, pilotless aircraft or for sci-fi autonomous fighters. Instead, it's for AI-powered pilot assistance software that cuts cockpit workload and speeds mission-critical decisions in high-risk operations. Think R2-D2, not HAL 9000. The system integrates flight data, weather, routing, and pilot inputs into real-time decision support. This is exactly what human pilots need when operating in contested, time-sensitive environments. Here is the strategic insight. Beacon deliberately chose limited AI autonomy over full automation, and the Pentagon validated this choice with a large check. The autonomy levels every executive needs to understand Most boardrooms discuss AI as if it's binary. You either have it or you don't. This is simplistic. AI autonomy operates on a spectrum, and understanding these levels is critical for any leadership team deploying AI systems. The framework comes from autonomous vehicles, but applies to AI systems in any industry: - Level 0 (no automation): Humans do everything. Traditional tools with no AI assistance. - Level 1 (driver assistance): AI provides information and basic support. Think spelling-check or fraud alerts that humans must act upon. The AI has no decision-making power. - Level 2 (partial automation): AI can execute specific tasks while humans maintain oversight and control. Examples: AI scheduling systems that suggest meetings but
Apr 28, 2026 · via jpost.com
Pet Health and AI News from NYC's Top Digital Cybersecurity Master's Program Artificial Intelligence Biotechnology Computer Science Cybersecurity Data Analytics and Visualization Digital Marketing and Media Mathematics Nursing Occupational Therapy Physician Assistant Physics Speech-Language Pathology How AI Is Revolutionizing Pet HealthJust as healthcare is becoming increasingly personalized for humans, AI is bringing similar advances to pet care. By using machine learning and predictive analytics, AI can analyze a pet’s unique characteristics – including breed, age, lifestyle, and genetics – to help inform customized wellness plans. These expanding capabilities are showing strong potential to improve how animals are examined, diagnosed, and treated.In veterinary clinics, research labs, and pet tech startups, AI is creating new opportunities to better understand and enhance animal health. The following discussion explores how AI is reshaping pet care and driving the next generation of veterinary innovation.AI Tools for Pet Health DiagnosticsArtificial intelligence is reshaping veterinary diagnostics by adding speed, consistency, and pattern recognition to the clinical process. While traditional diagnostics rely heavily on training, experience, and visual interpretation, AI helps surface insights that may be difficult to detect during routine exams, especially in early or complex cases.Rather than replacing veterinarians, AI acts as a support layer. It processes large volumes of data in seconds, compares findings against thousands of previous cases, and highlights areas that may require closer attention. The result is earlier detection, more confident diagnoses, and better-informed treatment decisions for pets.Imaging AnalysisMedical imaging plays a central role in veterinary diagnostics, but interpreting X-rays, MRIs, and ultrasounds can be time-consuming and subjective. AI-powered imaging tools improve this process by scanning images for subtle patterns that may not be immediately visible to the human eye.Trained on large datasets of veterinary images, these systems can identify early signs of disease, structural abnormalities, and progressive conditions. They flag areas
Apr 28, 2026 · via yu.edu
Inside Munger Elementary-Middle School on Martin Street on Detroit's southwest side, what looks like a horizontal iPad is installed just inside the school's front entrance. Every visitor to the school must present a valid form of identification, such as a driver's license, to have it scanned by the iPad and then have their face scanned. When the two images are cross-referenced and verified, the visitor's information is saved in the school's system for future visits. The system is part of a new facial recognition software system, called Visitor Aware, now in place at every Detroit Public Schools Community District school. The system matches visitors' faces against a valid form of identification before they're allowed into the school. When visitors are flagged, they may not be allowed in the building. Munger Elementary-Middle School Principal Donnell Burroughs said the software, which was officially rolled out earlier this school year, was needed for security. It's "more secure and safe" than a traditional sign-in sheet for visitors, Burroughs said. But critics have concerns about how facial recognition is being used in schools. Some cited privacy concerns and worried about inaccurate identifications, meaning someone is flagged when he or she shouldn't be. Others said they worry about the technology having a chilling effect, possibly discouraging some parents from visiting their children's schools. “What these systems are trying to do is to automate student security without human involvement," said Molly Kleinman, managing director of the University of Michigan's Science, Technology and Public Policy program. Detroit is one of 12 districts in Michigan that use Visitor Aware, according to its parent company Singlewire. Bloomfield Hills Schools also launched Visitor Aware last month. Both the Detroit school district and Bloomfield Hills Schools said it streamlines the process for allowing visitors into buildings and that keeping students and staff
