The main goal of multitask optimization (MTO) is the parallel optimization of multiple different tasks. However, since different tasks in the MTO problem usually have heterogeneous characteristics, it is difficult to realize the positive knowledge transfer among tasks, resulting in poor convergence. To cope with this problem, a multi-task particle swarm optimization with a heterogeneous domain adaptation strategy (MTPSO-HDA) is proposed to transfer positive knowledge among heterogeneous tasks. First, a nonlinear mapping between the source task and the target task is constructed based on the adaptive kernel function. Then, source tasks are mapped to the target task space to reduce the differences among heterogeneous tasks. Second, a multi-source domain adaptive strategy based on fitness landscape similarity is designed to implement domain adaptation. Then, the importance of each source domain is quantitatively described to reduce the differences between multiple source domains and a target domain and achieve domain adaptation among heterogeneous tasks. Third, a heterogeneous multitask particle swarm optimization mechanism is introduced to facilitate positive knowledge transfer among heterogeneous tasks. Then, an appropriate evolutionary mechanism is designed according to the fitness landscape similarity to achieve positive knowledge transfer. Finally, to assess the effectiveness of the MTPSO-HDA algorithm, some experiments are designed based on some benchmark problems and real-world application of wastewater treatment process. The results demonstrate that the proposed MTPSO-HDA algorithm can promote positive knowledge transfer among heterogeneous tasks to improve convergence.

We propose a non-cascaded and crosstalk-free multi-image encryption method based on optical scanning holography and 2D orthogonal compressive sensing. This approach enables the simultaneous recording and encryption of multiple plaintext images without mechanical scanning, while allows for independent retrieval of each image with exceptional quality and no crosstalk. Two features would bring about more substantial security and privacy. The one is that, by employing a sequence of pre-designed structural patterns as encryption keys at the pupil, multiple samplings can be achieved and ultimately the holographic cyphertext can be obtained. These patterns are generated using a measurement matrix processed with the generalized orthogonal one. As a result, one can accomplish the differentiation of images prior to the recording and thus neither need to pretreat the pending images nor to suppress the out-of-focus noise in the decrypted image. The other one is that, the non-cascaded architecture ensures that different plaintexts do not share sub-keys. Meanwhile, compared to 1D orthogonal compressive sensing, the 2D counterpart makes the proposed method to synchronously deal with multiple images of more complexity, while acquire significantly high-quality decrypted images and far greater encryption capacity. Further, the regularities of conversion between 1D and 2D orthogonal compressive sensing are identified, which may be instructive when to manufacture a practical multi-image cryptosystem or a single-pixel imaging equipment. A more general method or concept named synthesis pupil encoding is advanced. It may provide an effective way to combine multiple encryption methods together into a non-cascaded one. Our method possesses nonlinearity and it is also promising in multi-image asymmetric or public key cryptosystem as well as multi-user multiplexing.

Multitask optimization aims to solve multiple optimization problems in parallel utilizing a single population. However, if the computing resources are limited, allocating the same computing resources to different tasks will cause resource waste and make complex tasks difficult to converge to the optimal solution. To address this issue, a multitask particle swarm optimization with a dynamic on-demand allocation strategy (MTPSO-DA) is proposed to dynamically allocate computing resources. First, a task complexity index, based on convergence rate and contribution rate, is designed to evaluate the difficulty of solving different tasks. Then, the complexity of different tasks can be evaluated in real time. Second, the skill factor of the particle is extended to a time-varying matrix according to the task complexity index. Then, the recently captured feedback is stored to determine the computational resource demands of the task. Third, an on-demand allocation strategy, based on the time-varying matrix, is developed to obtain the skill factor probability vector utilizing the attenuation accumulation method. Then, computing resources can be allocated dynamically among different tasks. Finally, some comparative experiments are conducted based on the benchmark problem to evaluate the superiority of the MTPSO-DA algorithm. The results indicate that the proposed MTPSO-DA algorithm can achieve dynamic resource allocation.

Particle swarm optimization algorithm has become a promising approach in solving multitask optimization (MTO) problems since it can transfer knowledge with easy implementation and high searching efficiency. However, in the process of knowledge transfer, negative transfer is common because it is difficult to evaluate whether knowledge is effective for population evolution. Therefore, how to obtain and transfer the effective knowledge to curb the negative transfer is a challenging problem in MTO. To deal with this problem, a self-adjusting multitask particle swarm optimization (SA-MTPSO) algorithm is designed to improve the convergence performance in this article. First, a knowledge estimation metric, combining the decision space knowledge and the target space knowledge for each task, is designed to describe the effectiveness of knowledge. Then, the effective knowledge is obtained to promote the knowledge transfer process. Second, a self-adjusting knowledge transfer mechanism, based on the effective knowledge and the self-adjusting transfer method, is developed to achieve effective knowledge transfer. Then, the ineffective knowledge is removed to solve the negative transfer problem. Third, the convergence analysis is given to guarantee the effectiveness of the SA-MTPSO algorithm theoretically. Finally, the proposed algorithm is compared with some existing MTO algorithms. The results show that the performance of the proposed algorithm is superior to most algorithms on negative transfer suppression and convergence.

Abstract In this paper, a novel convolutional neural network for fast semantic segmentation is presented. Deep convolutional neural networks have achieved great progress in the task of vision scene understanding. While the increase of the accuracy mainly depends on the increase of depth and width. This slows down large networks and consumes power. A fast and efficient convolutional neural network, PGCNet, aiming at segmenting high‐resolution images with a high speed is introduced. Compared with the competitive methods, the generated model has high performance with fewer parameters and floating point operations. First, a lightweight general architecture pre‐trained on ImageNet is relied on as the main encoder. Then, a novel lateral connection module to better transmit features from encoder to decoder. Third, a powerful method termed as PGCN block to extract features of each block in the encoder is proposed and an edge decoder is applied as a supervision for pixels on the edge of stuff and things during training. Experiments show that this method has great advantages. Based on the proposed PGCNet, 75.8% mean IoU is achieved on the cityscapes test set and 35.4 Hz on a standard Cityscapes image on GTX1080Ti.