Search

Visual impairment community, especially blind people have a thirst for assistance from advanced technologies for understanding and answering the image. Through the development and intersection between vision and language, Visual Question Answering (VQA) is to predict an answer from a textual question on an image. It is essential and ideal to help blind people with capturing the image and answering their questions automatically. Traditional approaches often utilize the strength of convolution and recurrent networks, which requires a great effort for learning and optimizing. A key challenge in VQA is finding an effective way to extract and combine textual and visual features. To take advantage of previous knowledge in different domains, we propose BERT-RG, the delicate integration of pre-trained models into feature extractors, which relies on the interaction between residual and global features in the image and linguistic features in the question. Moreover, our architecture integrates a stacked attention mechanism that exploits the relationship between textual and visual objects. Specifically, the partial regions of images interact with partial keywords in question to enhance the text-vision representation. Besides, we also propose a novel perspective by considering a specific question type in VQA. Our proposal is significantly meaningful enough to develop a specialized system instead of putting forth the effort to dig for unlimited and unrealistic approaches. Experiments on VizWiz-VQA, a practical benchmark dataset, show that our proposed model outperforms existing models on the VizWiz VQA dataset in the Yes/No question type.

A conventional three-dimensional digital image correlation (DIC) based on painted speckle patterns has been used widely as a non-contact measurement method to study modal analysis of various structures. Surface treatment using painted speckle patterns might change the structural properties of flexible and lightweight structures such as artificial wings of flapping micro air vehicles. Furthermore, if materials and structures are being serviced, it is essential that their surfaces not be contaminated. We propose a new DIC method that is capable of preventing effects of painted speckle patterns on thin flexible structures during vibration measurement. We project a virtual speckle pattern onto the surface of structures instead of using painted speckle patterns. For a benchmark test, we demonstrate the effectiveness of the proposed virtual speckle pattern DIC method for measuring the structural characteristics of a cantilever beam. We then apply the proposed DIC method to the vibration measurement of an artificial flapping wing. Finite element analysis (FEA) of the artificial wing is performed in ABAQUS™ to obtain natural frequencies and mode shapes. The natural frequencies at the first three modes of the artificial wing obtained from the proposed method are higher than those obtained from the conventional 3D DIC because of the smaller weight of the unpainted wing. The results of the mass compensation, laser sensor, and finite element analysis prove the effectiveness of the proposed virtual speckle pattern DIC method. The proposed DIC method achieves high accuracy and prevents the effect of painted speckle patterns on the dynamic vibration measurement of a thin flexible structure such as an artificial flapping wing.

The haptic impression of textile surface properties has a decisive influence on its evaluation and ultimately on its acceptance and usability. Many solutions are used to replicate a static contour or shape, e.g. to feel controls on common touch displays. In contrast, this project investigates whether it is possible to simulate the roughness or friction behaviour of a textile surface using a commercially available mobile device.

This research proposes a modified metaheuristic optimization algorithm, named as improved stochastic fractal search (ISFS), which is formed based on the integration of the quasiopposition-based learning (QOBL) and chaotic local search (CLS) schemes into the original SFS algorithm for solving the optimal capacitor placement (OCP) in radial distribution networks (RDNs). The test problem involves the determination of the optimal number, location, and size of fixed and switched capacitors at different loading conditions so that the network total yearly cost is minimized with simultaneous fulfillment of operating constraints. Also, the hourly on/off scheduling plans of switched shunt capacitors (SCs) considering a modified cost objective function are obtained. The proposed ISFS algorithm has been tested on two IEEE 69-bus and 119-bus RDNs and a practical 152-bus RDN. For clarifying the effectiveness and validation of the ISFS, the simulated results have been compared with those of other previously utilized solution approaches in the literature as well as the original SFS. From result comparison, the proposed ISFS outperforms other previous approaches regarding solution quality and statistical performance for the compared cases, especially in the complex and large-scale networks. Notably, compared with the original SFS, the proposed ISFS shows a significantly better performance in all the tested cases.

