CryptoL: Towards Scale Dominance and Physics Constraints Mitigation in Financial Multivariate Time Series Forecasting
本文提出CryptoL框架，通过上下文归一化、适应性数值稳定及软可行性损失等方法解决加密货币预测中的规模异质性与物理约束问题。
0 citationsRead paper
Selected work
Representative Papers
Goat Optimization Algorithm: A Novel Bio-Inspired Metaheuristic for Global Optimization
To address the imbalance between exploration and exploitation and the tendency to converge prematurely to local optima in metaheuristic algorithms, this paper proposes the Goat Optimization Algorithm (GOA). Inspired by goats’ foraging behavior, parasite-avoidance strategies, and collective coordinated movement, GOA integrates three key mechanisms: adaptive foraging, elite-guided optimal-directional movement, and random jump-based escape. It further introduces a novel solution-filtering mechanism incorporating parasite avoidance to dynamically preserve population diversity and balance exploration versus exploitation. Evaluated on 23 standard benchmark functions, GOA demonstrates statistically significant improvements (Wilcoxon signed-rank test, *p* < 0.05) over PSO, GWO, GA, WOA, and ABC in convergence speed, global search capability, and solution accuracy. Results confirm GOA’s robustness in escaping local optima and its superior performance with statistical significance.
0 citationsRead paper
Intrusion Detection in IoT Networks Using Hyperdimensional Computing: A Case Study on the NSL-KDD Dataset
IoT edge devices face dual challenges of severe resource constraints and stringent real-time requirements for intrusion detection. Method: This paper pioneers the systematic application of Hyperdimensional Computing (HDC) to IoT intrusion detection, proposing a lightweight HDC framework that encodes NSL-KDD data into random high-dimensional vectors and employs binding and bundling operations to efficiently represent and classify normal traffic along with four attack types—DoS, Probe, R2L, and U2R. Integrated with tailored data preprocessing and class-balancing strategies, the framework achieves 99.54% accuracy on NSL-KDD. Contribution/Results: It significantly outperforms SVM, Random Forest, and shallow neural networks while requiring minimal parameters and exhibiting ultra-low inference latency. The model demonstrates strong generalization and enables real-time, on-device deployment—overcoming key limitations of conventional machine learning in edge-IoT scenarios.
0 citationsRead paper
Network Anomaly Detection for IoT Using Hyperdimensional Computing on NSL-KDD
Traditional intrusion detection systems (IDS) for IoT exhibit poor generalization on high-dimensional, complex network traffic and heavily rely on prior knowledge of known attacks. Method: This paper pioneers the application of Hyperdimensional Computing (HDC) to anomaly detection on the NSL-KDD dataset, proposing a lightweight HDC-based framework that integrates binary symbolic encoding, feature binding, and similarity matching—enabling efficient detection of both known and unknown attacks without training any classifier. Results: Evaluated on the KDDTrain+ subset, the method achieves 91.55% accuracy, significantly outperforming classical machine learning and deep learning baselines. By eliminating reliance on labeled attack samples, it enhances model robustness and generalization while maintaining computational efficiency. This work establishes a novel unsupervised/low-supervision detection paradigm tailored for resource-constrained IoT environments.
0 citationsRead paper
Recent publications
Latest Papers
CryptoL: Towards Scale Dominance and Physics Constraints Mitigation in Financial Multivariate Time Series Forecasting
本文提出CryptoL框架，通过上下文归一化、适应性数值稳定及软可行性损失等方法解决加密货币预测中的规模异质性与物理约束问题。
0 citationsRead paper
Goat Optimization Algorithm: A Novel Bio-Inspired Metaheuristic for Global Optimization
To address the imbalance between exploration and exploitation and the tendency to converge prematurely to local optima in metaheuristic algorithms, this paper proposes the Goat Optimization Algorithm (GOA). Inspired by goats’ foraging behavior, parasite-avoidance strategies, and collective coordinated movement, GOA integrates three key mechanisms: adaptive foraging, elite-guided optimal-directional movement, and random jump-based escape. It further introduces a novel solution-filtering mechanism incorporating parasite avoidance to dynamically preserve population diversity and balance exploration versus exploitation. Evaluated on 23 standard benchmark functions, GOA demonstrates statistically significant improvements (Wilcoxon signed-rank test, *p* < 0.05) over PSO, GWO, GA, WOA, and ABC in convergence speed, global search capability, and solution accuracy. Results confirm GOA’s robustness in escaping local optima and its superior performance with statistical significance.
0 citationsRead paper
Intrusion Detection in IoT Networks Using Hyperdimensional Computing: A Case Study on the NSL-KDD Dataset
IoT edge devices face dual challenges of severe resource constraints and stringent real-time requirements for intrusion detection. Method: This paper pioneers the systematic application of Hyperdimensional Computing (HDC) to IoT intrusion detection, proposing a lightweight HDC framework that encodes NSL-KDD data into random high-dimensional vectors and employs binding and bundling operations to efficiently represent and classify normal traffic along with four attack types—DoS, Probe, R2L, and U2R. Integrated with tailored data preprocessing and class-balancing strategies, the framework achieves 99.54% accuracy on NSL-KDD. Contribution/Results: It significantly outperforms SVM, Random Forest, and shallow neural networks while requiring minimal parameters and exhibiting ultra-low inference latency. The model demonstrates strong generalization and enables real-time, on-device deployment—overcoming key limitations of conventional machine learning in edge-IoT scenarios.
0 citationsRead paper
Network Anomaly Detection for IoT Using Hyperdimensional Computing on NSL-KDD
Traditional intrusion detection systems (IDS) for IoT exhibit poor generalization on high-dimensional, complex network traffic and heavily rely on prior knowledge of known attacks. Method: This paper pioneers the application of Hyperdimensional Computing (HDC) to anomaly detection on the NSL-KDD dataset, proposing a lightweight HDC-based framework that integrates binary symbolic encoding, feature binding, and similarity matching—enabling efficient detection of both known and unknown attacks without training any classifier. Results: Evaluated on the KDDTrain+ subset, the method achieves 91.55% accuracy, significantly outperforming classical machine learning and deep learning baselines. By eliminating reliance on labeled attack samples, it enhances model robustness and generalization while maintaining computational efficiency. This work establishes a novel unsupervised/low-supervision detection paradigm tailored for resource-constrained IoT environments.
0 citationsRead paper