Apr 27, 2026 · via govtech.com
More Videos
For the average price of a car in the US, you could buy 5 new Chinese EVsOakland Hills unveils new clubhouse that "feels familiar"Oakland Hills officials unveiled the new, 110,000-square-foot clubhouse Monday.Oakland Hills Country Club reopens after fire four years ago
Dearborn Heights to install 250 lit stop signsWhite House responds to Jimmy Kimmel's 'widow' joke about first ladyPalestinians repurpose Gaza rubble as U.S. rebuilding plan stallsJustice Department to close investigation into Fed Chair Jerome Powell
King Charles visits U.S. post-shooting and amid Iran tensionsDozens in New Mexico sought therapy for alleged abuse at Epstein ranchDetroit school adds facial recognition to safety proceduresBeverly Harper of Eastpointe hopes of unifying the country
Apr 27, 2026 · via detroitnews.com
Artificial intelligence has moved well beyond the pilot stage. One survey found that 88% of organizations have implemented AI in at least 1 business function. And engineering is no exception. Already, AI has reshaped how engineers work. Its applications range from design optimization to automation and beyond. It’s expanding the scope of existing jobs and raising the bar on the skills engineers need to stay competitive. Learn more about the impact of AI on engineering jobs, emerging career opportunities, and ways to thrive in a changing field. Key Points - Artificial intelligence has become fundamental enough to shift traditional engineering roles, changing how engineers work and the work itself. 88% of organizations now use AI in at least one business function. - Engineers are using AI to tackle complex real-world problems, with applications spanning predictive maintenance, design optimization, and automation. - AI’s impact on engineering is just beginning. But it’s already creating new job opportunities and demanding new skills to stay relevant. - AI likely won’t replace engineers, but it will affect some roles more than others. That makes adaptability 1 of the most valuable traits in the field right now. What Is Artificial Intelligence in Engineering? AI enables machines to learn, adapt, and perform tasks that typically require reasoning. In engineering, that capability translates into more powerful tools for analyzing systems and solving complex problems. Common artificial intelligence methods and capabilities used in engineering include: - Machine learning (ML): These systems train on data to perform specific tasks with minimal human oversight, learning and adapting from that data along the way. In engineering, ML is also used in automation, predictive maintenance, and systems optimization. - Computer vision: This AI capability interprets and extracts meaning from visual data such as images and video. In engineering, it can support real-time inspection,
Apr 27, 2026 · via intuit.com
Visual AI Apps Compared: What Actually Works in 2026? Which Visual AI Apps Actually Work in 2026? Visual AI is no longer a futuristic concept. It is already part of how people interact with the world around them. From identifying objects to understanding complex scenes, visual AI apps are changing how we search, learn, and make decisions. But with so many options available in 2026, one question keeps coming up: what actually works? The answer is not as simple as picking the most popular app. Different tools are built for different purposes, and understanding those differences can save you a lot of time and frustration. What People Expect from Visual AI Today A few years ago, basic image recognition felt impressive. Now, expectations are much higher. People want more than just a label. They want meaningful answers. When someone points their camera at something, they expect: - Accurate identification - Clear explanations - Context that helps them act on the information For example, identifying a plant is useful. Explaining whether it is safe, common, or valuable is far more helpful. This shift from recognition to understanding is what separates average apps from truly useful ones. The Main Types of Visual AI Apps Most visual AI tools today fall into a few clear categories. 1. Context-Aware Visual AI (Chance AI) Tools like Chance AI focus on understanding what you are looking at, not just recognizing it. Instead of only matching images, they aim to explain objects with context. This approach is useful when: - You do not know what something is - You need more than just a label - You want quick, meaningful answers It works well across categories like plants, products, tools, and unfamiliar objects, making it closer to a real-world assistant than a traditional search tool. 2. General
Apr 27, 2026 · via vocal.media
