MVFA: A Multi-View Text-Guided Multimodal Fusion LLM Adapter for Sentiment Analysis and Emotion Recognition
为解决多模态情感分析和情绪识别中跨模态融合的问题，提出MVFA框架，通过构建多视角文本引导的多模态融合，并在多个冻结的大语言模型上验证其有效性。
0 citationsRead paper
Selected work
Representative Papers
Breadth Beats Depth: Improving GCG-Based Jailbreak Optimization with Breadth-Oriented Suffix Search
该研究通过引入BOSS框架改进GCG方法，利用广度优先的后缀搜索策略解决现有方法过度关注易破解行为的问题，提高攻击成功率并减少优化时间。
0 citationsRead paper
OASIS: Optimizing Attacker Sequences for Hard-Label Black-Box Text Attacks
本文提出OURS方法，通过优化攻击序列来解决硬标签黑盒文本攻击问题，平衡攻击成功率与扰动，并在多个数据集上优于基线方法。
0 citationsRead paper
A Design Concept of Forecasting Software for Normalized Vector Autoregressions with Fat Tails and Stochastic Volatility
该论文提出了一套R软件包，利用先进的贝叶斯等模型解决宏观经济预测问题，并通过C++高效算法提高计算效率。
0 citationsRead paper
BrainLinear: A Linear Model for Brain Network Analysis in Sparse Tangent Subspaces
This study addresses model redundancy and the lack of connection-level interpretability in brain functional connectivity analysis by proposing a geometry-aware framework. By mapping connectivity matrices onto the tangent space of the Fréchet mean and selecting critical directions, the method employs a lightweight MLP for efficient classification. Empirical results demonstrate that sparse linear modeling within the tangent subspace effectively replaces complex interactions, significantly improving AUC and ACC while reducing runtime and GPU memory usage by 84% and 68.4%, respectively. By balancing high diagnostic performance with biological interpretability, this approach establishes an efficient paradigm for brain disease diagnosis.
0 citationsRead paper
Recent publications
Latest Papers
MVFA: A Multi-View Text-Guided Multimodal Fusion LLM Adapter for Sentiment Analysis and Emotion Recognition
为解决多模态情感分析和情绪识别中跨模态融合的问题，提出MVFA框架，通过构建多视角文本引导的多模态融合，并在多个冻结的大语言模型上验证其有效性。
0 citationsRead paper
Breadth Beats Depth: Improving GCG-Based Jailbreak Optimization with Breadth-Oriented Suffix Search
该研究通过引入BOSS框架改进GCG方法，利用广度优先的后缀搜索策略解决现有方法过度关注易破解行为的问题，提高攻击成功率并减少优化时间。
0 citationsRead paper
OASIS: Optimizing Attacker Sequences for Hard-Label Black-Box Text Attacks
本文提出OURS方法，通过优化攻击序列来解决硬标签黑盒文本攻击问题，平衡攻击成功率与扰动，并在多个数据集上优于基线方法。
0 citationsRead paper
A Design Concept of Forecasting Software for Normalized Vector Autoregressions with Fat Tails and Stochastic Volatility
该论文提出了一套R软件包，利用先进的贝叶斯等模型解决宏观经济预测问题，并通过C++高效算法提高计算效率。
0 citationsRead paper
BrainLinear: A Linear Model for Brain Network Analysis in Sparse Tangent Subspaces
This study addresses model redundancy and the lack of connection-level interpretability in brain functional connectivity analysis by proposing a geometry-aware framework. By mapping connectivity matrices onto the tangent space of the Fréchet mean and selecting critical directions, the method employs a lightweight MLP for efficient classification. Empirical results demonstrate that sparse linear modeling within the tangent subspace effectively replaces complex interactions, significantly improving AUC and ACC while reducing runtime and GPU memory usage by 84% and 68.4%, respectively. By balancing high diagnostic performance with biological interpretability, this approach establishes an efficient paradigm for brain disease diagnosis.
0 citationsRead paper