4748 papers
This paper introduces -GaBO, a novel family of Bayesian optimization algorithms that leverages information geometry to construct Matérn kernels and geometric optimizers tailored for the probability simplex, demonstrating superior performance over constrained Euclidean approaches in optimizing mixtures and robotic control tasks.
This paper introduces In-Context RLVR, a method that leverages a model's own in-context learning ability to measure "Demonstration Utility" via Evidence Gain, thereby implicitly reweighting rewards to prioritize high-quality reasoning traces over merely correct but flawed solutions during Reinforcement Learning with Verifiable Rewards training.
This paper introduces the smoothing pseudo-projector, a lightweight, multigrid-inspired module that corrects hidden representations in transformer-based models to suppress noise from label-irrelevant inputs, thereby improving training dynamics and robustness without altering the core architecture.
This paper proposes a novel hierarchical multi-task multi-fidelity (H-MT-MF) framework for Gaussian process-based surrogate modeling that unifies inter-task information sharing and fidelity-dependent uncertainty handling to significantly improve prediction accuracy and data efficiency in manufacturing systems with heterogeneous data sources.
This paper introduces HR-GAT, a hierarchical resolution graph attention network that leverages geospatial data to predict spectrum demand with 21% higher accuracy than baseline models, effectively addressing spatial autocorrelation challenges to enable more efficient spectrum sharing and policy-making.
The paper proposes GAST, a novel Parameter-Efficient Fine-Tuning method that unifies data-layer selection and layer-sparse strategies to adaptively match impactful data points with specific model layers, thereby overcoming the limitations of existing single-dimension approaches and achieving superior performance.
The paper introduces CarbonBench, the first standardized benchmark comprising over 1.3 million global observations from 567 sites, designed to rigorously evaluate and compare zero-shot spatial transfer learning methods for upscaling terrestrial carbon fluxes across diverse, unseen ecosystems and climate regimes.
The paper proposes MSSR, a memory-aware adaptive replay framework that estimates sample-level memory strength to dynamically schedule rehearsal intervals, effectively mitigating catastrophic forgetting while maintaining fast adaptation in continual LLM fine-tuning.
This paper introduces OptEMA, a novel adaptive Exponential Moving Average optimizer that achieves nearly optimal convergence rates in both stochastic and zero-noise regimes without requiring prior knowledge of Lipschitz constants or manual hyperparameter tuning.
This paper establishes that generative drifting is theoretically equivalent to score matching under Gaussian kernels, providing a spectral and variational framework that explains the empirical superiority of Laplacian kernels, proposes an exponential bandwidth annealing schedule to accelerate convergence, and proves the necessity of the stop-gradient operator through its connection to Wasserstein gradient flows.
The paper introduces SignalMC-MED, a comprehensive benchmark utilizing 22,256 synchronized single-lead ECG and PPG visits to evaluate biosignal foundation models across 20 clinical tasks, demonstrating that domain-specific models with multimodal fusion and full-duration signals outperform general time-series approaches while revealing that larger model sizes do not guarantee superior performance.
This paper introduces the Overfitting-Underfitting Indicator (OUI) as an efficient, early-stage metric based on hidden neuron activation patterns to distinguish optimal learning rates in PPO actor-critic training, demonstrating its superior ability to prune unpromising runs compared to traditional criteria by revealing distinct structural signatures in actor and critic networks.
This paper introduces "neural debuggers," a new class of language models that emulate traditional debugging interactions like setting breakpoints and stepping through code to enable both forward and inverse execution prediction, thereby laying the foundation for more powerful agentic coding systems and automated debugging.
This paper introduces a family of mean-normalized matrix operator norms to derive width-independent smoothness bounds for deep neural networks, leading to the development of MOGA, a row/column-normalized optimizer that enables stable hyperparameter transfer across model widths and outperforms Muon in speed while maintaining competitive performance.
The paper proposes C2FMAE, a coarse-to-fine masked autoencoder that resolves the tension between global semantics and local details in self-supervised learning by employing a cascaded decoder and progressive masking curriculum on a newly constructed multi-granular dataset to achieve hierarchical visual understanding and superior performance across various vision tasks.
This paper demonstrates that reasoning consistently enhances honesty in large language models by guiding them through a representational space where deceptive states are metastable and easily disrupted, ultimately nudging the models toward more stable, honest defaults.
This paper challenges the standard view of superposition in neural networks by demonstrating that, unlike in idealized uncorrelated settings where interference is merely noise, realistic feature correlations allow models to arrange features so that interference becomes constructive, thereby naturally forming the semantic clusters and cyclical structures observed in real language models.
The paper introduces Task-Aware Modulation with Representation Learning (TAM-RL), a novel framework that combines spatio-temporal representation learning with physically grounded constraints to significantly improve the accuracy and generalizability of global terrestrial carbon flux estimates compared to existing state-of-the-art methods.
This paper introduces two online neural network-based algorithms for change-point detection in large time series that demonstrate linear computational complexity, outperform existing methods on various datasets, and are proven to converge to optimal solutions under specific conditions.
This paper proposes a novel extension to semi-parametric Bayesian additive regression trees (BART) that resolves non-identifiability and bias issues by modifying tree-generation moves to allow shared covariates between linear and non-parametric components, thereby enabling the modeling of complex interactions while maintaining competitive performance across simulation and real-world applications.