Search

Deterministic optimization has been successfully applied to a series of design problems of square thin-walled energy absorption tubes and to a certain extent has fulfilled great expectations for the application of such structures in subway vehicles. However, most studies have not considered the uncertainty of parameters or the correlation of uncertainty parameters, leaving little or no tolerance and resulting in over-conservative structural design. This research proposes a multi-objective uncertain method with an ellipsoid-based model to address the effects of parametric uncertainties of a centrally symmetrical square tube with diaphragms (CSSTD) on design optimization, in which the ellipsoid model is adopted to describe the related uncertainty parameters. The nonlinear interval number programming method coupled with a reliability-based possibility degree of interval (RPDI) model is introduced to handle the transformation of uncertain optimization problems. Simultaneously, local-densifying technology is adopted to enhance the local accuracy of the approximate model. Finally, the outer layer of the micro multi-objective genetic algorithm (μMOGA) combined with the inner layer of the intergeneration projection genetic algorithm (IP-GA) is applied to solve the Pareto optimal solution set of the transformed deterministic optimization. The optimization results indicate that the proposed multi-objective uncertain optimization with an ellipsoid-based model not only guarantees the crashworthiness of the CSSTD, but also improves the design robustness, which means that the proposed method can provide insightful information for crashworthiness design of subways.

The transient characteristics of short‐circuit current (SCC) in doubly‐fed induction generator (DFIG) are quite different from the widely researched conventional synchronous machines. However, the second harmonic characteristics of SCC that impact the reliability of the transformer protection have not been accurately analysed. This paper presents the generation mechanism of second harmonic SCC in DFIG by deducing the dynamic response of rotor‐side converter, with particular attention to the influences of the voltage drop, harmonic component of stator voltage, frequency coupling, and deviation of the phase‐locked loop. When grid fault occurs, the terminal voltage sag leads to fluctuations in the active power, then the response of converter to power fluctuation and the frequency coupling leads to the second harmonic current in the abc reference frame. And the second harmonic current injected into the grid leads to fluctuations in the voltage disturbance term and deviation of the phase‐locked loop, which, in turn, affects the second harmonic current. Based on the analysis, the mitigation measures for the second harmonic current are proposed. Simulations are implemented to verify the theoretical analysis.

Aiming at the teaching challenge of cultivating integrated talents of Internet of things worldwide, and combining with the social background of fire rescue and management problems, We propose a virtual simulation teaching platform for the Internet of Things under the background of intelligent fire protection application. Based on the overall structure of the Internet of Things, different fire scenarios that cannot be simulated in real life are constructed through virtual simulation technology. OMNet++ technology is adopted to carry out virtual deployment of fire nodes for the overall structure of the building, and experiments such as cluster routing simulation, communication transmission simulation and wireless sensor node data acquisition simulation are designed. Meanwhile, a 3D fire data model is established using big data to simulate the best fire extinguishing scheme and the best escape strategy. From aspects of the Internet of Things system design and development, sensor principle and application, Internet of Things communication technology and Internet of Things data storage and application, we have realized the efficiency, innovation and challenge of the Internet of Things teaching.The virtual simulation teaching platform we built has been deployed and put into practical teaching, which has received positive response from students and achieved excellent teaching effect.

The echo state network (ESN) is a state-of-the art reservoir computing approach, which is particularly effective for time series forecasting problems because it is coupled with a time parameter. However, the linear regression algorithm commonly used to compute the output weights of ESN could usually cause the trained network over-fitted and thus obtain unsatisfactory results. To overcome the problem, we present four optimized ESNs that are based on the backtracking search optimization algorithm (BSA) or its variants to improve generalizability. Concretely, we utilize BSA and its variants to determine the most appropriate output weights of ESN given that the optimization problem is complex while BSA is a novel evolutionary algorithm that effectively unscrambles optimal solutions in complex spaces. The three BSA variants, namely, adaptive population selection scheme (APSS)–BSA, adaptive mutation factor strategy (AMFS)–BSA, and APSS&AMFS–BSA, were designed to further improve the performance of BSA. Time series forecasting experiments were performed using two real-life time series. The experimental results of the optimized ESNs were compared with those of the basic ESN without optimization, and the two other comparison approaches, as well as the other existing approaches. Experimental results showed that (a) the results of the optimized ESNs are more accurate than that of basic ESN and (b) APSS&AMFS–BSA–ESN nearly outperforms basic ESN, the three other optimized ESNs, the two comparison approaches, and other existing optimization approaches.

