We conduct an empirical analysis of Shanghai-Hong Kong Stock Connect to reveal the dynamic impacts of stock connect trading activity on the stock pool's Amihud illiquidity proxy, index return, and CNY-HKD exchange rate. From pairwise conditional g causality analysis, we note a mutual significant causal connection between northbound net buying volume and Shanghai stock exchange return on all frequency levels. Meanwhile, we find a significant causal impact on the Shanghai portfolio's liquidity from northbound net buying volume. And there is a significant causal impact from the southbound net buying volume on Hang Seng Index return. Both are significant at the low-frequency level. In particular, northbound trading activity stimulates the Shanghai portfolio's liquidity in the low trading activity regime from the threshold VAR analysis. In robust analysis, we find similar significant dynamic causal connection and stimulation effects for the northbound trades when replacing Amihud illiquidity with the turnover rate. The result might relate to the investment behaviors looking for opportunity in the low trading activity regime. In contrast, the investors' beliefs may vary in the high trading activity regime, which weakens the connection between trading activities and other factors like liquidity.

Based on the different premium volatility characteristics of various systematic factors in the A-share market, this paper constructs six representative high-frequency volatility prediction models that consider multiple complex risk structures. On this basis, a detailed comparative analysis of the differences in volatility characteristics among various factors is conducted, and the optimal prediction and early warning framework for the A-share market is proposed. Research shows that: 1) The volatility research results only for individual market indexes are not universally representative. 2) The fluctuation characteristics among different systematic factors and their respective optimal prediction model frameworks generally have significant differences, that is, there is no single fixed combination of model parameters. 3) Complex risk characteristics such as long memory, measurement errors, and high-frequency jump fluctuations obviously exist in the A-share market. The optimal forecast and early warning framework for the A-share market can be constructed by a combination of models that consider one or more of the above risk characteristics. The above conclusions have important practical reference value for the risk warning and prevention of the A-share market and the formulation of related policies.

A three-dimensional boundary-spanning technology search model including search depth, scope and height is established, and a quantitative calculation method is proposed to dynamically describe an organisation's technology search behaviour and demand characteristics. Organisations are clustered by types as technical, comprehensive, or professional using k-means based on technology search behaviour. Recommendation strategies for various types of organisations are proposed based on this, and the search and supply libraries of each organisation are built by considering their type and search contents. The semantic similarity between patents in different libraries is calculated using a Word2Vec and TextRank model to achieve patent recommendations. An empirical study of the robotics field shows a recommendation accuracy of 0.751, and the accuracy of the technical, comprehensive, and professional types is 0.8282, 0.5389 and 0.7723, respectively. This study considers an organisation's dynamic search behaviour and makes class-based recommendations, with a low computational complexity and strong interpretability.

With the advancement of society and science and technology, the demand for detecting small objects in practical scenarios becomes stronger. Such objects are only represented by relatively small coverage of pixels, and the features are degraded severely after being extracted by a deep convolutional neural network, which is detrimental to the detection performance for small objects. Therefore, an intuitive solution is to increase the resolution of small objects by cropping the original image. In this paper, we propose a simple but effective object density map guided region localization module (DMGRL) to locate and crop the regions of interest where small objects may exist. Firstly, the density map of the objects is estimated by object density map estimation network, and then the coordinates of the small object regions are calculated; Secondly, the continuous differentiable affine transformation is utilized to crop these regions so that the detector with DMGRL can be trained end-to-end instead of two-stage training. Finally, the all prediction results of input image and cropped region images are merged together to output the final detection results by non maximum suppression (NMS). Extensive experiments demonstrate the superior performance of the detector incorporated DMGRL.

This paper introduces the related concepts of the hybrid spherical-shaped dataset and proposes a new discriminant analysis method based on the spherical-shaped dataset (SDAM), then SDAM is further improved by the idea of the class cover and presents the nonlinear discriminant analysis method (NDAM). To demonstrate the effectiveness of these two methods, this work constructs seven hybrid spherical-shaped datasets and uses nine UCI datasets. Numerical experiments on these examples indicate that SDAM can preferably solve the discriminant problem for the hybrid sphericalshaped dataset, but this method does not always work well for real datasets; NDAM overcomes the drawbacks of SDAM and better solves the discriminative problem of real datasets. Besides, it has better stability.