11462 papers
This paper introduces a unified framework for ensembling diverse language models via -ensemble distributions and proposes a byte-level sequential Monte Carlo algorithm to sample from these distributions, effectively overcoming challenges like mismatched vocabularies and biased approximations to improve performance on structured text generation tasks.
The paper introduces OPSDC, an on-policy self-distillation method that trains reasoning models to generate more concise outputs by minimizing reverse KL divergence against their own "be concise" instructions, achieving significant token reduction and accuracy improvements on benchmarks like MATH-500 and AIME 2024 without requiring ground-truth answers or token budgets.
This paper introduces Latent Wasserstein Adversarial Imitation Learning (LWAIL), a novel framework that leverages a pre-trained, dynamics-aware latent space to achieve expert-level performance using only one or a few state-only expert demonstrations, thereby overcoming the data and action requirements of traditional imitation learning methods.
This paper proposes a physically constrained neural sequence learning framework that employs a Kraus-structured output layer to guarantee completely positive trace-preserving (CPTP) quantum state updates, demonstrating that a Kraus-LSTM architecture significantly outperforms unconstrained models and other backbones in reconstructing quantum trajectories under parameter drift.
This paper establishes a unified thermodynamic response framework for singular Bayesian models, demonstrating that posterior tempering induces a hierarchy of observables that naturally interpret complex learning-theoretic quantities like the real log canonical threshold and WAIC as free-energy derivatives, thereby revealing phase-transition-like structural reorganizations in models such as neural networks and Gaussian mixtures.
This paper introduces SurvHTE-Bench, the first comprehensive benchmark for evaluating heterogeneous treatment effect estimation in survival analysis, featuring a diverse suite of synthetic, semi-synthetic, and real-world datasets to enable rigorous, fair, and reproducible comparisons of causal survival methods.
This paper demonstrates that reasoning models often exhibit "performative chain-of-thought" by generating tokens without revealing their internal beliefs, yet activation probing can detect these hidden certainties early to enable significant token reduction while distinguishing genuine uncertainty in complex tasks.
This paper leverages Chinese open-weight LLMs that censor politically sensitive topics as a natural testbed to evaluate honesty elicitation and lie detection techniques, finding that methods like few-shot prompting and self-classification effectively increase truthful responses and detect falsehoods, though no approach completely eliminates deception.
This paper proposes a novel three-stage framework that combines inexpensive imperfect labels, supervised pretraining, and self-supervised refinement to achieve effective amortized optimization with significantly reduced costs and improved performance across challenging domains.
The paper introduces POET-X, a memory-efficient and scalable variant of the POET framework that utilizes optimized orthogonal equivalence transformations to enable the stable pretraining of billion-parameter large language models on a single GPU, overcoming the high memory and computational costs of the original implementation.
RoboPocket is a smartphone-based system that enhances robot imitation learning by using AR visual foresight to guide targeted data collection and asynchronous online finetuning, thereby doubling data efficiency and enabling instant policy iteration without requiring physical robot execution.
The paper proposes the Recurrent Action Transformer with Memory (RATE), a novel architecture that integrates a recurrent memory mechanism into transformers to overcome context length limitations in partially observable environments, demonstrating superior performance in memory-intensive offline reinforcement learning tasks while remaining competitive on standard benchmarks.
This paper introduces Crystal-GFN, a multi-environment, continuous-discrete GFlowNet that sequentially samples crystal structural attributes to efficiently generate diverse, valid materials with specific desirable properties and hard constraints, thereby accelerating the discovery of novel solid-state materials.
GeoTop is a mathematically principled framework that unifies Topological Data Analysis and Lipschitz-Killing Curvatures to resolve the diagnostic ambiguity of topologically equivalent structures by integrating robust topological signatures with precise geometric features, thereby achieving superior accuracy and interpretability in image classification tasks such as skin lesion diagnosis.
This paper presents a sample-optimal, locally differentially private algorithm for hypothesis selection that achieves the information-theoretic lower bound of using only rounds of interaction, thereby demonstrating the provable power of interactivity to overcome the sample complexity barrier inherent in non-interactive approaches.
This survey provides a comprehensive review and unified framework for Graph Neural Networks in EEG-based emotion recognition, categorizing existing methods by graph construction stages to offer guidance on their unique physiological foundations while outlining open challenges and future directions.
This paper investigates the applicability of classical generalization principles to list PAC learning, demonstrating that while uniform convergence remains equivalent to learnability, the sample compression conjecture fails as there exist list-learnable classes that cannot be compressed, even with arbitrarily large output lists.
This paper introduces the concept of agnostic tomography and presents an efficient algorithm that learns a stabilizer product state approximating an arbitrary quantum state as well as the best possible match within that class, running in polynomial time for constant fidelity.
This paper reviews and categorizes existing reward functions for reinforcement learning in autonomous driving into safety, comfort, progress, and traffic rule compliance, while highlighting their current limitations in standardization and context-awareness to propose future research directions for more robust and conflict-resolving reward designs.
This paper extends tracking guarantees for time-varying variational inequalities to non-monotone functions and periodic problems without sublinear solution paths, while also demonstrating that the associated discrete dynamical systems can exhibit either convergence or provably chaotic behavior.