Microwave photonics frequency measurement with improved accuracy based on an artificial neural network.

Photonics-assisted techniques for microwave frequency measurement (MFM) show great potential for overcoming electronic bottlenecks, with wild applications in radar and communication. The MFM system based on the stimulated Brillouin scattering (SBS) effect can measure the frequency of multiple high-frequency and wide-band signals. However, the accuracy of the MFM system in multi-tone frequency measurement is constrained by the SBS bandwidth and the nonlinearity of the system. To resolve this problem, a method based on an artificial neural network (ANN) is suggested, which can establish a nonlinear mapping between the measured two-tone signal spectra and the theoretical frequencies. Through simulation verification, the ANN optimized frequencies within the range of (0.5, 27) GHz of the MFM system show 79%, 76%, 70%, 44% reduction in errors separately under four spectral signal-to-noise ratios (SNR) conditions, 20 dB, 15 dB, 10 dB, 0 dB, and the frequency resolution is improved from 30 MHz to 10 MHz.

Kif15 deficiency contributes to depression-like behavior in mice.

Neuropsychiatric disorders have a high incidence worldwide. Kinesins, a family of microtubule-based molecular motor proteins, play essential roles in intracellular and axonal transport. Variants of kinesins have been found to be related to many diseases, including neurodevelopmental/neurodegenerative disorders. Kinesin-12 (also known as Kif15) was previously found to affect the frequency of both directional microtubule transports. However, whether Kif15 deficiency impacts mood in mice is yet to be investigated. In this study, we used the CRISPR/Cas9 method to obtain Kif15-/- mice. In behavioral tests, Kif15-/- female mice exhibited prominent depressive characteristics. Further studies showed that the expression of BDNF was significantly decreased in the frontal cortex, corpus callosum, and hippocampus of Kif15-/- mice, along with the upregulation of Interleukin-6 and Interleukin-1ß in the corpus callosum. In addition, the expression patterns of AnkG were notably changed in the developing brain of Kif15-/- mice. Based on our previous studies, we suggested that this appearance of altered AnkG was due to the maladjustment of the microtubule patterns induced by Kif15 deficiency. The distribution of PSD95 in neurites notably decreased after cultured neurons treated with the Kif15 inhibitor, but total PSD95 protein level was not impacted, which revealed that Kif15 may contribute to PSD95 transportation. This study suggested that Kif15 may serve as a potential target for future depression studies.

Label smoothing and task-adaptive loss function based on prototype network for few-shot learning.

Aiming at solving the problems of prototype network that the label information is not reliable enough and that the hyperparameters of the loss function cannot follow the changes of image feature information, we propose a method that combines label smoothing and hyperparameters. First, the label information of an image is processed by label smoothing regularization. Then, according to different classification tasks, the distance matrix and logarithmic operation of the image feature are used to fuse the distance matrix of the image with the hyperparameters of the loss function. Finally, the hyperparameters are associated with the smoothed label and the distance matrix for predictive classification. The method is validated on the miniImageNet, FC100 and tieredImageNet datasets. The results show that, compared with the unsmoothed label and fixed hyperparameters methods, the classification accuracy of the flexible hyperparameters in the loss function under the condition of few-shot learning is improved by 2%-3%. The result shows that the proposed method can suppress the interference of false labels, and the flexibility of hyperparameters can improve classification accuracy.