News Feed Hub

arxiv:2504.18541v1 announce type: new abstract: we present several algorithms to generate tables for asymmetric numeral systems and prove that they are optimal in terms of discrepancy. in turn, this gives rise to the strongest proven bound on entropy loss. we further give improved theoretical bounds for the entropy loss in tabled asymmetric numeral systems and a brief empirical evaluation of the stream variant.

arxiv:2504.18544v1 announce type: new abstract: generating synthetic tabular data can be challenging, however evaluation of their quality is just as challenging, if not more. this systematic review sheds light on the critical importance of rigorous evaluation of synthetic health data to ensure reliability, relevance, and their appropriate use. based on screening of 1766 papers and a detailed review of 101 papers we identified key challenges, including lack of consensus on evaluation methods, improper use of evaluation metrics, limited input from domain experts, inadequate reporting of dataset characteristics, and limited reproducibility of results. in response, we provide several guidelines on the generation and evaluation of synthetic data, to allow the community to unlock and fully harness the transformative potential of synthetic data and accelerate innovation.

arxiv:2504.18547v1 announce type: new abstract: pre-trained vision transformers have achieved remarkable performance across various visual tasks but suffer from expensive computational and memory costs. while model quantization reduces memory usage by lowering precision, these models still incur significant computational overhead due to the dequantization before matrix operations. in this work, we analyze the computation graph and propose an integerization process based on operation reordering. specifically, the process delays dequantization until after matrix operations. this enables integerized matrix multiplication and linear module by directly processing the quantized input. to validate our approach, we synthesize the self-attention module of vit on a systolic array-based hardware. experimental results show that our low-bit inference reduces per-pe power consumption for linear layer and matrix multiplication, bridging the gap between quantized models and efficient inference.

arxiv:2504.18556v1 announce type: new abstract: deep neural networks (dnns) are highly susceptible to adversarial samples, raising concerns about their reliability in safety-critical tasks. currently, methods of evaluating adversarial robustness are primarily categorized into attack-based and certified robustness evaluation approaches. the former not only relies on specific attack algorithms but also is highly time-consuming, while the latter due to its analytical nature, is typically difficult to implement for large and complex models. a few studies evaluate model robustness based on the model's decision boundary, but they suffer from low evaluation accuracy. to address the aforementioned issues, we propose a novel adversarial robustness evaluation metric, robustness difference index (rdi), which is based on sample clustering features. rdi draws inspiration from clustering evaluation by analyzing the intra-class and inter-class distances of feature vectors separated by the decision boundary to quantify model robustness. it is attack-independent and has high computational efficiency. experiments show that, rdi demonstrates a stronger correlation with the gold-standard adversarial robustness metric of attack success rate (asr). the average computation time of rdi is only 1/30 of the evaluation method based on the pgd attack. our open-source code is available at: https://anonymous.4open.science/r/rdi-b1da.

arxiv:2504.18560v1 announce type: new abstract: large language models (llms) have exhibited impressive natural language processing capabilities but often perpetuate social biases inherent in their training data. to address this, we introduce multilingual augmented bias testing (mla-bite), a framework that improves prior bias evaluation methods by enabling systematic multilingual bias testing. mla-bite leverages automated translation and paraphrasing techniques to support comprehensive assessments across diverse linguistic settings. in this study, we evaluate the effectiveness of mla-bite by testing four state-of-the-art llms in six languages -- including two low-resource languages -- focusing on seven sensitive categories of discrimination.

arxiv:2504.18562v1 announce type: new abstract: deep learning models, especially large transformers, carry substantial "memory" in their intermediate layers -- an \emph{internal world} that encodes a wealth of relational and contextual knowledge. this work harnesses that internal world for wildfire occurrence prediction by introducing a modular architecture built upon gemma 3, a state-of-the-art multimodal model. rather than relying on gemma 3's original embedding and positional encoding stacks, we develop a custom feed-forward module that transforms tabular wildfire features into the hidden dimension required by gemma 3's mid-layer transformer blocks. we freeze these gemma 3 sub-layers -- thus preserving their pretrained representation power -- while training only the smaller input and output networks. this approach minimizes the number of trainable parameters and reduces the risk of overfitting on limited wildfire data, yet retains the benefits of gemma 3's broad knowledge. evaluations on a moroccan wildfire dataset demonstrate improved predictive accuracy and robustness compared to standard feed-forward and convolutional baselines. ablation studies confirm that the frozen transformer layers consistently contribute to better representations, underscoring the feasibility of reusing large-model mid-layers as a learned internal world. our findings suggest that strategic modular reuse of pretrained transformers can enable more data-efficient and interpretable solutions for critical environmental applications such as wildfire risk management.