Figures Abstract Medical image classification requires models that effectively capture both fine-grained local patterns and global anatomical structures while maintaining computational efficiency for clinical deployment. Although state-of-the-art models such as MedMamba utilize State-Space Models (SSMs) to balance accuracy and efficiency, their sequential operations limit parallelism and increase runtime. To overcome these limitations, we propose MedSpectralNet, a lightweight Convolutional Neural Network (CNN) architecture that approximates self-attention with linear complexity to efficiently extract multi-frequency features. The model introduces a dual-stream feature extractor that processes global and local information in parallel, and a ContextGate block that adaptively fuses multi-scale representations. MedSpectralNet is evaluated across six benchmark datasets from MedMNIST (including BloodMNIST, BreastMNIST, DermaMNIST, PneumoniaMNIST, OrganCMNIST, and OrganSMNIST), MedSpectralNet achieves an average accuracy of 93.7% on OrganCMNIST and 98.0% on BloodMNIST, showing 1–4.3% relative accuracy gains when compared to larger transformer-based models. Importantly, it delivers this performance with only 8.5 million parameters, representing approximately 60% fewer parameters than MedMamba-T, which requires 14.5 million parameters. MedSpectralNet has also achieved high AUC values up to 0.999 across multiple classes, demonstrating state-of-the-art accuracy with substantially reduced computational cost and improved parallelization, which makes MedSpectralNet well-suited for real-time and resource-constrained classification-based medical applications. Citation: Afrin N, Fahim MA-NI, Alamro W, Allawi YM, Abadleh A, Sultan SM, et al. (2026) MedSpectralNet: A lightweight convolutional neural network architecture for multi-modal image classification. PLoS One 21(4): e0346128. https://doi.org/10.1371/journal.pone.0346128 Editor: Ali Mohammad Alqudah, University of Manitoba, CANADA Received: October 26, 2025; Accepted: March 16, 2026; Published: April 27, 2026 Copyright: © 2026 Afrin et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability: The data underlying the results presented in the study are available
Apr 27, 2026 · via journals.plos.org
Figures Abstract With the rapid development of wireless communication technologies, spectrum resources are becoming increasingly scarce, and spectrum monitoring technologies targeting control and interference suppression impose higher requirements on the real-time performance, reliability, and intelligence of signal detection, recognition, and key parameter extraction. Traditional signal processing methods heavily rely on operators’ prior knowledge, making it difficult to achieve intelligent spectrum monitoring, and often exhibit poor performance in complex electromagnetic environments with unknown signals or strong interference. Existing deep learning-based automatic modulation recognition techniques are more focused on signal recognition, with relatively limited research on detection and key parameter extraction. To address these challenges, this paper proposes a wideband signal processing frame based on semantic segmentation and signal spectrogram. The frame employs RepViT as the backbone network and achieves detection, recognition, and key parameter extraction of wideband signals through precise semantic segmentation of signal spectrogram. Experimental results on a large-scale synthetic dataset demonstrate that the proposed frame achieves a maximum signal recognition rate (mAcc) of 82.43% and an average signal recognition rate (aAcc) of 65.16% in multi-modulation scenarios and under different noise power levels. In terms of parameter extraction, the normalized root mean squared error (NRMSE) for time parameters (e.g., start time, and duration) is controlled within the ranges of 0.3%−2.8% and 0.4%−1.6%, respectively, while the NRMSE for frequency parameters (e.g., center frequency, and bandwidth) reaches 8.7% and 0.6% in multi-classification tasks, providing an effective reference solution for intelligent wireless signal analysis. Citation: Liu L, Zhu R, Chu P, Shi Z, Tian J, Zhang Y, et al. (2026) A frame of wideband wireless signal recognition and parameter extraction based on semantic segmentation. PLoS One 21(4): e0346685. https://doi.org/10.1371/journal.pone.0346685 Editor: Neng Ye, Beijing Institute of Technology, CHINA Received: December 15, 2025; Accepted: March 22, 2026; Published: April 27, 2026 Copyright: © 2026 Liu
Apr 27, 2026 · via journals.plos.org
JPLoft is recognized for delivering scalable, and enterprise-ready LLM solutions that empower organizations to turn data into intelligent business outcomes. DENVER, CO, UNITED STATES, February 27, 2026 /EINPresswire.com/ — As organizations worldwide accelerate digital transformation initiatives, JPLoft has announced a strategic expansion of its enterprise AI capabilities focused on advanced language intelligence and multimodal systems. The move reinforces the company’s commitment to building scalable, secure, and high-performance AI solutions tailored to complex business environments. Enterprises today are demanding smarter automation, deeper analytics, and more contextual intelligence to stay competitive. In response, JPLoft is reinforcing its technological foundation to build next-generation AI systems tailored to these evolving needs. These advanced solutions are designed to understand language, interpret visual data, and support high-value, data-driven decision-making across complex business environments. Meeting the Enterprise Demand for Language Intelligence The rapid growth of enterprise data—emails, documents, customer chats, reports, contracts, and multimedia content- has created both an opportunity and a challenge. Organizations need systems that can process, interpret, and extract insights from massive volumes of structured and unstructured information. To address this, JPLoft has positioned itself as a trusted LLM development company, helping enterprises design and deploy large language models