This paper investigates the effect of real-time sensor data variances on humans performing straightforward assembly tasks in a Virtual Reality-based (VR-based) training system. A VR-based training system has been developed to transfer color and depth images, and constructs colored point clouds data to represent objects in real-time. Various parameters that affect sensor data acquisition and visualization of remotely operated robots in the real-world are varied. Afterward, the associated task performance is observed. Experimental results from 12 participants performed a total of 95 VR-guided puzzle assembly tasks demonstrated that a combination of low resolution and uncolored points has the most significant effect on participants’ performance. Participants mentioned that they needed to rely upon tactile feedback when the perceptual feedback was minimal. The most insignificant parameter determined was the resolution of the data representations, which, when varied within the experimental bounds, only resulted in a 5% average change in completion time. Participants also indicated in surveys that they felt their performance had improved and frustration was reduced when provided with color information of the scene.

Copyright 2021 IEEE. Published in ICASSP 2021 - 2021 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), scheduled for 6-11 June 2021 in Toronto, Ontario, Canada. Personal use of this material is permitted. However, permission to reprint/republish this material for advertising or promotional purposes or for creating new collective works for resale or redistribution to servers or lists, or to reuse any copyrighted component of this work in other works, must be obtained from the IEEE. Contact: Manager, Copyrights and Permissions / IEEE Service Center / 445 Hoes Lane / P.O. Box 1331 / Piscataway, NJ 08855-1331, USA. Telephone: + Intl. 908-562-3966.; International audience; Matrix factorization with sparsity constraints plays an important role in many machine learning and signal processing problems such as dictionary learning, data visualization, dimension reduction. Among the most popular tools for sparse matrix factorization are proximal algorithms, a family of algorithms based on proximal operators. In this paper, we address two problems with the application of proximal algorithms to sparse matrix factorization. On the one hand, we analyze a weakness of proximal algorithms in sparse matrix factorization: the premature convergence of the support. A remedy is also proposed to address this problem. On the other hand, we describe a new tractable proximal operator called Generalized Hungarian Method, associated to so-called k-regular matrices, which are useful for the factorization of a class of matrices associated to fast linear transforms. We further illustrate the effectiveness of our proposals by numerical experiments on the Hadamard Transform and magnetoencephalography matrix factorization.

A thermal protection system (TPS) is designed and fabricated to protect a hypersonic vehicle from extreme conditions. Good condition of the TPS panels is necessary for the next flight mission. A loose bolted joint is a crucial defect in a metallic TPS panel. This study introduces an experimental method to investigate the dynamic characteristics and state of health of a metallic TPS panel through an operational modal analysis (OMA). Experimental investigations were implemented under free-free supports to account for a healthy state, the insulation effect, and fastener failures. The dynamic deformations resulted from an impulse force were measured using a non-contact three-dimensional point tracking (3DPT) method. Using changes in natural frequencies, the damping ratio, and operational deflection shapes (ODSs) due to the TPS failure, we were able to detect loose bolted joints. Moreover, we also developed an in-house program based on a modal assurance criterion (MAC) to detect the state of damage of test structures. In a damage state, such as a loose bolted joint, the stiffness of the TPS panel was reduced, which resulted in changes in the natural frequency and the damping ratio. The calculated MAC values were less than one, which pointed out possible damage in the test TPS panels. Our results also demonstrated that a combination of the 3DPT-based OMA method and the MAC achieved good robustness and sufficient accuracy in damage identification for complex aerospace structures such as TPS structures.

This paper presents an integration of Virtual Reality (VR) interfaces with the control system of a real-world mobile manipulator, ultimately facilitating a natural and intuitive method for human-robot interaction. VR’s ability to track movements in 3D space and translate performed motions provide an intuitive platform for users to explore and interact with the virtual environment. Coupled with intuitive controls, such as grabbing and pointing, the VR platform provides a compelling advantage that can be used to solve limitations of traditional remote robot teleoperation methods. This paper summarises the system implemented, which includes a simulation of the robot in Unity3d, as well as analyses critical results of accuracy and performance, from experiments with users of various experience levels. The method used for measuring accuracy with a simulated robot presented a utilitarian validation for contrasting the difference between 2D and VR 3D interfaces. Users’ performance and experience under various levels of control latency, which is a crucial factor in remote online robot control, were also measured.

In this paper, the mathematical modelling problem of thermoacoustic generator systems is investigated. An electrical circuit based method is developed by using the physical principles for each parts. Then, the modified dynamic equations as well as the compact electrical circuit are proposed to consider the controller. The proposed electrical circuit of whole of thermoacoustic generator system is employed in the mathematical modelling, which leads to simple and convenient implementation of next control designs. Finally, the effectiveness of the developed mathematical modelling is verified through simulation results.

