Yu Wu

The paper introduces SafeRedirect, a system-level defense that prevents frontier LLMs from generating harmful content during legitimate tasks that structurally require it, significantly reducing unsafe generation rates.

The paper proposes Family-Grouped Hierarchical Federated Learning (Family-FL) combined with a highly optimized Tiny CNN-LSTM model to enable privacy-preserving ECG monitoring on ultra-resource-constrained microcontrollers, significantly reducing communication overhead.

SARAD proposes a novel safety-aware hybrid framework that combines Large Language Models (LLMs) and Deep Reinforcement Learning (DRL) to improve autonomous driving decision-making by replacing random exploration with expert-guided decisions and adding collision prediction.

The paper distinguishes between a model's ability to generate useful updates for external agent components (harness-updating) and its ability to benefit from those updates (harness-benefit), finding that updating capabilities are surprisingly uniform while benefit is maximized in mid-tier models.

The paper introduces Contextual Belief Management (CBM) to address how LLMs should manage accumulating information over long interactions, showing that reinforcement learning significantly improves belief state accuracy.

AliMark proposes a novel watermarking framework that treats sentence-level watermarking as a bit sequence alignment problem, significantly enhancing robustness against structural text perturbations like sentence splitting and merging.

The paper introduces MusTBENCH, a new benchmark, and MusT, an optimization recipe, to rigorously test and improve the ability of Large Audio-Language Models (LALMs) to accurately ground their musical understanding in specific time segments of an audio track.

AliMark proposes a novel framework that enhances the robustness of sentence-level watermarking by reformulating the problem as a bit sequence encoding and alignment task, significantly improving resilience against structural text perturbations like sentence splitting and merging.

VISReg introduces a novel regularization technique that combines variance control with a Sliced-Wasserstein-based sketching objective to stabilize self-supervised learning, achieving state-of-the-art performance on out-of-distribution tasks.

The paper proposes OneReason, a framework that enhances the reasoning capability of generative recommendation models by focusing on improving item perception and structuring user behavior into coherent latent interests.