Search

Structured pruning is a representative model compression technology for convolutional neural networks (CNNs), aiming to prune some less important filters or channels of CNNs. Most recent structured pruning methods have established some criteria to measure the importance of filters, which are mainly based on the magnitude of weights or other parameters in CNNs. However, these judgment criteria lack explainability, and it is insufficient to simply rely on the numerical values of the network parameters to assess the relationship between the channel and the model performance. Moreover, directly utilizing these pruning criteria for global pruning may lead to suboptimal solutions, therefore, it is necessary to complement search algorithms to determine the pruning ratio for each layer. To address these issues, we propose ARPruning (Attention-map-based Ranking Pruning), which reconstructs a new pruning criterion as the importance of the intra-layer channels and further develops a new local neighborhood search algorithm for determining the optimal inter-layer pruning ratio. To measure the relationship between the channel to be pruned and the model performance, we construct an intra-layer channel importance criterion by considering the attention map for each layer. Then, we propose an automatic pruning strategy searching method that can search for the optimal solution effectively and efficiently. By integrating the well-designed pruning criteria and search strategy, our ARPruning can not only maintain a high compression rate but also achieve outstanding accuracy. In our work, it is also experimentally concluded that compared with state-of-the-art pruning methods, our ARPruning method is capable of achieving better compression results. The code can be obtained at https://github.com/dozingLee/ARPruning. [ABSTRACT FROM AUTHOR]

The cross-correlation technology is applied to decrease the noise figure of microwave photonic link(MPL) for the first time in this paper. Firstly, both effects of the intensity of the laser and the bias voltage of the EOM on the noise figure of MPL are analyzed and optimized. Then, a cascaded low-noise amplifier is added in front of MPL for lower noise figure. Finally, the double laser - double detector structure based on cross-correlation technology is used to offset different kinds of uncorrelated noise in the link and experiments show that the noise figure is further reduced by 3 dB. [ABSTRACT FROM AUTHOR]

A new type of structure for opto-electronic oscillator (OEO) is proposed and verified by experiments in this paper. Probing-signal control method for the long-term stability of opto-electronic oscillator (OEO) is proposed firstly and this new method is used to enhance long-term stability of OEO. Using this new method, a low-frequency signal is used as a reference source, getting through the microwave photonic link in OEO and extracting the time-delay fluctuation information and compensated ultimately. In addition, the long-fiber based OEO is injection-locked by a microwave source for suppressing side modes. Results show that the side mode suppression ratio characteristics are improved obviously after microwave source injection and the long-term stability is greatly enhanced by this probing-frequency method and the Allen Deviation of oscillation signal is reduced to 2.7 × 10 −12 during 1 s. [ABSTRACT FROM AUTHOR]

In this paper, the theory model of phase noise for microwave photonic link is derived and verified by experiments first, considering both types of flicker and white noises. In order to decrease the phase noise of link, low-phase noise amplifier and high-linear photodiode is applied in microwave photonic link. Based on this model, the effect of phase noise of microwave photonic link on the OEO is also analyzed and we found the lower-phase noise microwave photonic link results in better-characteristics OEO. [ABSTRACT FROM AUTHOR]