ImagenWorld benchmark with 3.6K condition sets for stress-testing image generation models across six core tasks with human evaluation.
Statistical guarantees for distributionally robust optimization using optimal transport divergences for adversarial robustness.
Markov chain-based car-following model for traffic flow using empirical probabilistic paradigm.
Physics-informed neural networks framework (Deflation-PINNs) that identifies multiple distinct solutions to nonlinear PDEs.
Privacy-preserving federated learning framework using flow-matching generation to improve robustness and aggregation in distributed training.
Transformer-based spatial upsampling of head-related impulse responses for binaural audio rendering.
Analysis of prompt injection attacks against LLM agents, tracking attack pipeline stages and defense mechanisms across five frontier models.
Modular transformer approach for efficient domain adaptation in optical character recognition with reduced computational requirements.
Research on how LLMs perform scientific reasoning tasks and how prompting affects their internal reasoning processes.
Training-free spectral modulation of diffusion model cross-attention via Fourier analysis for improved text-to-image control.
3D SONAR sensors for road condition monitoring and damage classification in adverse weather conditions.
Deep reinforcement learning framework using PPO to train virtual agents for guiding fish school collective motion.
Differentiable power-flow optimization using neural networks to replace Newton-Raphson methods for scalable grid simulation.
Domain adaptation method for quantum machine learning using classical shadows to handle distribution shift in quantum data.
Multi-view learning approach with prototypes for thyroid ultrasound classification with improved robustness across devices.
Multi-agent pipeline for literature analysis using Deleuzian ontology to identify non-linear patterns in research landscapes.
Framework for evolutionary GPU kernel optimization using evaluation-driven agent and evolutionary techniques for operator generation.
Ensemble learning system combining deep learning and traditional classifiers for brain tumor MRI classification.
Analysis of prompt framing artifacts in vision-language model evaluation on clinical neuroimaging tasks.
Survey of network performance modeling approaches comparing traditional simulation with deep learning methods.
Novel geometric optimization framework using affine normal descent with invariance under volume-preserving transformations.
Diffusion distillation approach using reinforcement learning to improve student model performance beyond teacher anchoring.
Study of coordinated user manipulation tactics in recommendation systems through collective rating and review mechanisms.
Analysis of instructional alignment in cybersecurity exercises using multimodal data and Bloom's taxonomy coding.
Medical AI Scientist system that autonomously generates hypotheses, conducts experiments, and writes manuscripts in clinical medicine.
Theoretical analysis of approximation constraints in narrow ResNets, proving inability to represent critical points.
Automated pipeline using diffusion models to detect and anonymize sensitive regions in images while preserving utility.
Genetic programming pipeline for automatically evolving interpretable composite features for music tagging tasks.
Study of demographic bias and fairness issues in facial recognition systems used by law enforcement.
Method for improving diversity in text-to-image diffusion transformers through contextual space repulsion to address typicality bias.
Paper examining financial insurance mechanisms for AI system failures and algorithmic liability in high-stakes domains.
Study on few-shot learning and RNNs applying asymptotic equipartition property from information theory to machine learning.
Theoretical analysis of inexact Langevin algorithm convergence for score-based generative models with KL divergence guarantees.
Survey on continual graph learning covering incremental learning from streaming graph data with experience and generative replay approaches.
Mathematical analysis of auto-differentiation reliability in neural-ODE training with high-order numerical methods.
Novel prior learning method for neural networks using structured posteriors to improve generalization and uncertainty estimation.
Proves asymptotic optimality of new restless bandit policies with O(1/√N) gap under unichain and aperiodicity conditions.
Theoretical analysis of sample complexity for model-based Q-learning, establishing finite-time convergence bounds for model-learning algorithms.
Paper proposing Explaining-Away Variational Autoencoders to improve uncertainty representations in deep generative models for visual inference tasks.
Survey of multimodal continual learning methods that enable models to learn from new data across multiple modalities while retaining previous knowledge without catastrophic forgetting.
Transformers learn variable-order Markov chains in-context with finite-sample accuracy analysis using context-tree weighting.
Analysis of Sharpness-Aware Minimization robustness to label noise through gradient down-weighting at element-wise level.
Scalable neural network verification using branch-and-bound with inferred cutting planes instead of external MIP solvers.
Wavelet subspace compression for optimizer states reduces memory during LLM training, improving upon low-rank approaches.
Dataset distillation for credit scoring models addressing class imbalance in pretrained models on tabular financial data.
Binned spectral power loss function for improved deep learning predictions of chaotic multiscale dynamical systems.
Multimodal drug-aware diffusion model for ECG generation in virtual clinical trials with demographic disentanglement.
Steering vectors applied to LLM activations for bias mitigation across social dimensions like age, gender, and race.
Large graph dataset and measurement framework for evaluating long-range interactions in graph representation learning.
Methods to measure faithfulness of concept-based explanations in deep vision models using surrogate models.