Diffusion system reconstructs 3D materials from views and text
Preprint results suggest texture-space diffusion can keep material outputs consistent across views and scale to 8K, but real-data tests remain small and inference is costly.
Search
2,451 results
321–340
Synthetic Spatial Data Linked to Higher Scores on AI Tests
Preprint: An automated multi-agent framework reports gains across spatial benchmarks, but its method depends on simulator scene metadata.
Strong Question-Answering Scores Did Not Reliably Predict Enterprise Agents
Preprint: A benchmark combining enterprise questions with simulated interactive tasks found that the leading agent changed with the task and model.
Unknown Horizons Block Zero-Regret Queue Scheduling
Preprint: A theoretical analysis finds that the best job to serve can change with time left, and no anytime policy can drive regret to zero against the finite-horizon optimum.
Study finds privacy method preserves most benchmark recourse
A preprint reports a formal edge-level privacy guarantee while retaining most of the available counterfactual recourse on benchmark graphs.
Robot Training Method Reports Higher Success in Obstacle Tests
An arXiv preprint reports higher success and lower policy-induced cost across two model variants, while transfer to dynamic obstacles remained unproven.
A model links channel-like opening time to current fluctuations
Preprint calculations connect residence time and integrated current in an asymmetric stochastic model; the biological examples remain model applications.
Quantum code design yields a 13,320-qubit memory on paper
Preprint study describes a flagged-code route to 64 logical qubits and a circuit-distance bound of at least 48, with simulations tied to a specified setup.
Quantum algorithm cuts circuit depth but leaves a costly trade-off
Preprint: A probabilistic quantum method is designed to avoid recursive depth, but simulations show that cumulative success still falls with each iteration.
Superconducting quantum machine records precision beyond classical limit
Preprint reports a TUR ratio of 1.71 plus or minus 0.17 at 81 mK, below the classical benchmark of 2.
Neural animation system adjusts motion transitions to fit each clip
This preprint describes a graph-guided neural system that varied transition length in tests, but the authors report failures on rare poses and no physics constraints.
SimpleEvol leads tests of simpler LLM heuristic design
A preprint reports that SimpleEvol had the highest reported ICE values in the two primary tasks, with weaker transfer on one out-of-distribution benchmark.
Map-Based AI System Leads Robot Navigation Tests, Preprint Says
The model combines a persistent spatial map with action prediction and reports strong results in simulation and limited robot trials.
Adaptive Rotation Improves iSOMA on Difficult Test Landscapes
Preprint: The modified optimizer posted its clearest gains on noiseless BBOB tests, while real-world and quantum-objective results were more mixed.
Cyber risk scores can rank the same vulnerabilities very differently
Preprint: A 2026 arXiv comparison of four vulnerability-scoring systems found overlap in some measures, but concluded they are not interchangeable.
Preprint: Codebook method leads in cross-modal AI tests
An arXiv preprint reports first-place results in seven of eight classification settings and most segmentation measures, but it took longer to train in the reported comparison.
AI method reports low collisions when rebuilding 3D scenes
Preprint: A point-cloud model completes missing object geometry while keeping the result aligned with surrounding scene geometry.
Spectral method shows mixed gains against vision-model hallucinations
Preprint: A training-free decoding method improved several benchmark measures, but results varied across models and tasks.
One AI Model Reports Stronger 2D and 3D Object-Use Predictions
Preprint reports cross-dataset gains, but the system has not been tested in closed-loop robot manipulation.
Holographic model links quantum turbulence to normal-fluid motion
Preprint: A two-dimensional simulation finds direction-dependent, non-Gaussian velocity fluctuations that weaken as temperature rises.