Research in psychology, pedagogy, and physiology has shown that learning a second language will be a great benefit to preschool children. Compared with adults, children have more advantages in learning, such as better pronunciation and intonation. However, because of children’s inattention, the effects of the second language education in many preschools are not ideal in China. To make children learn the second language effectively, it should be integrated into their daily lives and in line with their interests. In this paper, an educational robot system with object recognition technology is introduced, which aims to provide innovative second language learning services for preschool children in China. The proposed system combines object recognition and projection with English teaching and makes objects in daily life more interesting with projected animation to attract children’s attention. On the projection screen, children can interact with a robot by touch or movement and can easily trigger more interactive effects. To evaluate the effectiveness of the proposed system, we conducted an experiment, and the results showed that the system can improve the language learning efficiency for the Chinese preschool children by interacting with objects under proper guidance.

Learning high-quality sentence representations benefits a wide range of natural language processing tasks. Though BERT-based pre-trained language models achieve high performance on many downstream tasks, the native derived sentence representations are proved to be collapsed and thus produce a poor performance on the semantic textual similarity (STS) tasks. In this paper, we present ConSERT, a Contrastive Framework for Self-Supervised Sentence Representation Transfer, that adopts contrastive learning to fine-tune BERT in an unsupervised and effective way. By making use of unlabeled texts, ConSERT solves the collapse issue of BERT-derived sentence representations and make them more applicable for downstream tasks. Experiments on STS datasets demonstrate that ConSERT achieves an 8\% relative improvement over the previous state-of-the-art, even comparable to the supervised SBERT-NLI. And when further incorporating NLI supervision, we achieve new state-of-the-art performance on STS tasks. Moreover, ConSERT obtains comparable results with only 1000 samples available, showing its robustness in data scarcity scenarios., Accepted by ACL2021, 10 pages, 7 figures, 4 tables

The University Course Timetabling Problem (UCTP) is a highly constrained real-world combinatorial optimization task. The purpose of this study is to improve the scientific nature of constraints and the solving performance of algorithms to supplement the previous studies. First of all, through crawling and analyzing the university post bar data, the characteristics of teachers’ and students’ preferences different from those of previous studies are obtained, which are taken as the main constraints, and the improvement of class efficiency is also taken into account to establish a mathematical model. Secondly, according to the characteristics of the problem, an improved genetic algorithm with dynamic search and mandatory directional iteration is designed to solve such problems, which can significantly reduce the number of optimization iterations and obtain better solutions. Then, this paper applies the real data to conduct a case study. The genetic algorithm and particle swarm optimization are compared with the improved genetic algorithm in this paper. The multi-dimensional analysis proves the superiority of the improved algorithm. The results of this paper can not only effectively solve the UCTP, but also provide model and algorithm support for scheduling problems considering the distance factor.

Recently, conversational recommender system (CRS) has become an emerging and practical research topic. Most of the existing CRS methods focus on learning effective preference representations for users from conversation data alone. While, we take a new perspective to leverage historical interaction data for improving CRS. For this purpose, we propose a novel pre-training approach to integrating both item-based preference sequence (from historical interaction data) and attribute-based preference sequence (from conversation data) via pre-training methods. We carefully design two pre-training tasks to enhance information fusion between item- and attribute-based preference. To improve the learning performance, we further develop an effective negative sample generator which can produce high-quality negative samples. Experiment results on two real-world datasets have demonstrated the effectiveness of our approach for improving CRS., Comment: Accepted as CIKM short paper

Providing users with rich sensations is beneficial to enhance their immersion in Virtual Reality (VR) environments. Wetness is one such imperative sensation that affects users' sense of comfort and helps users adjust grip force when interacting with objects. Researchers have recently begun to explore ways to create wetness illusions, primarily on a user's face or body skin. In this work, we extended this line of research by creating wetness illusion on users' fingertips. We first conducted a user study to understand the effect of thermal and tactile feedback on users' perceived wetness sensation. Informed by the findings, we designed and evaluated a prototype---Mouille---that provides various levels of wetness illusions on fingertips for both hard and soft items when users squeeze, lift, or scratch it. Study results indicated that users were able to feel wetness with different levels of temperature changes and they were able to distinguish three levels of wetness for simulated VR objects. We further presented applications that simulated an ice cube, an iced cola bottle, and a wet sponge, etc, to demonstrate its use in VR.

