Search the full wire by company, model, lab, or keyword. Every story we have ever aggregated.
Just-in-Time diffusion-based defense mitigates adversarial attacks on deep neural networks for radio signal modulation classification with improved generalization.
Neural network deformation fields generate facial expression blendshapes for disconnected-part anime 3D models compatible with ARKit/MediaPipe standards.
GUIFlow framework accelerates vision-language model GUI agents via incremental inference that reuses overlapping multi-image representations.
Graph neural network approach for detecting illicit financial accounts via attentive message passing, addressing structural inconsistency in transaction networks.
Differential neural tangent kernel symmetries under orthogonal groups characterized for overparameterized physics-informed networks.
Information geometry analysis of neural representation formation trajectories using Bures geometry and quantum Fisher information.
Reversible pseudonymisation benchmark and engine for Russian/English documents sent to external LLMs, evaluating five tools and regex baseline.
Re-evaluation of neural operator baselines on The Well physics simulation benchmark reveals prior results were under-tuned.
WebViSE framework extracting reusable visual design patterns from open web to improve LLM website generation from natural-language specifications.
Neural Eigenfluid Bases apply learned spectral representations with differential constraints for reduced-order fluid simulation modeling.
Min-Marginal Power Index applies game-theoretic feature selection to multi-objective settings, labeling features harmful, neutral, or decisive via coalitional utility thresholds.
LLM-guided symbolic regression framework improves mathematical expression discovery by combining search-based methods with adaptive generation and reusing prior search insights.
FlashScript compiles successful agent runs into reusable neuro-symbolic programs, replacing repeated reasoning with parameterized code and LLM decision points.
Deploying a trained neural network involves more than translating each operator into a GPU kernel. It is a compilation problem with many legitimate answers. For... Deploying a trained neural network involves more than translating each operator into a GPU kernel. It is a compilation problem with many legitimate answers. For example, when NVIDIA TensorRT builds an engine, it makes interdependent decisions about graph-IR transformations, fusion, implementation selection, memory placement, and quantization. This process is called scheduling. Source
ZENO: neural operator framework for solving high-dimensional nonlinear filtering via Zakai equation SPDEs, combining Fokker-Planck dynamics with likelihood updates.
Neural LoFi extended with backward coupling to model cross-layer feature learning under joint training; spectral theory analysis in controlled regime.
HarnessMM reduces token consumption in multi-turn agents via decision reuse and recursive self-improvement of tool harness configurations.
ARENA neural architecture predicts solar flares from variable active-region imagery using multiple-instance learning without flare-origin labels.
Spike-and-slab Bayesian framework with PDMP sampling enables tractable inference for sparse neural networks via principled pruning.
Study investigates identifiability of hybrid deep generative models combining physics constraints with neural components via meta-learning.
LLM agents reuse docking probes and edits across protein pockets for structure-based drug design via experience replay.
Neural modeling of 3D Navier-Stokes turbulence using 28TB+ time-resolved DNS to evaluate numerical schemes for multi-scale flows.
Neural network approximation of symmetric polyconvex elastic energies using ReLU input-convex networks with monotonicity constraints.
Hybrid weather forecasting system combining neural networks with 4D-Var data assimilation for state-dependent model-error correction in the Integrated Forecasting System.
Group Relative Policy Optimization (GRPO) using pseudo-labels as rewards outperforms supervised fine-tuning for domain-adapted ASR, achieving 29-56% WER improvement over Whisper.
Analyzes phase transitions in approximation bounds for partially trainable neural networks versus fixed random features under spectral decay assumptions.
VERA dashboard operationalizes EU AI Act compliance for LLM deployments, mapping 18 technical requirements to measurable benchmarks for non-engineer stakeholders.
Production case study: moving compliance rules from prompts to structured retrieval context improves LLM pharmaceutical claims generation.
Neuro-symbolic household food recommender balancing dietary constraints and fairness via constrained fair division.