A replay-free method for continual learning scored higher across the reported benchmarks and showed no measured retention loss on TRACE, though its routing has not been tested under severe distribution shifts.
A shared humanoid policy reported higher success than LHM-Humanoid on seen and unseen object-transport tasks, but success fell as the number of objects increased.
A training-free approach that ranks visual tokens by both current relevance and changing importance performed strongly across several multimodal language-model benchmarks, including tests with 88.9% pruning.
A preprint describes a robot that watches for meaningful changes after human-object interactions and uses a vision-language model to plan assistance without a task request. The reported comparison is described as comparable, but the available extraction omits the numerical results and uncertainty estimates needed to judge how close the systems were.
A preprint reports valid-and-unique molecular yields and more distinct elite architectures, while a digit-image control favored Soft on distribution matching.
A new preprint describes LipDA, a detector that uses mismatches between lip motion and head movement to identify LipSync forgeries and estimate which generator family produced them.
A preprint reports favorable shifts on several gender-bias measures across three language models, while general-capability averages remained close to the corresponding base models.
A Dutch preprint links genetic scores associated with emotional stability and conscientiousness to earnings and household wealth, while leaving open whether the associations reflect personality or other genetic pathways.
An arXiv preprint reports higher within-model caption scores for full fine-tuning, while Video-LLaVA prompts remained below the strongest fine-tuned model and its full fine-tuning and LoRA tests were not completed.
A laboratory study reports finite-amplitude droplet pinch-off in deionized-water jets beyond the classical cutoff, while other high-wavenumber settings showed decay or spray.
A method called MOSAIC reported higher macro-client top-1 accuracy than FedDDA across domain-shift and joint-shift benchmark settings. The study points to a better average across clients rather than a universal win, with only three seeds and no confidence intervals reported.
A preprint describes TACD, a distillation method that limits terminal supervision and reports stronger benchmark results, faster generation and lower peak GPU memory across text-to-motion tests.
A preprint reports that PI-SCAN identified and characterized overlapping radio interference in simulated data, but its ability to detect every source fell as mixtures grew more crowded.
A modeled power grid reached 91.50% accuracy on the first 400 MNIST test images, while results varied with network changes, noise and the way images were represented.
A modeling study links fluctuating plankton growth and dispersal to abundance distributions, spatial correlations, Taylor scaling and patchiness, while exposing clear limits in seasonal and oligotrophic data.
A laboratory study of Enterobacter sp. SM3 found that swarm expansion varied with agar orientation, peaking at intermediate angles rather than changing steadily with tilt. The colonies also became more directional, with late growth concentrated toward the downhill front at some angles.
The uncertainty-guided system recorded higher segmentation metrics on an independent cohort, with its largest reported change in tumor core, while a structural tripwire routed 36 of 362 cases to full manual review.
A preprint describes an AI system that combines B-cell repertoire data with experimental feedback to identify broadly neutralizing antibody candidates against several viruses, but the evidence remains preclinical.
An inverse-design study examines how network degree and rewiring relate to modeled topology, chirality and resistance to selected perturbations, with small-world networks standing out in several tests.
CrashSim generated safety-critical scenarios that tracked real-world crash behavior and crash-type statistics more closely than competing methods. In closed-loop tests, the resulting nuCrash set produced much wider performance differences among five autonomous-driving planners.
A preprint testing FlyVision across computer-vision and biomedical datasets found that it nearly matched ResNet18 on MNIST with far fewer parameters, while larger tasks produced mixed results.
A case-based video framework reproduced broad movement trajectories in several tasks, but uneven errors and missing reference intervals exposed limits in detailed reconstruction.
A computational study proposes a coordination-and-distance rule for hydrogen in metal defects and uses it to reproduce a two-peak thermal-desorption profile in tungsten.
A theoretical note separates what Pontryagin duality and Fourier analysis provide from the physical choice that identifies a mathematical coordinate with momentum.
A preprint reports broad, large dielectric responses in pressed BiFeO3 nanopowders with nominal Sm substitutions. The strongest features depended on temperature, frequency and composition, while phase mixtures and particle agglomeration left the mechanism unresolved.
A mathematical analysis of two imaginary gas cells reports a modeled hot-to-cold molecular imbalance and uses it to interpret the Second Law and Loschmidt's paradox.
The tested methods reached more than 99% of the study's HVmax on several graph sets, while the preprint says they may not recover every best trade-off.
A reinforcement-learning retriever called MO-IKE reported higher combined scores than DR-IKE on Llama and Mistral tests, and recorded an overall score of 24.5 on UniEdit.
The proposed AL-LPAL method combines central concentration with very heavy tails and showed robustness in contaminated tests, while interval coverage remained imperfect.
Scientific publication5 min read
Showing 30 of 2037 articles
Research fields
Explore science by topic
Start broad, then move into one of thirty focused subject desks.