10552 papers
This paper proves that standard RLHF alignment is inherently shallow because gradient signals vanish once a sequence's harmfulness is determined, and it proposes a recovery penalty objective to ensure alignment gradients persist throughout the entire generation process.
This paper introduces Osmosis Distillation, a novel model hijacking attack that exploits synthetic datasets generated by dataset distillation methods to compromise deep learning models in transfer learning with high success rates using only a few poisoned samples while maintaining utility on original tasks.
This paper introduces ReCouPLe, a lightweight framework that leverages natural language rationales as causal guidance to train reward models that are robust to spurious correlations and capable of zero-shot transfer to novel tasks, significantly outperforming baselines in reward accuracy and downstream policy performance under distribution shifts.
This paper presents a scalable, interpretable framework that achieves 80.4% accuracy in classifying eight professional baseball pitch types using only monocular 3D body kinematics, revealing that upper-body mechanics—particularly wrist position and trunk tilt—are the primary predictors while establishing an empirical ceiling for grip-based distinctions.
This paper introduces a feature-centric framework demonstrating that the noise required for differential privacy in two-layer neural networks degrades fairness and robustness by disrupting feature learning dynamics, as quantified by the feature-to-noise ratio, while also revealing the limitations of public pre-training strategies under distribution shifts.
FedAFD is a unified multimodal federated learning framework that enhances both client and server performance by employing a bi-level adversarial alignment and granularity-aware fusion for personalized local learning, alongside a similarity-guided ensemble distillation mechanism to effectively handle model heterogeneity and modality discrepancies.
This paper demonstrates that for high-dimensional random data, gradient descent on shallow ReLU networks exhibits an implicit bias that approximates the minimum -norm solution with high probability, bridging the gap between worst-case non-existence and exact orthogonality results through a novel primal-dual analysis.
This paper presents U-Parking, a distributed autonomous parking system that combines Ultra-Wideband-assisted robust localization with Large Language Model-driven intelligent planning to achieve reliable automated parking in challenging indoor environments, as demonstrated on real vehicles.
This paper introduces VPWEM, a non-Markovian visuomotor policy that combines a sliding window of recent observations with a Transformer-based episodic memory compressor to efficiently retain long-term context for robotic control, achieving significant performance improvements over state-of-the-art baselines on memory-intensive manipulation tasks while maintaining constant computational costs.
The paper introduces EVMbench, a benchmarking framework that evaluates the capabilities of frontier AI agents in detecting, patching, and exploiting smart contract vulnerabilities within a realistic local Ethereum environment, revealing their ability to successfully execute end-to-end attacks against live blockchain instances.
This paper introduces BandPO, a novel reinforcement learning algorithm that replaces PPO's fixed clipping mechanism with a dynamic, probability-aware operator to resolve the exploration bottleneck and entropy collapse caused by suppressing high-advantage low-probability actions, thereby achieving superior stability and performance across diverse models.
This paper proposes a Semantic Communication-Enhanced U-Shaped Split Federated Learning (SC-USFL) framework for vehicular networks that integrates a semantic communication module and a network status monitor to reduce communication overhead, enhance label privacy, and adaptively optimize transmission rates under dynamic channel conditions.
This paper addresses the challenge of person detection and tracking from an overhead crane LiDAR by curating a new annotated dataset, evaluating adapted 3D detectors like VoxelNeXt and SECOND with integrated tracking algorithms, and demonstrating high accuracy and real-time feasibility to bridge the gap between standard driving benchmarks and industrial overhead sensing.
This paper introduces -Reasoner, a novel framework that enhances LLM reasoning by integrating differentiable gradient descent on token logits during inference, thereby shifting from inefficient discrete search to efficient first-order optimization to achieve significant accuracy gains with reduced computational costs.
The paper introduces TimeWarp, a benchmark that evaluates web agents across evolving UI versions to expose their vulnerability to design changes, and proposes TimeTraj, a plan distillation algorithm that significantly improves agent robustness by training on trajectories collected from multiple web versions.
This study demonstrates that Transformer-based neural networks equipped with evidential output layers outperform LSTM and GRU models in predicting blood glucose and identifying adverse glycemic events for Type 1 diabetes by providing superior accuracy and well-calibrated uncertainty estimates validated on the HUPA-UCM dataset.
This paper introduces WaterSIC, a novel linear layer quantization algorithm that achieves information-theoretically near-optimal performance by allocating different quantization rates to weight columns via a waterfilling strategy, thereby significantly outperforming existing methods like GPTQ and establishing new state-of-the-art results for LLMs across 1 to 4-bit quantization rates.
This paper demonstrates that replaying generic pre-training data during fine-tuning significantly improves performance on target tasks by enhancing data efficiency and preventing catastrophic forgetting, even for less related domains.
The paper introduces Mixture of Universal Experts (MOUE), a novel Mixture-of-Experts architecture that scales model capacity by converting depth into "virtual width" through a universal expert pool shared across layers, utilizing a staggered rotational topology and specialized routing mechanisms to overcome scalability limits and outperform traditional MoE baselines.
This paper proposes a functionality-oriented model merging method that computes a weighted Karcher mean on the Fisher--Rao manifold via a practical fixed-point algorithm, effectively overcoming the representation collapse and scalability limitations of existing Euclidean-space approaches when combining multiple heterogeneous LLMs.