Focusing on higher organisms, this paper analyzes and develops a comprehensive review of the problems in bionic computing and also proposes and expounds some new views and insights, from the perspective of collective intelligence as a whole, which includes swarm intelligence and crowd intelligence. On the basis of an overview on the research progress of bionic computation in higher organisms (including fundamental higher organisms, regular higher organisms and quasi-man organisms), the reflux phenomenon in the research on the trend of making algorithms marked by “zoo algorithm” in swarm intelligence optimization is found. A reasonable interpretation of the reasons for the formation of the trend of making algorithms from both the bionic-computational dimension and the problem-method dimension. Furthermore, the overall idea of problem solving is given, and the two main development directions of bionic computing for collective intelligence are refined and formed. Emphasis on the expansion of bionic behavior towards cooperative behavior is dominant in the direction of collective intelligence bionic computing development. Aiming at the difficulties existing in the research of swarm intelligence optimization, five bottlenecks that need to be focused on to achieve breakthroughs are proposed. Based on the overall view of “metaphorical bionic computing-normative bionic computing-complex bionic computing”, the new paradigm of intelligent computing of complex bionic computing is advocated, which can guide the direction for higher organism bionic computing.

To explore the behavioral characteristics of massive crowds under the active interaction of millions of streamers and viewers in the field of live streaming, this paper summarized the temporal patterns of live streaming workload and user behavior characteristics of the live streaming platform, taking Douyu and Huya live streaming platforms as examples, a statistical analysis of 123 consecutive days, involving more than 2.4 million anchors, and more than 726 million live streaming data. The live streaming workload has obvious intra-day and intra-week effect. Different live streaming modes have significant differences in live streaming characteristics such as the average number of viewers and followers. The lifetime of streamers and the number of viewers conform to a power law distribution. With the development of the platform, there is a strong linear correlation between the number of streamers and viewers, but its volatility is gradually increasing, reflecting the increasingly strong heterogeneity and non-uniformity of the system. It is of great significance for understanding user behavior patterns in complex systems of live streaming, mining user distribution laws and changing trends, and designing business models such as personalized recommendations.

In order to promote the development of autonomous driving technology, this paper focuses on the lane-changing decision-making behavior of autonomous vehicles. First, the lane-changing intention is quantified objectively, and then the lane-changing collision probability and the lane-changing dynamic risky coefficient are introduced. Based on the game theory, the decision-making behavior model of the lane-changing game for autonomous vehicles is established. Besides, speed gains are considered as the objective of game gains. Therefore, autonomous vehicles can change lanes in a coordinated, safe and reasonable manner. Finally, with SUMO software, the traditional LC2013 lane-changing model and the decision-making behavior model of the lane-changing game are used for simulation experiments and comparative analysis. The results show that the decision-making behavior model of the lane-changing game has higher stability, reliability, safety and lane utilization.

Aiming at the problems of inaccurate keywords extraction and only considering single correlation in subject words extraction, a subject words extraction algorithm combining integration idea with complex network is proposed. Firstly, the keywords of topic data are extracted through the integration algorithm to improve the accuracy of keywords extraction. Secondly, the traditional word co-occurrence formula is improved to calculate the co-occurrence degree of keywords, and a keywords co-occurrence network is established. Based on the network, the optimal connected subgraph is obtained. At the same time, the importance of keywords is measured by taking the centrality of node degree as the weight, and the subject words are mapped. Finally, the micro-blog topic data set is used to verify the example, which proves that the algorithm is effective and better than the traditional word co-occurrence algorithm, and it is applied in the Qingdao community topic data set.

Complex criminal cases today present systematic characteristics of multi-factor coupling and dynamic evolution, and their investigation process faces the challenge of nonlinear information integration. Criminal suspects use anti-detection methods such as changing clothes and shoes, facial obstruction, and posture camouflage, combined with complex environmental interference, which significantly reduces the practical effectiveness of single technical means such as face recognition and video structuring. In order to resolve this problem, this article focuses on the actual needs of suspect identification and tracking, breaks through the recognition bottleneck of a single modality, systematically integrates multi-information such as video structuring, face recognition, and gait recognition, and proposes a multi-information fusion video investigation system with gait recognition as the core, which realizes the dual characterization of suspect behavior patterns and identity characteristics, and provides a new technical path for improving identity recognition capabilities and the efficiency of solving complex cases.