Visual Question Answering (VQA), the novel task among the intersection between Computer Vision (CV) and Natural Language Processing (NLP), extracts answers from features of both questions and images. The current approaches in VQA rely on the combination between convolution and recurrent networks, which leads to the huge number of parameters for learning phase. With the success of employing pre-trained models, we integrate BERT [1] for embedding text and two models: ResNets [2] and VGG [3] for embedding image. In addition, we also propose to take advantages of fine-tuning techniques and stacked attention mechanism to combine textual and visual features in a novel learning phase considered its ability to reduce the size of models. To demonstrate our model’s performance, we conduct experiments in the VizWiz VQA Challenge 2020. According to the experimental results, the proposed approach outperforms existing methods for Yes-No questions on VizWiz VQA dataset

Guaranteeing data transmission between modules is the key for application development of modular robotic systems. In a multi-channel modular robotic system, intersection modules play an essential role of communication channel selection in controlling data flow toward desired destinations. The gradient-based routing algorithm is an ideal solution to create an one-way communication path from any robotic module to a designated destination. To create bi-directional communication for a communication path of robotic configuration, virtual pheromone-based routing algorithm is a promising mechanism for intersection modules due to its simplicity and distributivity. In this paper, we address a virtual pheromone based network flow control based on the integration of gradient and virtual pheromone-based routing algorithms. We validated this method through an education and entertainment application using our newly developed modular robotic system.

This paper describes how Rule Input Network Generator (RING), a network generation computational tool, can be adopted to generate a variety of complex biochemical reaction networks. The reaction language incorporated in RING allows representation of chemical compounds in biological systems with various structural complexity. Complex molecules such as oligosaccharides in glycosylation pathways can be described using a simplified representation of their monosaccharide building blocks and glycosidic bonds. The automated generation and topological network analysis features in RING also allow for: (1) constructing biochemical reaction networks in a rule-based manner, (2) generating graphical representations of the networks, (3) querying molecules containing a particular structural pattern, (4) finding the shortest synthetic pathways to a user-specified species, and (5) performing enzyme knockout to study their effect on the reaction network. Case studies involving three biochemical reaction systems: (1) Synthesis of 2-ketoglutarate from xylose in bacterial cells, (2) N-glycosylation in mammalian cells, and (3) O-glycosylation in mammalian cells are presented to demonstrate the capabilities of RING for robust and exhaustive network generation and the advantages of its post-processing features.

Recently research in opinions summarization focuses on rating expressions by aspects and/or sentiments they carry. To extract aspects of an expression, most studies require a predefined list of aspects or at least the number of aspects. Instead of extracting aspects, we rate expressions by aspect similarity recognition (ASR), which evaluates whether two expressions share at least one aspect. This subtask relaxes the limitation of predefining aspects and makes our opinions summarization applicable in domain adaptation. For the ASR subtask, we propose an attention-cell LSTM model, which integrates attention signals into the LSTM gates. According to the experimental results, the attention-cell LSTM works efficiently for learning latent aspects between two sentences in both settings of in-domain and cross-domain. In addition, the proposed extractive summarization method using ASR shows significant improvements over baselines on the Opinosis corpus.

Neural networks (NNs) have archived high accuracy in many fields of machine learning such as image recognition. Since computations of NNs are heavy tasks, GPUs have been widely used to accelerate them. However, the problem sizes of NNs that can be computed are limited by GPU memory capacity.

In this paper, a novel approach of sensor placement is proposed for the purpose of maximizing fault detectability and isolability. This new approach rests on the basic fact that faults are embedded in the analytical redundancy relations (ARRs) and that the occurrence of a fault will change the consistency of the corresponding ARRs. Based on these basic facts, the minimal isolating (MI) set is introduced to formulate the full/maximal isolability which is the constraint for sensor placement. Consequently, the optimization problem for sensor placement is reformulated as searching an MI set which is related to the least number of candidate sensors. To find the optimal MI set, a low complexity dynamic programming (LCDP) algorithm is developed on the fault set ${\cal F}$ that consists of system faults and sensor faults. However, sensor faults are varied as different candidate sensors are used. Therefore, another dedicated procedure is proposed to handle this issue. A case study shows that the proposed approach outperforms an existing sensor placement approach in terms of efficiency.

