3802 papers
This study demonstrates that quantum entanglement serves as a functional resource in competitive reinforcement learning, enabling hybrid quantum-classical agents trained on the game Pong to consistently outperform separable quantum circuits and match or exceed classical baselines by learning structurally distinct features that better model dynamic agent interactions.
This paper introduces HyMEM, a hybrid self-evolving structured memory system that combines discrete symbolic nodes with continuous embeddings in a graph format to significantly enhance the performance of open-source GUI agents, enabling them to match or surpass strong closed-source models on complex, long-horizon tasks.
GaLoRA is a parameter-efficient framework that integrates structural information into large language models for text-attributed graph node classification, achieving state-of-the-art performance with only 0.24% of the parameters required for full fine-tuning.
This paper introduces a regime-aware in-context learning framework that leverages pretrained large language models to forecast financial volatility by dynamically adapting to nonstationary market conditions through oracle-guided, regime-specific demonstrations without requiring parameter fine-tuning.
This paper introduces a search procedure to identify near-optimal learning rate schedule shapes across various workloads, revealing that while warmup and decay are robust features, commonly used schedules are suboptimal and the ideal shape is significantly influenced by hyperparameters like weight decay.
This paper introduces a flexible stochastic beam search sampling method for masked language models that optimizes protein properties by evaluating entire-sequence neighborhoods, demonstrating through extensive in silico and in vitro antibody engineering experiments that the choice of sampling strategy is at least as critical as the model itself.
This paper introduces a data-driven integration kernel framework that enhances the interpretability and efficiency of nonlocal operator learning in climate modeling by separating nonlocal information aggregation via learnable weighting functions from local nonlinear prediction, thereby achieving competitive performance with fewer parameters and clearer physical insights.
NasoVoce is a nose-mounted speech interface that fuses acoustic and vibration signals to enable robust, discreet, and always-available voice interaction for AI, effectively overcoming the limitations of existing silent and whispered speech recognition methods.
This paper introduces FairFAL, an adaptive federated active learning framework that leverages lightweight prediction discrepancy and prototype-guided pseudo-labeling to dynamically select between global and local query models, effectively addressing the challenges of extreme non-IID data and global class imbalance to achieve superior performance over state-of-the-art methods.
This paper addresses the fixed-budget best-arm identification problem in non-stationary linear bandits by establishing a tighter, arm-set-dependent lower bound on error probability and proposing the algorithm, which utilizes an Adjacent-optimal design to achieve minimax-optimal performance that fully leverages the geometric structure of the arm set.
HEAL is an RL-free distillation framework that overcomes the "Teacher Ceiling" of standard rejection sampling by synergizing entropy-guided repair, uncertainty filtering, and progressive curriculum learning to transfer reasoning capabilities from Large Reasoning Models to smaller students.
This paper introduces Causal Concept Graphs (CCG), a framework that combines task-conditioned sparse autoencoders with differentiable structure learning to map causal dependencies between interpretable latent features in LLMs, demonstrating through the Causal Fidelity Score that graph-guided interventions significantly enhance stepwise reasoning performance compared to existing tracing and random baselines.
This paper establishes a new scaling law for Mixture-of-Experts models by deriving an explicit power-law formula for the optimal compute allocation ratio between expert and attention layers, enabling more efficient model design under fixed computational budgets.
This paper proposes a variance-aware adaptive weighting strategy that dynamically adjusts training weights based on loss variance across noise levels to address imbalanced training dynamics in diffusion models, resulting in improved generative performance and training stability on CIFAR datasets.
This paper introduces Graph-GRPO, an online reinforcement learning framework that enhances Graph Flow Models through analytical transition probabilities and a localized refinement strategy, achieving state-of-the-art performance in graph generation and molecular optimization tasks.
This paper analyzes the learning dynamics of two-layer over-parameterized linear networks under label noise SGD, revealing a two-phase process where noise drives the transition from the lazy to the rich regime to improve generalization, a mechanism that also extends to Sharpness-Aware Minimization (SAM).
This paper introduces PP-LUCB, a cost-efficient algorithm that optimally combines biased LLM-generated proxy scores with selective human audits to identify the best service system configuration while providing statistically valid confidence guarantees and significantly reducing audit costs.
The paper introduces STemDist, the first dataset distillation method designed for spatio-temporal forecasting that simultaneously compresses both spatial and temporal dimensions through a hybrid cluster-level and subset-based approach, achieving significantly faster training, reduced memory usage, and lower prediction errors compared to existing methods.
This paper proposes a domain-adaptive framework featuring degradation-stage synchronized batch sampling and a cross-domain aligned fusion large autoencoder to overcome distribution mismatches and temporal dependency limitations in health indicator learning, achieving significant performance improvements on industrial datasets.
This paper proposes Weighted improved Greedy Sampling (WiGS), a reinforcement learning-based active learning framework that dynamically adapts the balance between feature-space diversity and output-space uncertainty to outperform static multiplicative methods, particularly in domains with irregular data density.