Recently, significant progress has been made in sequential recommendation with deep learning. Existing neural sequential recommendation models usually rely on the item prediction loss to learn model parameters or data representations. However, the model trained with this loss is prone to suffer from data sparsity problem. Since it overemphasizes the final performance, the association or fusion between context data and sequence data has not been well captured and utilized for sequential recommendation. To tackle this problem, we propose the model S^3-Rec, which stands for Self-Supervised learning for Sequential Recommendation, based on the self-attentive neural architecture. The main idea of our approach is to utilize the intrinsic data correlation to derive self-supervision signals and enhance the data representations via pre-training methods for improving sequential recommendation. For our task, we devise four auxiliary self-supervised objectives to learn the correlations among attribute, item, subsequence, and sequence by utilizing the mutual information maximization (MIM) principle. MIM provides a unified way to characterize the correlation between different types of data, which is particularly suitable in our scenario. Extensive experiments conducted on six real-world datasets demonstrate the superiority of our proposed method over existing state-of-the-art methods, especially when only limited training data is available. Besides, we extend our self-supervised learning method to other recommendation models, which also improve their performance., Comment: Accepted as CIKM2020 long paper

Weakly supervised machine reading comprehension (MRC) task is practical and promising for its easily available and massive training data, but inevitablely introduces noise. Existing related methods usually incorporate extra submodels to help filter noise before the noisy data is input to main models. However, these multistage methods often make training difficult, and the qualities of submodels are hard to be controlled. In this paper, we first explore and analyze the essential characteristics of noise from the perspective of loss distribution, and find that in the early stage of training, noisy samples usually lead to significantly larger loss values than clean ones. Based on the observation, we propose a hierarchical loss correction strategy to avoid fitting noise and enhance clean supervision signals, including using an unsupervisedly fitted Gaussian mixture model to calculate the weight factors for all losses to correct the loss distribution, and employ a hard bootstrapping loss to modify loss function. Experimental results on different weakly supervised MRC datasets show that the proposed methods can help improve models significantly.

Verifying fact on semi-structured evidence like tables requires the ability to encode structural information and perform symbolic reasoning. Pre-trained language models trained on natural language could not be directly applied to encode tables, because simply linearizing tables into sequences will lose the cell alignment information. To better utilize pre-trained transformers for table representation, we propose a Structure-Aware Transformer (SAT), which injects the table structural information into the mask of the self-attention layer. A method to combine symbolic and linguistic reasoning is also explored for this task. Our method outperforms baseline with 4.93% on TabFact, a large scale table verification dataset.

This study proposes a tabletop projection device that can be implemented by combining physical objects with interactive projections. Users can interact on kraft papers using daily tools, such as marker pens, toothbrushes, colored blocks, and square wooden blocks. The input of the proposed device is a multifunction sensor, and the output is a tabletop projector. Using MagicPAPER, four types of interactions are implemented, namely drawing, gesture recognition, brushing, and building blocks. The abstract and poster discuss the design motivations and system descriptions of MagicPAPER.

Limb motion controlled robots have been developed and applied in many fields that affect every aspect of people's life. This paper presents a wearable massage robot named Anmoji, which is controlled by motions of a user's head, shoulders, and ankles. Anmoji is controlled in real time and able to provide functions including head massage, neck massage, back massage, leg massage, foot massage and multi-functional holder. Anmoji contributes a novel combination between massage and limb controlled robot, enabling a host of assistive interactions with the massage actions.

The emotion affect every aspect of our daily lives. Because of the high temporal resolution and low cost, EEG is widely used in the fields of emotion recognition. This paper studies the effects of different EEG feature combinations and channel selections on emotion recognition. Three effective features including differential entropy (DE), 1st difference (1st) and fractal dimension (FD) were extracted from the EEG signals, and their performances in the three situations of four emotion classifications, two emotions classification on valence and two emotions classification on arousal were calculated and compared by SVM. Two channel selection methods, including the mean relief channel selection algorithm and the common channel selection algorithm, were used to select the best channel. The results showed that when selecting the top 10 channels, the accuracy of the four emotional states classification rate was approximately 95%. This is significant for reducing the number of electrodes and reducing the complexity of brain-computer interface applications in the future.

