Binary neural networks (BNNs) have been applied in limited resources and mobile devices because of their extreme model compression ability. However, manually designing suitable architectures is challenging given the specialized structure of binarized operations. Neural architecture search (NAS) provides a promising approach for designing high-performance BNN architectures. In practice, various situations require networks with different parameter sizes and performance levels. Therefore, this article proposes a multiobjective evolutionary NAS algorithm for BNNs based on a two-stage training strategy (MO-TS-BNAS) to solve these problems. First, the ApproxSign function is used to approximate the gradient error in the training of BNNs. To avoid the small model trap problem, two auxiliary objectives are introduced in nondominated sorting to retain larger models with similar errors. Then, a two-stage training strategy with flexible use of auxiliary objectives is proposed, forming the selection mechanism in environmental selection. The path dropout method is used in the second stage to prevent hypernetwork overfitting. In addition, the mini-batch gradient descent strategy is improved to speed up individual architecture evaluation and reduce time cost in the search process. Finally, the full precision baseline search space is binarized for general experimental comparison. Our MO-TS-BNAS algorithm balances the two different objective functions of the model size and error. A large number of experiments are carried out on the CIFAR10 and ImageNet datasets, and the results show the effectiveness of the proposed method.

Abstract The generative adversarial network (GAN) is one of the most promising methods in the field of unsupervised learning. Model developers, users, and other interested people are highly concerned about the GAN mechanism where the generative model and the discriminative model learn from each other in a gameplay manner, which generates a causal relationship among output features, internal network structure, feature extraction process, and output results. Through the study of the interpretability of GANs, the validity, reliability, and robustness of the application of GANs can be verified, and the weaknesses of the GANs in specific applications can be diagnosed, which can provide support for designing better network structures. It can also improve security and reduce the decision‐making and prediction risks brought by GANs. In this article, the study of the interpretability of GANs is explored, and ways of the evaluation of the application effect of GAN interpretability techniques are analyzed. Besides, the effect of interpretable GANs in fields such as medical treatment and military is discussed, and current limitations and future challenges are demonstrated.

A two-phase constraint-handling technique is integrated into the evolutionary algorithms to solve constrained optimization problems (called TPDE) in this article. In phase one, denoted as the exploration phase, an exterior penalty function method with the dynamic penalty coefficients is developed to compare any two candidate solutions, which aims to push the population into the feasible region. To reduce the computational burden, in phase two, denoted as the exploitation phase, an interior penalty function method with the dynamic penalty coefficients is developed, which enhances the search ability by using the information of constraints in feasible solutions. During the optimization process, differential evolution is adopted as the search algorithm to produce the offspring population. Experiment results on four benchmark test suites, namely, IEEE CEC 2006, IEEE CEC 2010, IEEE CEC 2017, and IEEE CEC 2020, indicate that TPDE is competitive with other popular algorithms.

Summary: Gene expression changes over the lifespan and varies among different tissues or cell types. Gene co-expression also changes by sex, age, different tissues or cell types. However, gene expression under the normal state and gene co-expression in the human brain has not been fully defined and quantified. Here we present a database named Brain EXPression Database (BrainEXP) which provides spatiotemporal expression of individual genes and co-expression in normal human brains. BrainEXP consists of 4567 samples from 2863 healthy individuals gathered from existing public databases and our own data, in either microarray or RNA-Seq library types. We mainly provide two analysis results based on the large dataset: (i) basic gene expression across specific brain regions, age ranges and sexes; (ii) co-expression analysis from different platforms. Availability and implementation: http://www.brainexp.org/. Supplementary information: Supplementary data are available at Bioinformatics online.

Inspired by the fact that the division of labor and cooperation play extremely important roles in the human history development, this paper develops a novel artificial bee colony algorithm based on information learning (ILABC, for short). In ILABC, at each generation, the whole population is divided into several subpopulations by the clustering partition and the size of subpopulation is dynamically adjusted based on the last search experience, which results in a clear division of labor. Furthermore, the two search mechanisms are designed to facilitate the exchange of information in each subpopulation and between different subpopulations, respectively, which acts as the cooperation. Finally, the comparison results on a number of benchmark functions demonstrate that the proposed method performs competitively and effectively when compared to the selected state-of-the-art algorithms.

The artificial bee colony (ABC) algorithm is a relatively new optimization technique which has been shown to be competitive to other population-based algorithms. However, ABC has an insufficiency regarding its solution search equation, which is good at exploration but poor at exploitation. To address this concerning issue, we first propose an improved ABC method called as CABC where a modified search equation is applied to generate a candidate solution to improve the search ability of ABC. Furthermore, we use the orthogonal experimental design (OED) to form an orthogonal learning (OL) strategy for variant ABCs to discover more useful information from the search experiences. Owing to OED's good character of sampling a small number of well representative combinations for testing, the OL strategy can construct a more promising and efficient candidate solution. In this paper, the OL strategy is applied to three versions of ABC, i.e., the standard ABC, global-best-guided ABC (GABC), and CABC, which yields OABC, OGABC, and OCABC, respectively. The experimental results on a set of 22 benchmark functions demonstrate the effectiveness and efficiency of the modified search equation and the OL strategy. The comparisons with some other ABCs and several state-of-the-art algorithms show that the proposed algorithms significantly improve the performance of ABC. Moreover, OCABC offers the highest solution quality, fastest global convergence, and strongest robustness among all the contenders on almost all the test functions.