دانلود رایگان مقاله انگلیسی ارائه مدولار شبکه های عصبی لایه بندی شده - الزویر 2018

abstract

Layered neural networks have greatly improved the performance of various applications including image processing, speech recognition, natural language processing, and bioinformatics. However, it is still difficult to discover or interpret knowledge from the inference provided by a layered neural network, since its internal representation has many nonlinear and complex parameters embedded in hierarchical layers. Therefore, it becomes important to establish a new methodology by which layered neural networks can be understood. In this paper, we propose a new method for extracting a global and simplified structure from a layered neural network. Based on network analysis, the proposed method detects communities or clusters of units with similar connection patterns. We show its effectiveness by applying it to three use cases. (1) Network decomposition: it can decompose a trained neural network into multiple small independent networks thus dividing the problem and reducing the computation time. (2) Training assessment: the appropriateness of a trained result with a given hyperparameter or randomly chosen initial parameters can be evaluated by using a modularity index. And (3) data analysis: in practical data it reveals the community structure in the input, hidden, and output layers, which serves as a clue for discovering knowledge from a trained neural network.

6. Conclusion

Layered neural networks have achieved a significant improvement in terms of classification or regression accuracy over a wide range of applications by their ability to capture the complex hidden structure between input and output data. However, the discovery or interpretation of knowledge using layered neural networks has been difficult, since its internal representation consists of many nonlinear and complex parameters. In this paper, we proposed a new method for extracting a modular representation of a trained layered neural network. The proposed method detects communities of units with similar connection patterns, and determines the relational structure between such communities. We demonstrated the effectiveness of the proposed method experimentally in three applications. (1) It can decompose a layered neural network into a set of small independent networks, which divides the problem and reduces the computation time. (2) The trained result can be estimated by using a modularity index, which measures the effectiveness of a community detection result. And (3) providing the global relational structure of the network would be a clue to discover knowledge from a trained neural network.