arxiv:2504.18563v1 announce type: new abstract: text-to-image diffusion models are increasingly vulnerable to backdoor attacks, where malicious modifications to the training data cause the model to generate unintended outputs when specific triggers are present. while classification models have seen extensive development of defense mechanisms, generative models remain largely unprotected due to their high-dimensional output space, which complicates the detection and mitigation of subtle perturbations. defense strategies for diffusion models, in particular, remain under-explored. in this work, we propose spatial attention unlearning (sau), a novel technique for mitigating backdoor attacks in diffusion models. sau leverages latent space manipulation and spatial attention mechanisms to isolate and remove the latent representation of backdoor triggers, ensuring precise and efficient removal of malicious effects. we evaluate sau across various types of backdoor attacks, including pixel-based and style-based triggers, and demonstrate its effectiveness in achieving 100% trigger removal accuracy. furthermore, sau achieves a clip score of 0.7023, outperforming existing methods while preserving the model's ability to generate high-quality, semantically aligned images. our results show that sau is a robust, scalable, and practical solution for securing text-to-image diffusion models against backdoor attacks.

arxiv:2504.18564v1 announce type: new abstract: recent research has focused on exploring the vulnerabilities of large language models (llms), aiming to elicit harmful and/or sensitive content from llms. however, due to the insufficient research on dual-jailbreaking -- attacks targeting both llms and guardrails, the effectiveness of existing attacks is limited when attempting to bypass safety-aligned llms shielded by guardrails. therefore, in this paper, we propose dualbreach, a target-driven framework for dual-jailbreaking. dualbreach employs a target-driven initialization (tdi) strategy to dynamically construct initial prompts, combined with a multi-target optimization (mto) method that utilizes approximate gradients to jointly adapt the prompts across guardrails and llms, which can simultaneously save the number of queries and achieve a high dual-jailbreaking success rate. for black-box guardrails, dualbreach either employs a powerful open-sourced guardrail or imitates the target black-box guardrail by training a proxy model, to incorporate guardrails into the mto process. we demonstrate the effectiveness of dualbreach in dual-jailbreaking scenarios through extensive evaluation on several widely-used datasets. experimental results indicate that dualbreach outperforms state-of-the-art methods with fewer queries, achieving significantly higher success rates across all settings. more specifically, dualbreach achieves an average dual-jailbreaking success rate of 93.67% against gpt-4 with llama-guard-3 protection, whereas the best success rate achieved by other methods is 88.33%. moreover, dualbreach only uses an average of 1.77 queries per successful dual-jailbreak, outperforming other state-of-the-art methods. for the purpose of defense, we propose an xgboost-based ensemble defensive mechanism named eguard, which integrates the strengths of multiple guardrails, demonstrating superior performance compared with llama-guard-3.

arxiv:2504.18565v1 announce type: new abstract: uncontrollable autonomous replication of language model agents poses a critical safety risk. to better understand this risk, we introduce replibench, a suite of evaluations designed to measure autonomous replication capabilities. replibench is derived from a decomposition of these capabilities covering four core domains: obtaining resources, exfiltrating model weights, replicating onto compute, and persisting on this compute for long periods. we create 20 novel task families consisting of 86 individual tasks. we benchmark 5 frontier models, and find they do not currently pose a credible threat of self-replication, but succeed on many components and are improving rapidly. models can deploy instances from cloud compute providers, write self-propagating programs, and exfiltrate model weights under simple security setups, but struggle to pass kyc checks or set up robust and persistent agent deployments. overall the best model we evaluated (claude 3.7 sonnet) has a >50% pass@10 score on 15/20 task families, and a >50% pass@10 score for 9/20 families on the hardest variants. these findings suggest autonomous replication capability could soon emerge with improvements in these remaining areas or with human assistance.

