All papers

This paper formulates tilted sampling as an online learning problem for sampling from Gibbs distributions and establishes logarithmic regret bounds.

This paper analyzes the theoretical Lagrangian structure of cost-penalized heuristics in active feature acquisition under hard budget constraints.

This paper introduces a prior-data fitted network that performs simultaneous Bayesian variable selection and nonlinear prediction in a single forward pass.

This paper introduces a spatio-temporal spectral sheaf Laplacian and filter to jointly capture space and time dependencies on graphs.

This paper presents an uncertainty-aware federated learning method to manage power and computing allocation for extended reality devices under state uncertainties.

This paper combines Gaussian process references with targeted diffusion model corrections to infer scientific physical fields from sparse data.

This paper applies deep learning to classify topological equivalence classes of finite group actions on Riemann surfaces.

This paper introduces an interface-based graph representation and encoder for scalable weight-space learning across diverse neural network architectures.

This paper investigates direct key-value cache substitution between frozen language models and finds geometric similarity fails to ensure consumability.

This paper investigates unlabeled data selection strategies for semi-supervised QSAR modeling using large chemical repositories.

This paper explores internal-state sharing in multi-agent language models through a central pool rather than pairwise transfer bridges.

This paper introduces usage-aware Bayesian experimental design to minimize risk when experimental data is processed by unknown or external decision makers.

This paper presents a pipeline parallelism training system that balances data workloads across heterogeneous hardware while reducing communication overhead.

This paper tests a convolutional neural network technosignature detector against synthetic injections in real radio data to reveal out-of-distribution failure modes.

This paper optimizes the predefined phase-length schedules of Successive Rejects algorithms to improve error exponents in fixed-budget best-arm identification.

This paper proposes an attention-based noise-aware training method to improve the robustness of Transformer models deployed on analog compute-in-memory hardware.

This paper evaluates probabilistic AI weather forecasting models against operational physics-based ensembles on tropical cyclone tracking, intensity, and structure.

This paper presents a diffusion-based posterior inference framework to reconstruct physical scenes from ubiquitous wireless channel measurements.

This paper introduces a temporal difference active learning framework to adaptively learn experiment selection policies for budgeted perturbation discovery.

Reorganizes tensor operations in physics-informed DeepONet architectures to eliminate redundancy and reduce training costs for higher-dimensional problems.

Provides a GPU-accelerated Huffman compression framework for tensors to lower memory overhead during model training.

Analyzes generalization and memorization dynamics in exponential Hopfield networks trained with softmax attention over noisy examples.

Derives theoretical high-probability generalization guarantees and early stopping bounds for randomized block coordinate descent on nonconvex problems.

Shares hardware-model co-design techniques and analytical kernel selection for training large recommendation models on AMD GPUs.

Proposes a single-increment rejection sampling algorithm to correct token correlation bias efficiently in parallel masked diffusion.

Introduces a tabular foundation model pretrained on synthetic priors to predict survival curves in-context.

Introduces an algorithm to evaluate multiple target policies simultaneously in contextual linear bandits with shared sampling.

Presents a GPU format that losslessly compresses sparse activations to reduce memory usage during neural network training.

Applies topology-inspired loss functions to fine-tune super-resolution neural networks on scientific wind-field datasets.

Analyzes score centering and covariance estimands to maintain reward shift invariance under regularized importance sampling.

Analyzes rational strategy synthesis and equilibrium guarantees for stochastic multi-agent systems with probabilistic temporal logic objectives.

Establishes a training approach to ensure backward compatibility and prevent false positives across versions of neural audio watermarking.

Develops a robust reinforcement learning framework for dynamic treatment policies under patient measurement and model uncertainty.

Introduces a graph-based retrieval method that traces multi-hop library dependencies to find lemmas for formal theorem proving.

Uses flow matching and distillation to generate full multivariate time-series trajectory forecasts rapidly.

Applies persistent homology to encode global structural connectivity into learned representations of mechanical metamaterials.

Introduces local appearance purification to filter alignment targets and improve representation learning during diffusion model training.

Selects maximally informative pixel tokens directly from vision transformers to train segmentation heads under tight annotation budgets.

Evaluates on-policy distillation versus supervised fine-tuning in vision-language models, analyzing how teacher reliability impacts student behavior.

Combines visual tracking and vocal fingerprinting to identify and monitor individual dogs across real-world multi-dog video recordings.

Introduces a Shapley-based attribution method that jointly explains spatial and spectral channel importance in multispectral imagery.

Proposes invertible bi-Lipschitz adapters to align pretrained latent spaces with semantic representations for flow matching without retraining encoders.

Presents an auditable framework that uses risk budgeting to selectively deploy generative enhancers over base restoration models in low-light images.

Introduces a label-efficient self-supervised learning framework that splits volumetric CT scans into local slice and global context representations.

Proposes a framework that increases image diffusion throughput by fusing and processing multiple generation samples simultaneously in manifold space.

Identifies early-stage issues with classifier-free guidance in flow matching models and proposes a correction for controllable image editing.

Validates an unsupervised domain adaptation model to classify normal and abnormal retinal development across diverse optical coherence tomography systems.

Proposes a single-stage joint diffusion method to forecast solar irradiance directly from sky images without requiring a separate regression model.

The paper applies implicit neural representations with temporal consistency constraints to register periodic medical image sequences across time.

The paper introduces a multiple instance learning method to recover coexisting secondary lesions from whole-slide pathology images labeled only by dominant severity.

The paper introduces a compact vision-language-action model using register tokens as persistent memory to reduce control oscillations in robotics.

The paper presents a refiner that models motion proposals as heterogeneous temporal graphs to improve trajectory forecasting for autonomous vehicles.

The paper investigates learning physical damping properties from impact audio and distills this knowledge into vision models.

Evaluates classical machine learning models and preprocessing choices for detecting cryptojacking attacks across two benchmark proxy datasets.

Introduces a hybrid neural network fusing temporal, contextual, and relational features with explainable mechanisms for secure cloud decision systems.

Presents a hardware-software contract framework to characterize microarchitectural side-channel leakage at the instruction set architecture level.

Evaluates a software-defined radio implementation for verifying combinatorial watermarks in satellite positioning, navigation, and timing signals.

Presents a machine-learning-assisted physical-layer attack that nullifies authentic GPS signals without requiring victim-side channel information.

Studies the resilience of the Lightning Network during a client node shutdown to measure the practical value of client diversity.

Demonstrates side-channel and covert-channel security risks in hardware-assisted page migration on unified CPU-GPU superchips.