Low rank optimization for efficient deep learning:making a balance between compact architecture and fast training

Deep neural networks(DNNs)have achieved great success in many data processing applications.However,high computational complexity and storage cost make deep learning difficult to be used on resource-constrained devices,and it is not environmental-friendly with much power cost.In this paper,we focus on low-rank optimization for efficient deep learning techniques.In the space domain,DNNs are compressed by low rank approximation of the network parameters,which directly reduces the storage requirement with a smaller number of network parameters.In the time domain,the network parameters can be trained in a few subspaces,which enables efficient training for fast convergence.The model compression in the spatial domain is summarized into three categories as pre-train,pre-set,and compression-aware methods,respectively.With a series of integrable techniques discussed,such as sparse pruning,quantization,and entropy coding,we can ensemble them in an integration framework with lower computational complexity and storage.In addition to summary of recent technical advances,we have two findings for motivating future works.One is that the effective rank,derived from the Shannon entropy of the normalized singular values,outperforms other conventional sparse measures such as the?_1 norm for network compression.The other is a spatial and temporal balance for tensorized neural networks.For accelerating the training of tensorized neural networks,it is crucial to leverage redundancy for both model compression and subspace training.

Optimizing Rank of Landscape Planning Works of Urban Wetland Park

Classifying and ranking the huge amounts of landscape planning works of urban wetland park is always difficult due to the multi-functions （ecological, leisure, educational and disaster prevention） of the urban wetland park. Therefore, an optimizing rank system is urgently needed. Analytic Hierarchy Process （AHP） and Technique for Order Preference by Similarity to Ideal Solution （TOPSIS） models were used to rank the planning works of 30 urban wetland park based on four mainly factors, which included landscape ecological planning, landscape planning, ecological planning and economic planning. The study indicated that the AHP- TOPSIS model had good discrimination in the classification and ranking of landscape planning works of urban wetland park and it was also applicable to the planning works of other urban greenbelts.

Exploiting global information in complex network repair processes

Robustness of complex networks has been studied for decades,with a particular focus on network attack.Research on network repair,on the other hand,has been conducted only very lately,given the even higher complexity and absence of an effective evaluation metric.A recently proposed network repair strategy is self-healing,which aims to repair networks for larger components at a low cost only with local information.In this paper,we discuss the effectiveness and efficiency of self-healing,which limits network repair to be a multi-objective optimization problem and makes it difficult to measure its optimality.This leads us to a new network repair evaluation metric.Since the time complexity of the computation is very high,we devise a greedy ranking strategy.Evaluations on both real-world and random networks show the effectiveness of our new metric and repair strategy.Our study contributes to optimal network repair algorithms and provides a gold standard for future studies on network repair.

OPTIMISM IN THE RANKS

As the 17th National Congress of the ruling Communist Party of China closed on October 21,a grassroots ethnic minority Party member looks at past progress and challenges facing future development