61 papers
FedEMA-Distill is a robust and communication-efficient federated learning framework that leverages server-side exponential moving average smoothing and ensemble knowledge distillation from compressed client logits to achieve superior accuracy, faster convergence, and Byzantine resilience under non-IID data conditions without requiring client-side software modifications.
This paper introduces AttnBoost, an interpretable framework that integrates feature-level attention into gradient boosting to dynamically prioritize relevant variables like promotions and seasonality, thereby improving both sales forecasting accuracy and actionable insights for retail supply chain management.
This paper introduces EDINET-Bench, a challenging open-source benchmark derived from ten years of Japanese financial reports to evaluate LLMs on complex tasks like fraud detection and earnings forecasting, revealing that current models struggle significantly without specialized scaffolding and highlighting the need for more realistic evaluation frameworks.
The paper introduces GIT-BO, a high-dimensional Bayesian optimization framework that leverages the zero-shot inference capabilities of the TabPFN v2 tabular foundation model combined with an active-subspace mechanism to achieve superior performance and runtime efficiency across diverse benchmarks up to 500 dimensions without requiring online retraining.
This paper presents a high-resolution, US-scale digital similar that fuses diverse datasets to model interactions between livestock, wild birds, and humans, enabling the evaluation and validation of spillover risks for highly pathogenic avian influenza (H5N1) to guide targeted surveillance efforts.
This paper addresses the problem of set-membership localization using noisy range measurements by characterizing the nonconvex set of consistent points and developing efficient convex programming methods to compute tight, guaranteed outer-bounding boxes or ellipsoids, as well as inner approximations, without relying on semidefinite programming relaxations.
The paper introduces FMint-SDE, a multimodal foundation model based on decoder-only transformers that leverages in-context learning and error correction to provide a general-purpose, efficient, and accurate solution for simulating stochastic differential equations across diverse scientific domains.
This paper argues that as embodied AI crosses critical thresholds in dexterity, generalization, and reliability, it will trigger a phase transition in economic geography by dismantling the century-old Fordist paradigm of centralized mega-factories and replacing labor-driven site selection with a new topology defined by demand-proximal micro-manufacturing and machine-optimal environmental conditions.
This paper proposes a privacy-preserving, offline Central Bank Digital Currency (CBDC) payment model for resource-constrained IoT devices that integrates Secure Elements, lightweight Zero-Knowledge Proofs, and intermittent synchronization to enable secure, cash-like transactions while preventing double-spending and ensuring AML/CFT compliance without continuous internet connectivity.
This paper establishes a biorthogonal harmonic analysis framework for non-normal directed graphs that utilizes dual eigenbases to define an exact energy-preserving transform, derives variational bounds and reconstruction guarantees controlled by eigenvector geometry, and validates these theoretical results through simulations linking filtering robustness to the graph's departure from normality.
This paper proposes a reinforcement learning framework using Proximal Policy Optimization (PPO) to dynamically weight LLM-generated trading alphas, demonstrating that while it may not always maximize cumulative returns, it significantly improves risk-adjusted performance and stability compared to traditional baseline strategies.
This paper proposes and validates a dynamic cryptocurrency portfolio optimization strategy that integrates technical indicators (RSI and SMA) with AI-driven sentiment analysis (VADER and Google Gemini) into a mean-variance framework, demonstrating superior risk-adjusted returns and cumulative growth compared to traditional benchmarks while highlighting the need for enhanced risk management to mitigate drawdowns during market stress.
This paper proposes a novel framework that leverages large language models to automatically generate interpretable, sentiment-aware formulaic alphas, which serve as enhanced features for Transformer-based models to significantly improve stock price prediction accuracy and transparency.
This paper proposes a novel parameter-scaled continuum surface flux (CSF) approach that significantly improves the accuracy of melt pool temperature predictions and reduces computational costs in metal additive manufacturing simulations by overcoming the limitations of classical CSF methods regarding extreme temperature gradients and material property ratios.
This paper validates the legitimacy and effectiveness of a newly proposed high-frequency fast direct solver for electromagnetic integral equations on non-canonical geometries by leveraging semiclassical microlocal results to analyze its asymptotic spectral properties.
This paper introduces a Fourier Neural Operator-based surrogate model that rapidly and accurately predicts angle- and energy-resolved proton transport and neutron production in proton therapy, achieving Monte Carlo-level accuracy within seconds to enable real-time adaptive range verification and neutron dose estimation.
This paper proposes an improved objective function for the optAPM code, featuring a simplified hotspot cost formulation and pre-interpolation of trail data, to reduce modeling errors and enhance the accuracy of absolute plate motion reconstructions over geological timescales.
MOOSE-Star is a unified framework that overcomes the mathematical intractability of directly training scientific discovery models by decomposing the generative reasoning process into tractable subtasks and employing motivation-guided hierarchical search, thereby enabling scalable training and continuous test-time scaling while reducing complexity from exponential to logarithmic.
SpecBridge is a novel framework that improves small-molecule identification from mass spectrometry by fine-tuning a spectral encoder to align with a frozen molecular foundation model's latent space, achieving significant accuracy gains over existing baselines while maintaining parameter efficiency.
This paper extends the CNF framework to solve coupled and non-linear Partial Differential Equations for forward, inverse, and discovery tasks, while providing a comprehensive survey, implementation, and evaluation of various neural PDE solvers to address challenges like the curse of dimensionality and high computational costs.