arxiv:2504.18566v1 announce type: new abstract: distributed denial of service (ddos) attacks represent a persistent and evolving threat to modern networked systems, capable of causing large-scale service disruptions. the complexity of such attacks, often hidden within high-dimensional and redundant network traffic data, necessitates robust and intelligent feature selection techniques for effective detection. traditional methods such as filter-based, wrapper-based, and embedded approaches, each offer strengths but struggle with scalability or adaptability in complex attack environments. in this study, we explore these existing techniques through a detailed comparative analysis and highlight their limitations when applied to large-scale ddos detection tasks. building upon these insights, we introduce a novel generative adversarial network-based feature selection (ganfs) method that leverages adversarial learning dynamics to identify the most informative features. by training a gan exclusively on attack traffic and employing a perturbation-based sensitivity analysis on the discriminator, ganfs effectively ranks feature importance without relying on full supervision. experimental evaluations using the cic-ddos2019 dataset demonstrate that ganfs not only improves the accuracy of downstream classifiers but also enhances computational efficiency by significantly reducing feature dimensionality. these results point to the potential of integrating generative learning models into cybersecurity pipelines to build more adaptive and scalable detection systems.

here’s what crypto investors make of president donald trump’s first 100 days back in office.

volatile post-earnings moves offer the potential for quick gains.

shares of starbucks have fallen 8% this year, as investors worry about tariffs and how they will affect consumer spending.

amazon probably won’t be hiking prices too much if it’s declining to show how much tariffs are impacting costs.

a mizuho analyst was impressed by paypal’s performance on several closely watched metrics, including transaction margin dollars, which measures the profitability of payment volume.

amazon founder jeff bezos has cozied up to president donald trump since the republican won a second white house term.

“my son just said, ‘dad didn’t want them to get anything.’”

lutnick said a trade deal is "done, done, done, done," but he declined to say which country he was talking about.

pfizer’s first-quarter revenue missed analyst estimates as it faces an “uncertain and volatile external environment.”

amazon said tuesday it considered displaying import charges on items sold via its site for ultradiscount items, but that the plan "was never approved and not going to happen."

the 2025 met gala—themed "superfine: tailoring black style"—is an ode to the black dandy. ahead of the event on may 5, here's everything you need to know about black dandyism.

cardi b's fashion and beauty met gala looks are always one of the most anticipated of the night. here, we take a look back at the rapper's best beauty moments.

on monday, liberal party candidate mark carney defeated conservative leader pierre poilievre to maintain his post as prime minister.

charli xcx, meryl streep as aslan, and emma mackey as the white witch—the four-time oscar-nominated auteur’s reimagining of the childhood classic will be one to remember.

here’s how luxury sparkling teas are transforming the no-alcohol space.

acielle tanbetova is on the ground in the italian paradise to capture the latest chanel collection.

the princess of wales’s favored silhouette of denim has been an ambient presence in just about every casual outfit she has ever been photographed in.

in honor of prince william and kate middleton’s wedding anniversary, we look back at the couple’s best moments.

from adidas sambas to nike cortez and dries van noten’s trainers, here are six sneaker outfits that prove the comfortable shoe can be worn all spring & summer.

new technique does away with the need for strong magnetic fields to control spin currents the post layer-spintronics makes its debut appeared first on physics world.

a research team from the hefei institutes of physical science of the chinese academy of sciences has developed a novel large-scale compound cryopump (multi-stage cryopump) capable of separating fuel particles from helium ash.

the beloved italian pasta cacio e pepe is perhaps best known for two things: being delicious and being frustratingly difficult to cook. at first glance, it looks like a simple recipe, containing only three ingredients: pasta, pecorino romano cheese, and black pepper. but as anyone who has tried to make it will know, the cheese will often clump when added to the hot pasta water, turning what is supposed to be a smooth, creamy sauce into a stringy, sticky mess.

iulia georgescu tumbles down the rabbit-hole to explore the curiosity that is a quantum cheshire cat the post curiouser and curiouser: delving into quantum cheshire cats appeared first on physics world.