ive Summarization is the specific task in text generation whose popular approaches are based on the strength of Recurrent Neural Network. With the purpose to take advantages of Convolution Neural Network in text representation, we propose to combine these above networks in our encoder to capture both the global and local features from the input documents. Simultaneously, our model also integrates the reinforced mechanism with the novel reward function to get the closer direction between the learning and evaluating process. Through the experiments in CNN/Daily Mail, our models gains the significant results. Especially, in ROUGE-1 and ROUGE-L, it outperforms the previous works in this task with the expressive improvement (39.09% in ROUGE-L F1-score).

Using modular robotics enables an innovative learning and teaching method in education and entertainment. In this paper, we present our unique feature modular robotic system, named moreBot. The robotic system can be used as an educational and entertaining tool for numerous “learning through playing” scenarios. We describe the design, implementation of the moreBot and demonstrate the key characteristics through a number of education and entertainment applications.

Variable structure control with sliding mode can provide good control performance and excellent robustness. Unfortunately, the chattering phenomenon investigated due to discontinuous switching gain restricting their applications. In this paper, a chattering free improved variable structure control (IVSC) for a class of mismatched uncertain interconnected systems with an unknown time-varying delay is proposed. A sliding function is first established to eliminate the reaching phase in traditional variable structure control (TVSC). Next, a new reduced-order sliding mode estimator (ROSME) without time-varying delay is constructed to estimate all unmeasurable state variables of plants. Then, based on the Moore-Penrose inverse approach, a decentralized single-phase robustness sliding mode controller (DSPRSMC) is synthesized, which is independent of time delays. A DSPRSMC solves a complex interconnection problem with an unknown time-varying delay term and drives the system’s trajectories onto a switching surface from the initial time instance. Particularly, by applying the well-known Barbalat’s lemma, the chattering phenomenon in control input is alleviated. Moreover, a sufficient condition is established by using an appropriate Lyapunov theory and linear matrix inequality (LMI) method such that a sliding mode dynamics is asymptotically stable from the beginning time. Finally, a developed method is validated by numerical example with computer simulations.

The key task in extractive summarization is to determine the importance of the sentence in the input. Several recent studies have focused on comparing the similarity between sentences to assess the significance of them efficiently. Each comparison method has its strengths and weaknesses. In this paper, we propose the combination of similarity measures for sentence comparison. Experiments conducted on both English and Vietnamese datasets demonstrate the efficiency of our proposed approach. Our model outperforms the recent works in English with the significant improvement (9.4 ROUGE-2 F1-score) and achieves the competitive result in Vietnamese.

Neighbourhood connection and network topology in modular robotic systems are frequently varied by users. Inter-communication of modular robotic systems is a challenge due to uncertainties of network topology. In this paper, we present a novel distributed communication protocol for modular robotic systems. The protocol is designed to deal with rapid changes of system reconfiguration. We demonstrated the effectiveness of the developed protocol through experiments on the real-world modular robotic system.

Cross-language summarization is the novel topic which is extremely practical and necessary for capturing, tracing, and retrieving the huge data. Especially, for many low-resource languages as Vietnamese, Chinese, ..., there are not any previous works to solve this problem as well as datasets. Therefore we propose to apply Phrase-based Compressive Summarization for English-Vietnamese. This model takes advantages of the relation between translation and summarization phases to overcome the popular drawback in most antecedent researches. Besides, the bilingual corpus for English-Vietnamese summarization built manually on the dataset is extremely helpful for a lot of later works. In this dataset, our system achieves approximately 37 % in ROUGE-1 score which is equivalent to systems on other language pairs. This significant and encouraging result proves the effectiveness of our approach and the quality of our manual datasets in English-Vietnamese.

The paper present a strategy of artificial lighting control to respond the natural state of daylight based on fuzzy logic control. After building the fuzzy logic controller, the simulations using Matlab program in two scenarios are built. For each scenario, the output responses were surveyed with the whole range of daylight corresponding to each of input statement. The simulation results show that the output of the fuzzy controller is good responsive the equilibrium when the input illumination changes, while the 4 zones scenario have more sensitive result than the 3 zones scenario.

