GONE: Structural Knowledge Unlearning via Neighborhood-Expanded Distribution Shaping
本文针对大语言模型中知识遗忘问题,提出GONE基准及NEDS框架,有效处理结构化数据中的直接事实移除、推理泄露和灾难性遗忘。
本文针对大语言模型中知识遗忘问题,提出GONE基准及NEDS框架,有效处理结构化数据中的直接事实移除、推理泄露和灾难性遗忘。
This paper characterizes the structure of split comparability graphs and establishes an upper bound on their permutation representation number. We first provide an exact combinatorial characterization—via a necessary and sufficient condition on vertex labelings—yielding the first precise structural description of this graph class. Building on this, we prove that the permutation representation number of any split comparability graph is at most three. As a corollary, the dimension of any split poset is at most three, and we supply a purely combinatorial proof independent of the Dushnik–Miller theorem. Our approach integrates split graph decomposition, transitive orientations, poset dimension theory, and permutation graph representation techniques. The key innovation lies in establishing a direct correspondence between vertex labelings and comparability structure, thereby unifying the interpretation of permutation representation number and poset dimension. This resolves a fundamental gap in the representation theory of split graphs.
Early detection of drought stress is critical for minimizing crop losses, yet subtle phenotypic changes necessitate non-invasive aerial imaging and advanced modeling. This paper proposes an interpretable Vision Transformer (ViT)-driven framework tailored for potato crops. We introduce two novel architectures: a ViT-SVM hybrid model and an end-to-end ViT classifier—the first integration of ViT with SVM for agricultural stress recognition. Leveraging transfer learning and multispectral/RGB drone imagery, our method localizes key stress indicators—including leaf wilting and canopy texture degradation—via attention maps. Experimental results demonstrate significant improvements in detection accuracy and provide full interpretability of model decisions through visualized attention mechanisms. The framework enables real-time, trustworthy drought monitoring and management in field conditions. (136 words)
MightyPPL工具扩展了MITL模型检测,首次支持MTL属性及Pnueli和Past模态,通过改进架构实现了更优性能。
该研究通过引入AlphaRJM,利用Reward-Jump Memory和基于SDE的回报评估器解决公式alpha发现中反馈延迟问题,提高股票预测性能。
MightyPPL工具扩展了MITL模型检测,首次支持MTL属性及Pnueli和Past模态,通过改进架构实现了更优性能。
该研究通过引入AlphaRJM,利用Reward-Jump Memory和基于SDE的回报评估器解决公式alpha发现中反馈延迟问题,提高股票预测性能。
本文提出了一种针对右删失数据的时间依赖双向部分AUC和部分Youden指数的非参数估计方法,以评估生物标志物的预测性能。
研究使用GPT-5.5评估了10364页手写物理答案的评分,通过两轮改进后与官方评分高度一致,适用于辅助高风险考试评分。
研究开发了低资源语言Mizo的自动语音识别系统,通过收集17.62小时语音数据并使用Whisper和SraVaani 1.0模型进行微调,显著降低了词错误率。