researchers have demonstrated a new quantum sensing technique that widely surpasses conventional methods, potentially accelerating advances in fields ranging from medical imaging to foundational physics research, as shown in a study published in nature communications.

researchers at rice university have developed a new machine learning (ml) algorithm that excels at interpreting the "light signatures" (optical spectra) of molecules, materials and disease biomarkers, potentially enabling faster and more precise medical diagnoses and sample analysis.

an atomic clock research team from the national time service center of the chinese academy of sciences has proposed and implemented a compact optical clock based on quantum interference enhanced absorption spectroscopy, which is expected to play an important role in micro-positioning, navigation, timing (μpnt) and other systems.

first report by the national quantum mission also warns that india needs to improve its quantum security the post india must boost investment in quantum technologies to become world leader, says report appeared first on physics world.

it would be difficult to understand the inner workings of a complex machine without ever opening it up, but this is the challenge scientists face when exploring quantum systems. traditional methods of looking into these systems often require immense resources, making them impractical for large-scale applications.

he qinglin's group at the center for quantum materials science, school of physics, has reported the first observation of non-reciprocal coulomb drag in chern insulators. this breakthrough opens new pathways for exploring coulomb interactions in magnetic topological systems and enhances our understanding of quantum states in such materials. the work was published in nature communications.

the federal emergency management agency (fema) has been cleared of wrongdoing in a probe focusing on anti-trump bias, according to a monday court document. according to a justice department email in the document, a fema probe investigating bias against trump-supporting homes allegedly displayed by a florida-based fema employee in the wake of hurricane milton and…

house republicans are proposing placing fees on electric vehicles to fund transportation department programs as part of the tax reform bill up for consideration. this week, rep. sam graves (r-mo.), who leads the house transportation and infrastructure committee, will take up the proposal to boost funding in various transportation sectors and add a new $200…

novo nordisk is partnering with online telehealth companies to sell a direct-to-consumer, lower cost version of its popular weight-loss drug wegovy. the pharmaceutical company announced it would be partnering with hims & hers., lifemd and ro. to provide all five dose strengths of its weight-loss glp-1 medication wegovy at a reduced cost of $499. all…

these orders and deals threaten to corrode the rule of law, chill access to quality legal representation and backfire on the firms.

president trump will speak before national guard troops at selfridge national guard base in harrison township, mich., on tuesday afternoon. his remarks come as the president marks his first 100 days in office, which includes executive actions that have weighed heavily on the military and scrutiny over defense secretary pete hegseth’s use of signal chats…

over the past few years, these buoys were used by texas across the rio grande to combat the flow of illegal immigration. the move generated controversy.

illinois gov. jb pritzker’s (d) political team is dialing up the heat on the trump administration, announcing a new video series on tuesday that seeks to show how illinois residents are negatively impacted by president trump. “i had a national security job lined up. i passed all my prerequisite testing. i went through all the…

president trump said amazon ceo jeff bezos did the right thing after the e-commerce giant pushed back on reports that it was planning to display how much tariffs add to costs. trump and bezos spoke on tuesday morning about the reports of amazon considering a tariff display, a source familiar told the hill, and, later…

republican lawmakers have proposed legislation that would allow felony murder charges and life or death sentences for those who sell or give out fentanyl that leads to an overdose death.

the neighborhood homes investment act is a bipartisan bill that would create a new tax credit to bridge the gap between the cost of building or repairing a home and its value, helping to revitalize struggling urban and rural communities while making homeownership accessible for first-time buyers.

using the james webb space telescope, astronomers have peered into galaxies that existed at cosmic noon to hunt ravenously feeding black holes and bursts of starbirth.

one of the most famous findings in physics could be wrong – the double-slit experiment was long thought to confirm that light can be a wave, but its results can be fully explained using only quantum particles

putting air filters in classrooms seems to boost student attendance, which may be due to them reducing levels of air pollution, pollen, pathogens or all three

'this is a true, true sea monster.' the post geologists accidentally found a monstrous mosasaur fossil in mississippi mud appeared first on popular science.

