All papers

This paper analyzes how unsupervised deep networks under-represent high-frequency signals when learning representations from raw time-domain EEG data.

This paper evaluates neural network models to filter background noise in multi-telescope gamma-ray event detection.

This paper presents a moment-based anomaly detection framework for high-dimensional mixed data using dependence-aware aggregation across features.

The paper introduces structured random transform constructions to accelerate 1-bit quantized Johnson-Lindenstrauss compression for key-value caches.

The paper presents a generative flow matching surrogate to simulate bubbly flow interface deformation and trajectories rapidly.

The paper introduces an R package implementing nature-inspired optimization algorithms for spatial fuzzy clustering.

The paper combines neural operator priors with sequential Monte Carlo to improve sampling efficiency when solving partial differential equations.

The paper shows that resampling the decoded answer state improves recursive reasoning recovery more efficiently than perturbing latent states.

The paper presents an epigraph lifting and difference-of-convex scheme to solve inverse problems regularized by neural networks.

The paper proposes a black-box framework correcting time series foundation model forecasts using retrieved historical analogues and error residuals.

The paper introduces an interpretable causal forest extension using functional decomposition to explain covariate contributions to treatment effect heterogeneity.

The paper develops a dissimilarity metric for temporal networks based on how epidemic spreading processes unfold across them.

The paper proposes an adaptive treatment selection algorithm controlling false discovery rates under partial feedback with minimax optimality.

The paper introduces a trust-calibrated reinforcement learning framework coupling vision-language-action policy optimization with an evolving simulator.

The paper derives asymptotic risk bounds for constant and warmup-stable-decay learning rate schedules in streaming least-squares regression.

The dissertation presents offline-trained neural implicit representations of reachability corrections for real-time safe robot navigation in dynamic environments.

The paper proposes a neural architecture and reinforcement learning method to optimize market making under non-Markovian Hawkes order-flow dynamics.

The paper surveys theoretical relationships linking machine learning algorithms, data compression methods, and computational meta-complexity.

The paper proves a representation theorem and certified generalization bounds for learning functions governed by graded sparse interactions.

The paper uses a self-supervised convolutional neural network as a preconditioner to accelerate iterative solvers for parametric Helmholtz equations.

The paper establishes two provably convergent differentiable algorithms for pairwise conditional random field regularization in image and signal classification.

The paper introduces a computationally efficient method combining prediction error methods and moving horizon estimation to identify nonlinear dynamic models.

The paper derives PAC-Bayesian generalization bounds that substitute standard relative entropy with general optimal transport metrics.

The paper analyzes human gaze sequences across datasets using motion tokens to demonstrate domain-dependent local transition patterns.

The paper proposes using normalizing flows to evaluate high-dimensional Markov Chain Monte Carlo convergence from multivariate joint posteriors.

The paper presents a language-aware framework using language models and mixture-of-experts adaptors to improve speech recognition during code-switching.

The paper introduces a data-anchored penalty to prevent distributional bias and leakage in physics-informed diffusion models.

The paper introduces a multiscale octree flow framework for generating three-dimensional molecular structures and atom placements.

The paper proposes a hyperbolic aggregation framework for self-supervised brain wave representation learning across heterogeneous electrode montages.

The paper explores prefill-decode disaggregation architectures to speed up rollout generation during agentic reinforcement learning training.

The paper proposes an implicit integration approach for flow-based reinforcement learning policies to enable stable off-policy optimization.

The paper introduces an advantage-guided policy optimization framework that decouples policy representation from reinforcement learning policy updates.

The paper combines natural language reasoning with executable program evidence using reference anchoring to improve mathematical problem solving.

The paper presents a causal action geometry framework to estimate stable decision-response shapes from non-stationary time series.

The paper introduces an industrial time-series forecasting benchmark to evaluate foundation models on multi-rate and multivariate data.

The paper studies sequence-length parameter bounds and representation geometry required for attention layers to perform counting tasks.

The paper applies optimal transport with Gaussian mixtures to improve flow matching sample efficiency without requiring training labels.

The paper introduces a model that predicts wireless propagation across varying physical scenes, transceiver positions, and antenna configurations.

The paper investigates information-theoretic and geometric compression mechanisms over time within recurrent neural networks.

The paper introduces an orthogonal parameter alignment approach for low-rank adaptation to mitigate catastrophic forgetting in continual learning.

The paper develops an amortized variational inference method to estimate latent point process intensities from replicated event patterns.

The paper analyzes the spectral properties and limits of multinomial and residual resampling in high-dimensional particle filtering.

The paper introduces a nested H-score method for distributed feature extraction that avoids redundancy with receiver-side auxiliary information.

The paper proposes an amortized adaptive importance sampling method using a lightweight neural network to iteratively update proposal distributions.

The paper transfers quadrupedal locomotion skills from single side-view videos to robots by pinning unobserved lateral motion to stance width.

The paper presents an invertible normalizing flow method for two-dimensional topographic manifold learning and data visualization.

The paper integrates matrix sketching via frequent directions into adaptive optimization methods to reduce preconditioning computation overhead.

The paper theoretically analyzes how one-layer transformers learn in-context nearest-neighbor prediction when using chain-of-thought reasoning.

The paper evaluates machine learning models on metadata and acoustic features to predict audio popularity on an online sharing platform.

The paper proposes a dynamic auxiliary search strategy to guide symbolic regression solvers by constructing a high-potential functional subspace.

The paper evaluates neural network architectures to downscale ERA5 atmospheric data to high-resolution satellite precipitation estimates across West Africa.

The paper develops a calibrated barycenter framework to accelerate multiobjective optimization over divergence matching problems.

The paper introduces a center-estimated neural posterior estimator to improve sample-size extrapolation in simulation-based inference.

The paper analyzes when feature standardization improves risk bounds for rare subgroups under finite stochastic gradient training budgets.

The paper proposes a post-training tail correction method for diffusion and flow models using radial-angular extreme value decomposition.

The paper introduces GpFlash, an open library designed to speed up Gaussian process training and prediction while reducing memory footprint.

The paper develops a formal performance theory analyzing skill allocation and workflow design in joint human-machine systems.

The paper presents PivotRL, a system that dynamically shifts GPU resources among rollout, reward modeling, and training during post-training.

The paper demonstrates that small-batch stochastic gradient noise restores sequential feature learning in the Muon optimizer.

The paper analyzes sequence composition strategies during electronic health record foundation model pretraining to improve downstream clinical prediction.