This study considers an extensive coordinated capacitated dynamic lot-size problem that contains a delivery scheduling problem. The coordinated capacitated dynamic lot-size and delivery problem (CCLSDP) assumes that a family of products is subject to capacity constraints, and the demand without backorders is dynamic or time-varying but deterministic. The objective is to determine a replenishment and delivery schedule that minimizes the total system costs, including the family and minor setup costs, delivery costs, and inventory holding costs. We develop a six-phase heuristic (SPH) to solve the CCLSDP. The SPH contains six phases that can efficiently reschedule replenishment and delivery planning. Each phase of the SPH evaluates the cost and feasibility of the shifting operations, and determines whether to reschedule the system planning through a series of sequential shifting operations. To further enhance the overall performance, we embed SPH as a lifting procedure into evolutionary algorithms contributing to explore the search space and escapes the local optimum. Experimental results of randomly generated instances that were over a wide range of parameters indicate that the proposed heuristics could efficiently and effectively solve the complicated CCLSDP.

In the real business situation, suppliers usually provide retailers with forward financing to decrease inventory or increase demand. Moreover, some heterogeneous goods are not allowed to transport together, or a penalty cost is incurred when heterogeneous goods are transported at the same time. This research proposes a practical multi-item joint replenishment problem (JRP) by considering trade credit and grouping constraint in accordance with the practical situation. The JRP aims to find reasonable item replenishment frequencies and each group’s basic replenishment cycle time so that the overall cost can be minimized. Four intelligent algorithms, which include an advanced backtracking search optimization algorithm (ABSA), genetic algorithm (GA), differential evolution (DE) and backtracking search optimization algorithm (BSA), are provided to solve this problem. Findings of contrastive example verify that ABSA is superior to GA, DE, and BSA, which have been validated to be effective algorithms. Randomly generated problems are used to test the performance of ABSA. Results indicate ABSA is more effective and stable to resolve the proposed JRP than the other algorithms. ABSA is a good solution for the proposed JRP with heterogeneous items under trade credits.

As the most common writing material in our daily life, paper is an important carrier of traditional painting, and it also has a more comfortable physical touch than electronic screens. In this study, we designed a shadow-art device for human-computer interaction called MagicPAPER, which is based on physical touch detection, gesture recognition, and reality projection. MagicPAPER consists of a pen, kraft paper, and several detection devices, such as AirBar, Kinect, LeapMotion, and WebCam. To make our MagicPAPER more interesting, we developed more than a dozen applications that allow users to experience and explore creative interactions on a desktop with a pen and a piece of paper. Results of our user study showed that MagicPAPER has received positive feedback from many different types of users, particularly children.

As the most common writing material in our daily life, paper is an important carrier of traditional painting, and it also has a more comfortable physical touch than electronic screens. In this study, we designed a shadow-art device for human--computer interaction called MagicPAPER, which is based on physical touch detection, gesture recognition, and reality projection. MagicPAPER consists of a pen, kraft paper, and several detection devices, such as AirBar, Kinect, LeapMotion, and WebCam. To make our MagicPAPER more interesting, we developed thirteen applications that allow users to experience and explore creative interactions on a desktop with a pen and a piece of paper.

Touch interaction is usually implemented on a computer screen or a projected surface. However, tactile quality of paper is attractive and intuitive for people to manipulate in traditional paintings and storytelling. In order to enable touch interaction on paper, we present MagicPAPER, a shadow-art device that consists of a pen, kraft paper, a projector, a gesture sensing device, and a software system. With the purpose of illustrating how MagicPAPER effectively integrates the art-design process with the computer and communication technology, we have developed eight interactive applications simply based on kraft paper and mark pen: the Dimension Door, Balls, Palette, Flat Cube, Piano, Color potion, Fast mask- changing, and Candle. Finally, we experimented with MagicPAPER and demonstrated its interactive effects by detecting the gesture of a finger and pen on kraft paper and projecting it into eight interactive applications. Our research results show that MagicPAPER successfully carries out paper-based touch interactions.

Glass chip in liquid injection must be avoided according to the medicine regulations. We develop a system that can detect glass chip automatically. The system imitates the artificial glass chip inspection method which is done by swaying and vibrating the container. An industrial camera acquires a sequence of images during the process. The image data are sent to the computer and our algorithm analyses the data. Image binarization, mathematic morphological filtering and connected area analysis are used to detect glass chip candidate areas. Then we use optical flow method to detect moving object. Finally, we analyze the results comprehensively to determine whether there is glass chip in the current sample.