scientists in china have refuelled a thorium reactor on the fly for the first time. the breakthrough is paving the way for working reactors that are significantly safer than conventional alternatives.

power blackouts that left millions of people across spain and portugal without electricity may have been caused by a bizarre atmospheric phenomenon, though the true cause is yet to be determined.

set on the edge of the lake near the la tene archeological site in switzerland, the latenium museum has a large number of artifacts spanning the bronze age and iron age. the la tene culture covered much of central europe from approximately 500 b.c. to 1 b.c. active archeology continues in the lake area near the museum. it's a great introduction to a celtic culture that was in contact with the greco-roman world. early stuff goes back even farther into the bronze age, and the cast swords on display are great examples of the craft.

greenbrier cove is a trailhead for numerous popular trails within the great smoky mountains national park, and also home to some of the park's most accessible opportunities for water play. the road to the greenbrier cove, its ranger station and picnic area turns off of u.s. 321 midway between gatlinburg and the cosby campground. the partially paved road runs alongside the little pigeon river, that stretch of which is essentially a natural water park layered with small rapids and swimming holes. the road and informal trails alongside it provide multiple access points to the river. the best strategy is to stake out a stretch of river as a home base, but take a tube and enjoy the stretches of rapids. it's easty to float down a section, paddle back to shore, then use the trail or road to quickly walk back up and ride the stretch again. the park provides a ranger station and picnic area with additional amenities and a more formal place to enjoy a meal. numerous trails branch from the road as well, including routes to the appalachian trail. greenbrier also is the trailhead for the very popular ramsay cascades trail, which ascends four miles past especially big trees to the highest waterfall that's accessible by trail in the park.

among korkeasaari zoo’s many historical landmarks, the bear castles are among the oldest. built during the early 1900s, these three cages were once a central attraction at the zoo, showcasing polar bears and finland’s native brown bears. their design now offers a glimpse into the past’s standards for animal welfare. the bear castles were constructed with sturdy stone walls and iron bars. the enclosures were small and lacked the naturalistic elements we now recognize as essential for animal well-being. as attitudes toward animal care evolved, so did the standards at the zoo. in the 1950s, the old bear cages were retired, and their residents were relocated to more spacious and enriching new exhibits. today, the old bear cages remain on the island. they are not used as animal enclosures, but preserved as historical artifacts and places for art installations. they serve as a reminder of how far we have come in understanding and respecting the needs of animals in human care.

"what are sometimes called superbugs — microbes resistant to every available drug — are not merely the stuff of nightmares. they are taking lives right now."

what does the future of space exploration look like? at the 2025 first robotics world championship in houston, nasa gave student robotics teams and industry leaders a first-hand look—complete with lunar rovers, robotic arms, and real conversations about shaping the next era of discovery.  nasa engaged directly with the artemis generation, connecting with more than […]

firefly aerospace's alpha rocket suffered an anomaly during its sixth-ever launch today (april 29), resulting in the loss of its payload, a lockheed martin technology demonstration.

houston company axiom space announced on tuesday (april 29) that it's targeting may 29 for the launch of its fourth astronaut mission to the international space station.

six different rockets launched toward orbit during an approximately 17.5-hour stretch, topping by two the previous record for most liftoffs in a 24-hour span.

amateur astrophotographer dr. michele hernandez baylisshas captured incredible views of the croc's eye and whirlpool galaxies from her vermont-based backyard observatory.

landing on the moon is not easy, particularly when a crew or spacecraft must meet exacting requirements. for artemis missions to the lunar surface, those requirements include an ability to land within an area about as wide as a football field in any lighting condition amid tough terrain. nasa’s official lunar landing requirement is to […]

jared isaacman faces a vote by the u.s. senate committee on commerce, science, and transportation to forward his confirmation as the new nasa administrator.

today’s rather aquatic-themed nasa/esa hubble space telescope image features the spiral galaxy messier 77, also known as the squid galaxy, which sits 45 million light-years away in the constellation cetus (the whale). the designation messier 77 comes from the galaxy’s place in the famous catalog compiled by the french astronomer charles messier. another french astronomer, pierre méchain, discovered the […]

at nasa’s michoud assembly facility in new orleans, the liquid hydrogen tank for the artemis 3 space launch system rocket was moved into the factory's final assembly area on april 22, 2025.

