Bai Liang

Feature analysis and extraction for audio automatic classification

Feature analysis and extraction are the foundation of audio automatic classification. This paper divides audio streams into five classes: silence, noise, pure speech, speech over background sound and music. We present our work on audio feature analysis and extraction on the frame level and clip level. Four new features are proposed, including silence ratio, pitch frequency standard deviation, harmonicity ratio and smooth pitch ratio. We have presented an SVM based approach to classification. The effectiveness of the features is evaluated in experiments. Experiment results show that the features we selected and proposed are rational and effective.

Controllability and Directionality in Complex Networks

We investigate how to assign link directions in a given complex network to enhance its controllability. Based on the node residual degree, a method of assigning link direction is proposed. Numerical simulation demonstrates that the method outperforms the random direction assignment method in enhancing the controllability of the network. Furthermore, robustness of control is also improved.

Optimization of robustness of network controllability against malicious attacks

As the controllability of complex networks has attracted much attention recently, how to design and optimize the robustness of network controllability has become a common and urgent problem in the engineering field. In this work, we propose a method that modifies any given network with strict structural perturbation to effectively enhance its robustness against malicious attacks, called dynamic optimization of controllability. Unlike other structural perturbations, the strict perturbation only swaps the links and keeps the in- and out-degree unchanged. A series of extensive experiments show that the robustness of controllability and connectivity can be improved dramatically. Furthermore, the effectiveness of our method is explained from the views of underlying structure. The analysis results indicate that the optimization algorithm makes networks more homogenous and assortative.

Smart Rewiring: Improving Network Robustness Faster*

Previous work puts forward a random edge rewiring method which is capable of improving the network robustness noticeably while it lacks further discussions about how to improve the robustness faster. In this study the detailed analysis of the structures of improved networks show that regenerating the edges between high-degree nodes can enhance the robustness against a targeted attack. Therefore, we propose a novel rewiring strategy based on regenerating more edges between high-degree nodes, called smart rewiring, which could speed up the increase of the robustness index effectively. The smart rewiring method also explains why positive degree-degree correlation could enhance network robustness.

Enhancing Complex Network Controllability by Rewiring Links

We propose a method to enhance controllability of directed network by rewiring links while keeping the total number of links unchanged. Numerical simulation on two canonical network models (ER and SF) demonstrates that this method is effective. We show that the improved controllability is mainly determined by the mean degree and does not depend on the initial degree distribution and the size of network. Furthermore, we discuss the correlation between controllability and heterogeneity.