Search

In this paper, we present a novel system (denoted as Polaca) to generate poetic Chinese landscape painting with calligraphy. Unlike previous single image-to-image painting generation, Polaca takes the classic poetry as input and outputs the artistic landscape painting image with the corresponding calligraphy. It is equipped with three different modules to complete the whole piece of landscape painting artwork: the first one is a text-to-image module to generate landscape painting image, the second one is an image-to-image module to generate stylistic calligraphy image, and the third one is an image fusion module to fuse the two images into a whole piece of aesthetic artwork., Comment: Accepted by IJCAI 2022

Artificial intelligence is a bellwether of today’s new technological revolution. As a branch of artificial intelligence, machine learning faces two main problems in practical application: 1) data owned by most enterprises are difficult to aggregate; and, 2) big data owners pay more and more attention to data privacy and security, which leads to the problem of data island. Federated learning (FL), as a distributed FL paradigm, which can enable all parties to achieve the purpose of co-building models while ensuring data privacy and exposure, provides a possible solution to the mentioned problems of machine learning. The FL takes full advantage of participants’ data and computing power and builds a more robust machine learning model without sharing the data. In an environment with strict data regulation, the FL can effectively solve the key problems, such as those related to data privacy and data rights. However, most of the existing FL-based frameworks do not pay attention to the impact of data source distribution on FL training. Therefore, this paper proposed a data-oriented FL framework called the Federated AI Engine(FAE), which can solve FL problems without leaving the data in control. The proposed framework provides a method that can be used to verify FL quickly for researchers who intend to try federal learning.

Sequential recommendation and group recommendation are two important branches in the field of recommender system. While considerable efforts have been devoted to these two branches in an independent way, we combine them by proposing the novel sequential group recommendation problem which enables modeling group dynamic representations and is crucial for achieving better group recommendation performance. The major challenge of the problem is how to effectively learn dynamic group representations based on the sequential user-item interactions of group members in the past time frames. To address this, we devise a Group-aware Long- and Short-term Graph Representation Learning approach, namely GLS-GRL, for sequential group recommendation. Specifically, for a target group, we construct a group-aware long-term graph to capture user-item interactions and item-item co-occurrence in the whole history, and a group-aware short-term graph to contain the same information regarding only the current time frame. Based on the graphs, GLS-GRL performs graph representation learning to obtain long-term and short-term user representations, and further adaptively fuse them to gain integrated user representations. Finally, group representations are obtained by a constrained user-interacted attention mechanism which encodes the correlations between group members. Comprehensive experiments demonstrate that GLS-GRL achieves better performance than several strong alternatives coming from sequential recommendation and group recommendation methods, validating the effectiveness of the core components in GLS-GRL.

Real-world integrated personalized recommendation systems usually deal with millions of heterogeneous items. It is extremely challenging to conduct full corpus retrieval with complicated models due to the tremendous computation costs. Hence, most large-scale recommendation systems consist of two modules: a multi-channel matching module to efficiently retrieve a small subset of candidates, and a ranking module for precise personalized recommendation. However, multi-channel matching usually suffers from cold-start problems when adding new channels or new data sources. To solve this issue, we propose a novel Internal and contextual attention network (ICAN), which highlights channel-specific contextual information and feature field interactions between multiple channels. In experiments, we conduct both offline and online evaluations with case studies on a real-world integrated recommendation system. The significant improvements confirm the effectiveness and robustness of ICAN, especially for cold-start channels. Currently, ICAN has been deployed on WeChat Top Stories used by millions of users. The source code can be obtained from https://github.com/zhijieqiu/ICAN.

In this article, an operation strategy of digital micromirror device (DMD) and the maximum intensity projection (MIP) image processing method are proposed to improve the contrast of images in confocal microscopy. First, the configuration of DMD is demonstrated and the effect of scanning unit size on image performance is analyzed. Then, the image processing method MIP is applied. According to the MIP method, only the maximum intensity projection point of the same pixel is chosen from every image, and the maximum intensity projection point exactly corresponds to the positon where mirror is at "on" position during the scanning process in DMD-based confocal microscope system,. Thus, high contrast of images can be achieved by using MIP. Finally, experiments are conducted to verify imaging performance by changing the parameter of scanning unit size and applying a MIP image processing technique. The results show that DMD scanning unit size and MIP image processing techniques play important roles in improving image contrast. Smaller scanning unit size of DMD improves axial contrast but greatly decreases the signal to noise ratio, which thus leads to reduced image contrast. Larger scanning unit size produces a better signal to noise ratio, thus better image contrast. However, a large S will sacrifice the processing time. Therefore, DMD scanning unit size should be smaller on the premise that image contrast can be satisfied. RESEARCH HIGHLIGHTS: Effect of DMD scanning unit size setting on image contrast is analyzed and verified. The maximum intensity projection (MIP) is investigated to improve the image contrast. Experiments are conducted to verify the enhancement of the image contrast.

The effect of different atomic substitutions at Mn sites on the magnetic and magnetocaloric properties in Ni50Mn35Co2Sn13 alloy has been studied in detail. The substitution of Ni or Co for Mn atoms might lower the Mn content at Sn sites, which would reduce the d-d hybridization between Ni 3d eg states and the 3d states of excess Mn atoms at Sn sites, thus leading to the decrease of martensitic transformation temperature TM in Ni51Mn34Co2Sn13 and Ni50Mn34Co3Sn13 alloys. On the other hand, the substitution of Sn for Mn atoms in Ni50Mn34Co2Sn14 would enhance the p-d covalent hybridization between the main group element (Sn) and the transition metal element (Mn or Ni) due to the increase of Sn content, thus also reducing the TM by stabilizing the parent phase. Due to the reduction of TM, a magnetostructural martensitic transition from FM austenite to weak-magnetic martensite is realized in Ni51Mn34Co2Sn13 and Ni50Mn34Co2Sn14, resulting in a large magnetocaloric effect around room temperature. For a low field change of 3 T, the maximum ∆SM reaches as high as 30.9 J/kg K for Ni50Mn34Co2Sn14. A linear dependence of &Delta, SM upon &mu, 0H has been found in Ni50Mn34Co2Sn14, and the origin of this linear relationship has been discussed by numerical analysis of Maxwell&rsquo, s relation.