SearchWiki: Learning to Build and Navigate Knowledge Wikis for Active Information Seeking
为解决信息检索中忽略文档结构的问题,提出SearchWiki框架构建分层知识维基,并训练WikiResearcher-9B模型通过多轮工具使用进行有效导航和检索。
为解决信息检索中忽略文档结构的问题,提出SearchWiki框架构建分层知识维基,并训练WikiResearcher-9B模型通过多轮工具使用进行有效导航和检索。
This study addresses the challenge of detecting sophisticated cyber threats—such as DDoS attacks, brute-force attempts, and web-based intrusions—by proposing a novel intrusion detection approach that integrates topological data analysis (TDA) with long short-term memory (LSTM) networks. The method leverages persistent homology to extract topological features from network traffic, including Betti curves and persistence diagrams, and for the first time embeds these features into a deep learning framework where they are complementarily fused with LSTM-captured temporal dependencies. Evaluated on the CIC-IDS2017 dataset using five-fold cross-validation, the proposed model achieves perfect performance with both AUC and F1-score of 1.000, significantly outperforming baseline models such as TDA combined with random forest and isolation forest, thereby demonstrating the efficacy and superiority of jointly modeling topological and temporal characteristics for intrusion detection.
This study addresses the challenge of stock return prediction in non-stationary financial markets by proposing an adaptive Financial Transformer model. The architecture incorporates a market state encoder and an adaptive gating attention mechanism to dynamically integrate eleven semantically grouped financial indicators, thereby modulating temporal modeling according to prevailing market conditions. A finance-aware composite objective function is introduced to jointly optimize prediction error, directional accuracy, and the Sharpe ratio, while correcting sequence alignment bias commonly present in backtesting procedures. Empirical evaluations demonstrate that the proposed method achieves competitive predictive performance across multiple stocks, reduces model complexity by 15.2%, and significantly enhances both parameter efficiency and the interpretability of trading signals.
本文提出了一种物理约束的深度神经网络模型,用于预测二元混合物中的相分离演化,通过直接在输出上施加守恒约束,实现了长时间稳定和准确预测。
This study addresses the vulnerability of traditional momentum strategies to the “winner’s curse,” which often leads to severe drawdowns during market reversals. To mitigate this issue, the authors propose the AEGIS framework, which innovatively decouples momentum signals from risk exposure. The approach employs a volatility-adjusted momentum filter to identify robust trends, integrates a minimax correlation algorithm for structured diversification, and dynamically optimizes the Sortino ratio via sequential least squares programming (SLSQP) to adapt to varying market regimes. Backtesting over a 20-year period demonstrates that the strategy matches the returns of the Nasdaq-100 while significantly reducing downside volatility and generating substantial alpha relative to the S&P 500, thereby achieving dual objectives: efficient participation in bull markets and effective drawdown control in bear markets.
为解决信息检索中忽略文档结构的问题,提出SearchWiki框架构建分层知识维基,并训练WikiResearcher-9B模型通过多轮工具使用进行有效导航和检索。
This study addresses the challenge of detecting sophisticated cyber threats—such as DDoS attacks, brute-force attempts, and web-based intrusions—by proposing a novel intrusion detection approach that integrates topological data analysis (TDA) with long short-term memory (LSTM) networks. The method leverages persistent homology to extract topological features from network traffic, including Betti curves and persistence diagrams, and for the first time embeds these features into a deep learning framework where they are complementarily fused with LSTM-captured temporal dependencies. Evaluated on the CIC-IDS2017 dataset using five-fold cross-validation, the proposed model achieves perfect performance with both AUC and F1-score of 1.000, significantly outperforming baseline models such as TDA combined with random forest and isolation forest, thereby demonstrating the efficacy and superiority of jointly modeling topological and temporal characteristics for intrusion detection.
This study addresses the challenge of stock return prediction in non-stationary financial markets by proposing an adaptive Financial Transformer model. The architecture incorporates a market state encoder and an adaptive gating attention mechanism to dynamically integrate eleven semantically grouped financial indicators, thereby modulating temporal modeling according to prevailing market conditions. A finance-aware composite objective function is introduced to jointly optimize prediction error, directional accuracy, and the Sharpe ratio, while correcting sequence alignment bias commonly present in backtesting procedures. Empirical evaluations demonstrate that the proposed method achieves competitive predictive performance across multiple stocks, reduces model complexity by 15.2%, and significantly enhances both parameter efficiency and the interpretability of trading signals.
本文提出了一种物理约束的深度神经网络模型,用于预测二元混合物中的相分离演化,通过直接在输出上施加守恒约束,实现了长时间稳定和准确预测。
This study addresses the vulnerability of traditional momentum strategies to the “winner’s curse,” which often leads to severe drawdowns during market reversals. To mitigate this issue, the authors propose the AEGIS framework, which innovatively decouples momentum signals from risk exposure. The approach employs a volatility-adjusted momentum filter to identify robust trends, integrates a minimax correlation algorithm for structured diversification, and dynamically optimizes the Sortino ratio via sequential least squares programming (SLSQP) to adapt to varying market regimes. Backtesting over a 20-year period demonstrates that the strategy matches the returns of the Nasdaq-100 while significantly reducing downside volatility and generating substantial alpha relative to the S&P 500, thereby achieving dual objectives: efficient participation in bull markets and effective drawdown control in bear markets.