In order to ensure that urban emergency supplies can be delivered to the demand point in a timely and accurate manner, especially for special emergency supplies with a short life cycle′ perishable goods, its timeliness requirements are higher. Based on the emergency scenario of sudden public health events, this paper establishes a multi-objective location model of urban perishable goods distribution center with the goal of minimizing transportation time and transportation cost and maximizing relative coverage area. The Harris Hawk optimization algorithm (HHO) is improved to achieve an effective solution to the multi-objective location problem of perishable goods distribution center. In order to verify the effectiveness of the model, a district in Shanghai is selected as a research example. The results show that the improved HHO algorithm can solve the location model of perishable goods distribution center under actual urban road conditions, and can provide an intuitive multi-objective location optimization scheme.

To scientifically evaluate the resilience of China′s industrial chain and promote high-quality economic development, this paper employs complex network theory and utilizes data from China′s multi-regional input-output table to construct a simulated attack model of the industrial chain network, thereby measuring the resilience of the industrial chain. The results show that the industrial chain layout of 31 provinces in China exhibits obvious local correlation attributes, and presents a collaborative development characteristic led by key regions and industries; The overall resilience of China′s industrial chain is strong, and during the inspection period, the overall resilience index of China′s industrial chain shows an upward trend; There is a significant gap in the resilience of the industrial chain between regions and industries in China.

In order to better reveal the transmission mechanism of COVID-19, this paper proposes the SEAIHR dynamic model by analyzing the transmission characteristics of COVID-19, considering the self-healing of the hidden lurks and the early isolation of the lurks, introducing “h hospitalization isolation”, “recessive cure”, considering the change of prevention and control intensity, and introducing “morbidity status”. Using the real epidemic data and considering the changes of parameters in different stages, a multi model comparative test was conducted. The experimental results showed that the fitting and prediction accuracy of the SEAIHR model was significantly improved, and the fitting error was 34.4%~72.8% lower than that of the classical model in the early and middle stages of the epidemic, providing reference and guidance for epidemic prevention and control.

To improve the ability to predict the evolution trend of public opinion, a public opinion evolution trend prediction model based on an improved wolf pack algorithm and optimized long-short term memory neural network is proposed. Use Halton Sequence to initialization to improve population diversity. Design step factor to perform Gauss-Sine perturbation transformation to improve wolf group exploration and development capabilities. Combine with the spiral in the whale optimization algorithm to improve the siege mechanism to enhance the local search ability of wolves. The bidirectional memory population is used to increase the cooperative ability of the wolf pack. The improved wolf pack algorithm (IWPA) is applied to the hyperparameter prediction of the LSTM neural network. Using keywords such as “COVID-19” and “Food Safety”, the experiment proves that the IWPA-LSTM neural network public opinion evolution prediction model has good accuracy and generality. The model is suitable for the prediction of various public opinion evolution trends.

SeqSLAM is a widely used loop closure detection algorithm in mobile robot and autonomous vehicle field. It could recognize revisited places by comparing sequences of images even under dramatic changes of season, illumination, and weather. However, SeqSLAM is vulnerable to viewpoint changes. In addition, SeqSLAM compares sequences of images by brute force method, which prevents its real-time application to large-scale image datasets. To address these problems, we first represent each image by a kind of low dimensional description — vector of locally aggregated descriptors (VLAD) which is robust to viewpoint changes, and then replace the brute force method by an approximate nearest neighbor search algorithm — hierarchical navigable small world graphs (HNSW). Tests on publicly available datasets show that, the improved SeqSLAM with VLAD and HNSW could obtain much better detection results in the respect of precision-recall evaluation and the search time is reduced by orders of magnitude. We make code publicly available at https://github.com/qipengliuNTU/Efficient-SeqSLAM-with-HNSW.