Yang Wang

FedFG introduces a robust federated learning framework using flow-matching generation to simultaneously enhance client privacy and defend against sophisticated poisoning attacks.

The paper introduces Dummy-Aware Weighted Attack (DAWA), a novel evaluation method that significantly reduces the reported robustness of Dummy Classes-based defenses by simultaneously targeting both the true and dummy class labels.

The paper introduces ProjLens, an interpretability framework that reveals that backdoor vulnerabilities in Multimodal Large Language Models (MLLMs) are encoded within a low-rank subspace of the projector, causing a measurable semantic shift in poisoned inputs.

This study investigated user reactions to inferred personal information from their own ChatGPT histories, finding that acceptability is governed by context-sensitive norms regarding generation, retention, and transmission, rather than just the inference content.

The paper proposes a universal graph backdoor defense framework that addresses feature-based graph backdoor attacks, which are more challenging than traditional subgraph-based attacks, by leveraging local feature consistency.

FedMPT introduces a novel federated learning framework for Multi-Label Recognition (MLR) using Vision-Language Models (VLMs) by leveraging generalizable conditions to mitigate label overfitting and improve robustness.

The paper introduces Harness-Bench, a diagnostic benchmark that measures how different system 'harnesses' affect LLM agent performance in realistic workflows, showing that agent capability must be reported at the model-harness configuration level.

The paper introduces an operator-level factorial benchmark for molecular MPNNs, finding that message construction (specifically concatenation-based mixing) is the primary determinant of performance, rather than the complexity of the node update mechanism.

DeMaVLA is a generalizable Vision-Language-Action foundation model designed for deformable object manipulation, achieving strong real-world performance on folding tasks by leveraging large-scale real-world data and corrective learning.

Weak self-training on synthetic data can amplify a language model's existing capabilities, but this effect is strictly dependent on the compatibility between the source and student models, not on the data's intrinsic quality.

The paper introduces MechVQA, a comprehensive dataset and benchmark for mechanical drawing understanding, and proposes the MechVL model, which significantly improves Multimodal LLMs' performance on these specialized tasks.

The paper addresses the reliability of open-weight LLMs for power system code generation by identifying structured API-knowledge boundary errors and proposing a boundary-aware intervention that significantly boosts accuracy without fine-tuning.

The paper introduces a new benchmark for multi-target cross-lingual summarization (MTXLS) and proposes an activation steering method that significantly improves LLM performance by guiding the generation process using English representations.

The paper introduces TabPrep, a feature engineering pipeline that systematically improves performance across various tabular machine learning models by addressing structural data patterns ignored by current benchmarks.

HumanNOVA introduces a photorealistic, universal, and rapid model capable of generating high-quality 3D human avatars from a single input RGB image.

MOSS-Audio is a unified audio-language model designed for comprehensive understanding of speech, environmental sounds, and music, achieving strong performance across various audio-grounded tasks.