6592 papers
This paper introduces a normalized confidence scoring framework based on output anchor tokens to detect LLM errors without external validation, revealing that while supervised fine-tuning yields well-calibrated confidence, reinforcement learning methods induce overconfidence, and proposing post-RL self-distillation to restore reliability for applications like adaptive retrieval-augmented generation.
The paper introduces Structure-Aware Set Transformers (STAR), a novel architecture that enhances asynchronous clinical time series modeling by integrating parameter-efficient soft attention biases for temporal locality and variable-type affinity, thereby outperforming existing grid-based and set-based baselines on ICU prediction tasks while providing interpretable insights into temporal and variable interactions.
LegoNet is a post-training compression technique that clusters 4x4 weight blocks across entire neural network architectures to achieve memory footprint reductions of up to 128x with negligible accuracy loss, without requiring any retraining or architectural modifications.
This paper addresses the lack of systematic evaluation in Multi-Agent Deep Reinforcement Learning for C-V2X resource allocation by introducing a disentangled benchmark suite of interference games and diverse datasets to isolate specific challenges, ultimately identifying policy robustness and generalization across vehicular topologies as the primary hurdle and demonstrating the superiority of actor-critic methods over value-based approaches.
To address the complexity gap between StarCraft II's full game and its mini-games, this paper introduces the Two-Bridge Map Suite, an open-source, lightweight benchmark that isolates tactical navigation and combat skills to enable accessible reinforcement learning research under realistic compute budgets.
This paper introduces a practical method for obtaining finite-sample valid p-values for feature-level hypothesis testing in tabular data by combining the Conditional Randomization Test with the TabPFN foundation model, enabling statistical inference without model retraining or parametric assumptions.
This paper introduces CapTrack, a capability-centric framework that redefines LLM forgetting as systematic behavioral drift rather than mere knowledge loss, revealing through a large-scale study that post-training significantly degrades robustness and default behaviors, with instruction fine-tuning causing the most pronounced effects.
This paper introduces DeepScope, a deep learning-based system that analyzes microscopic images of unincubated water samples to detect fecal contamination in seconds with 93% accuracy and a cost of $0.44 per test, significantly outperforming traditional incubation methods in speed and affordability.
The paper demonstrates that unlike in domains with external verifiers, scaling inference compute through crowd wisdom strategies fails to improve LLM truthfulness in unverified settings because correlated model errors and the inability to distinguish social prediction from truth verification cause aggregation to reinforce shared misconceptions rather than identify correct answers.
The paper introduces OptiRoulette, a stochastic meta-optimizer that dynamically selects update rules from a pool during training, demonstrating significantly faster convergence and higher test accuracy across multiple image-classification benchmarks compared to a standard AdamW baseline.
This paper proposes a unified linear representation for diffusion models and flow matching to theoretically demonstrate that the often weak correlation between noisy data and predicted targets in existing methods may adversely impact the learning process.
This paper proposes Annealed Co-Generation (ACG), a framework that replaces high-dimensional joint diffusion modeling with a low-dimensional, pairwise approach coupled through a three-stage annealing process to achieve efficient and consistent multivariate co-generation for scientific applications like flow-field completion and antibody generation.
RACER is a novel, model-agnostic routing framework that addresses misrouting risks in multi-LLM systems by formulating the problem as -VOR to generate calibrated, variable-sized model sets with rigorous distribution-free risk control, thereby enhancing downstream accuracy and cost-performance trade-offs.
The paper introduces Evo, a novel large language model that unifies autoregressive and diffusion-based generation within a continuous evolutionary latent framework, enabling adaptive balancing of planning and refinement to achieve state-of-the-art performance across diverse benchmarks while maintaining fast inference speeds.
This paper proposes a novel topology-aware framework that leverages domain-specific foundation models, a graph tokenizer for multiplex connectivity, and knowledge distillation to achieve robust zero-shot interaction prediction in multiplex biological networks, outperforming state-of-the-art methods.
The paper introduces NAT (Not All Tokens Are Needed), a token-efficient reinforcement learning framework that utilizes unbiased partial-token gradient estimation via Horvitz-Thompson reweighting to achieve full-sequence performance with significantly reduced compute and memory costs by updating policies on only a subset of generated tokens.
GraphSkill is an agentic framework that improves complex graph reasoning by leveraging hierarchical document retrieval and self-debugging with generated test cases, validated on a new comprehensive dataset.
This paper reveals that state-of-the-art Process Reward Models (PRMs) are systematically exploitable by adversarial optimization, functioning primarily as fluency detectors rather than reasoning verifiers due to a critical dissociation between stylistic changes and ground-truth accuracy, prompting the release of a diagnostic framework and benchmark to address these vulnerabilities.
This paper empirically demonstrates that a Transformer model utilizing self-attention mechanisms outperforms traditional ARIMA and recurrent neural network approaches (LSTM, BiLSTM) in short-term power load forecasting on PJM data, achieving a superior 3.8% MAPE and highlighting the effectiveness of attention-based architectures for capturing complex temporal patterns.
This survey comprehensively reviews the methodological advances and diverse applications of Flow-GRPO, a framework that extends Group Relative Policy Optimization to align large-scale flow matching models with human preferences across various generative tasks and modalities.