tailored to their operational requirements. These advanced systems are built to summarize documents, generate reports, power intelligent chat interfaces, automate knowledge management, and enhance internal collaboration. Unlike generic AI tools, JPLoft focuses on domain-specific customization. Each model is trained and fine-tuned to align with industry language, regulatory requirements, and organizational workflows. This ensures relevance, accuracy, and long-term performance sustainability. A Structured Approach to AI Implementation Deploying large-scale AI systems requires more than model training. It demands structured planning, infrastructure readiness, and performance governance. JPLoft follows a phased implementation framework to ensure reliable enterprise adoption. The process includes: Business objective mapping Data readiness assessment Model selection
Apr 27, 2026 · via blufftontoday.com
Australia plans biometric liveness detection refresh for national digital ID Australia plans to contract a biometric liveness detection capability to support the country’s national digital ID and protect it against advanced spoof attacks and emerging fraud threats. The Australian Taxation Office (ATO), which manages myID within the Australian Governments Digital ID System (AGDIS), has published an RFI for biometric liveness detection. The government is also in the process of opening up the AGDIS to private sector identity verification providers. The “Biometric Verification Capabilities to Support myID” must come as a SaaS solution, and support peak workloads of 10,000 verifications per hour with 95th percentile responses within one second. The myID system is working towards adding two new identity proofing levels. IP1+ involves the confirmation of the individual’s name and date of birth through identity document verification. IP2+ includes authentication with Australia’s Face Verification Service (FVS) plus liveness detection. ATO is hoping for responses from suppliers with identity verification expertise, particularly related to liveness detection and facial image capture, biometric matching and credential validation through technologies such as NFC. The agency specifically wants to address advancements in liveness detection since 2021, when its current capability was sourced from iProov. Scalable biometric authentication is also a concern, given that myID is up to 14 million users, and the ATO also wants a capability to assess non-Australian ID documents for Australians living overseas. Requirements include sufficient image quality capture for biometric comparison, based on ISO/IEC 29794-5. Biometric presentation attack detection (PAD) that “meets at least Evaluation Assurance Level 2 (Level B) as defined by ISO/IEC 30107-3:2023 and the Digital ID (Accreditation) Data Standards” must be present and integrated as part of a “single continuous process.” A qualified third party must attest to that level of PAD compliance. The face biometrics matching algorithm must
Apr 27, 2026 · via biometricupdate.com
Olga Cronin Monday 27 April — Last week the Independent Examiner of Security Legislation Judge George Birmingham published his first annual report. In it, he looked at three pieces of security legislation and made recommendations in respect of each: - Interception of Postal Packets and Telecommunications Messages (Regulation) Act 1993; - Criminal Justice (Surveillance) Act 2009; and - Communications (Retention of Data) Act 2011. Although many of his recommendations are to be welcomed, we are very concerned about Judge Birmingham’s position on encryption in particular. On encryption, he concedes that “very many people in many different walks of life on a daily basis use encryption and are encouraged to do so. It has brought significant additional security to financial transactions and communications.” This is correct. Journalists use encryption to protect their sources and patients use it to communicate with their doctors. Undoubtedly, our own politicians, their staff, garda, and intelligence officers themselves use encrypted messaging services because of its security. Encryption protects and secures the processing of our data when we are online banking, online shopping, accessing health data and carrying out our employment. In essence, it is essential for our collective cybersecurity. But Judge Birmingham recommends that legislation be passed to provide “for lawful access to all communications, including encrypted communications, incorporating appropriate safeguards”. ICCL understands that encryption presents a challenge for law enforcement but this position misunderstands the issue and what’s at stake. Creating “appropriate safeguards” won’t solve the fundamental problem - any piercing of encryption on the supposed basis of creating pathways solely for An Garda Síochána introduces vulnerabilities that would be exploited by others, ultimately undermining the security and privacy of everyone. As we’ve previously said, the issue at stake here is not a simple trade-off between individual freedoms, such as privacy and expression, and State
Apr 27, 2026 · via iccl.ie
Why Elon Musk and Sam Altman are fighting over OpenAI
Musk, who co-founded the company that created ChatGPT with Altman, wants more than $130 billion in damages in a lawsuit that could shakeup the artificial intelligence landscape. The BBC's Lily Jamali explains why the two tech giants are facing off in court.
Apr 27, 2026 · via bbc.com