new data from the agency’s jovian orbiter sheds light on the fierce winds and cyclones of the gas giant’s northern reaches and volcanic action on its fiery moon. nasa’s juno mission has gathered new findings after peering below jupiter’s cloud-covered atmosphere and the surface of its fiery moon, io. not only has the data helped […]

one court said fcc violated right to trial, but other courts haven't ruled yet.

gavin kliger helped oversee mass firings at the consumer financial protection bureau and could be violating federal ethics laws.

marvel is hoping to not have to give away too much publicly because it could ruin future surprises.

the rocket's first stage may have exploded moments after it separated from the upper stage.

openai ceo sam altman said tuesday the company is “rolling back” the latest update to the default ai model powering chatgpt, gpt-4o, after complaints about strange behavior, in particular extreme sycophancy. “[w]e started rolling back the latest update to gpt-4o last night,” altman wrote in a post on x. “[i]t’s now 100% rolled back for free [chatgpt] […]

dying for sex, the handmaid’s tale, and a thousand blows are just a few of the shows you should be watching on hulu this month.

caastle, the embattled fashion startup whose board of directors accused its founder, christine hunsicke, of financial misconduct, is starting to face lawsuits from a partner and a supplier over missed payments and more allegations of fraud. as first reported by axios and by suits seen by techcrunch, caastle is being sued by p180, a vehicle […]

star trek, d&d, and back to the future may have gotten fancier set ups, but the straw hats are drawing the biggest crowds at universal's after-hours event.

"this was never approved and is not going to happen," says amazon.

pegatron says the uncertainty makes it hard to make contingency plans.

geospace model version 2.0 begins operations in support of power gridson february 3, 2021*, geospace model version 2.0, which is part of the university of michigan’s space weather modeling framework, will commence operations in support of customers affected by geomagnetic disturbances. this model replaces version 1.5 which has been in operations since november 2017. the geospace model (swmf) is a coupled global magnetohydrodynamic (mhd) model of earth’s geospace environment that extends from near earth’s surface to 32 earth radii toward the sun on the dayside and 120 earth radii into earth’s magnetotail on the nightside. the geospace model utilizes three components of the university of michigan's space weather modeling framework (swmf). the model components are a global mhd magnetosphere, the rice convection model for the inner magnetosphere, and the ridley ionosphere model. model predictions include ground magnetic disturbances resulting from geospace interactions with the solar wind. such magnetic disturbances induce a geoelectric field that can drive currents (geomagnetically induced currents or gics) in large-scale electrical conductors, such as the power grid, and have the potential, during disturbed times, to damage such systems. short-term advanced warning from the model provides forecasters and power grid operators with situational awareness about harmful currents and allows time to mitigate the problem and maintain the integrity of the electric power grid. the most significant changes between version 1.5 and 2.0 are: increased resolution for solving the mhd equations in targeted regions of the grid, moving from ~ 1 million to 1.9 million grid cells improved auroral oval specification and more realistic representation of magnetospheric current systems new tail composition settings to better represent current systems responsible for the disturbance storm time index (dst) a new method for calculating a predicted-estimated kp (global geomagnetic activity index) based on magnetic variations from the geospace model processed by a kp algorithm that is the same as the one swpc uses for calculating kp from ground-based magnetometer stations. the geospace model products available on swpc website include: geospace geomagnetic activity plot, which shows solar wind input to the model as well as predicted kp and dst compared to the observed values geospace ground magnetic perturbation maps, which illustrate magnetic perturbations over north america, the polar regions, and the entire earth geospace magnetosphere movies, which show model density, velocity, and pressure in equatorial and meridian views in earth’s magnetosphere *delayed from january 28 because it changes are not permitted during severe weather in order to assure continuity of service.

the sun (euv) coronal mass ejections the aurora goes x-ray flux goes proton flux updated time: noaa scales geomagnetic storms kp kp = 5 (g1) kp = 6 (g2) kp = 7 (g3) kp = 8, 9- (g4) kp = 9o (g5)