Textiles are normally incompressible: when we try to compress them, they immediately buckle. Unfortunately, many cloth simulation solvers disregard this fact. In this paper, we present an efficient method to model buckling using distance contraint. This constraint is formulated as a linear complementarity problem (LCP) and can be easily integrated within a collision handling process.

We applied the log function of the queues into Approximate Dynamic Programming to build a better single-intersection traffic light controller. We then use a distributed network of this controller to control multiple intersection. The simulated results on a big road network showed that the log function is more effective than the direct queues.

BitTorrent is one of the dominant traffic generating applications in the Internet today. The ability to identify BitTor-rent traffic in real-time could allow network operators to better manage network traffic and provide a better service to their customers. In this paper we analyse the statistical properties of BitTorrent traffic and select four features that can be used for real-time classification using Machine Learning techniques. We then train and test a classifier using the C4.5 algorithm. Our results show that based on statistics calculated on 150-packet sub-flows, we can classify BitTorrent traffic with Recall of 98.2% and Precision of 96.5%. We then show that 98.1% of sub-flows from other client-server bulk transfer applications are correctly classified as non-BitTorrent.

In this paper, the frequently disconnected problem in VANETs is solved by a novel cell area design with decision points at cell borders. At decision points, a car gets local road information under inter-vehicle communication (IVC), otherwise roadside-to-vehicle communication (RVC). Moreover, in order to decrease the usage of RVC, we utilize a transfer estimation based on low level carrier sense with large carrier sense range (CSR). Simulation results confirm the reduction of RVC by applying low level carrier sense.

In this paper we present a method for fast energy minimization of virtual garments. Our method is based upon the idea of multiresolution particle system. When garments are approximately positioned around a virtual character, their spring energy may be high, which will cause instability or at least long execution time of the simulation. An energy minimization algorithm is needed; if a fixed resolution is used, it will require many iterations to reduce its energy. Even though the complexity of each iteration is O(n), with a high resolution mass-spring system, this minimization process can take a whole day. The hierarchical method presented in this paper is used to reduce significantly the execution time of the minimization process. The garments are firstly discretized in several resolutions. Once the lowest resolution particles system is minimized (in a short time), a higher resolution model is derived, then minimized. The procedure is iterated up to the highest resolution. But at this stage, the energy to minimize is already much lower so that minimization takes a reasonable time.

In this paper, we report on our research and progress in Vietnamese Broadcast News transcription, with an emphasis on efficient modeling for more accurate recognition. In the acoustic modeling area, this was achieved through a re-alignment process, which considers all pronunciations for each word and outputs the pronunciation that best matches the acoustic data. The effectiveness of acoustic adaptation is greatly increased through unsupervised clustering of test data. In language modeling, we explored the use of non-broadcast-news training data as well as the adaptation to topic. Experimental results showed significant improvements in which the achieved WAR measured on a 1h test set was 84.2%, which gained absolutely 5.4% improvement over the baseline result (Nguyen and Vu, 2006 and Huynh et al., 2005).

This paper presents the testing methodology of the UltraSPARC T1 microprocessor and the challenges to productizing the industry's first 8 cores and 32 thread microprocessor. The challenges include effectively testing 8 individual cores that share common functional blocks (L2 cache, cache crossbar, floating point unit), reducing 8times conventional production test time and tester memory while achieving the same coverage, and productizing 6 core or 8 core devices at the targeted frequency. Highlights of DFT features designed to overcome the challenges are presented followed by a production flow together with key learning and future improvement plans

In this paper, we present an easy method for automatic 3D garment pre-positioning. Given a digital 3D mannequin and a set of 2D patterns, the problem consists in sewing automatically the patterns around the mannequin and give an initial position of the garment that the mannequin has to wear. Our method is based upon the use of a figurine. A figurine of a virtual garment, which is always present in the garment design, is associated to the 2D patterns used for producing the garment. This figurine is a 2D drawing which gives the flavor and the style of the future garment. The figurine is the key element of the technique presented in this paper. It can be stored in a 2D CAD file with the 2D patterns of garments or in a database. In this method, we use two figurines, corresponding to the front and the back view of the future garment and a standard silhouette. The designer just needs to work once on the garment construction to allow the automatic pre-positioning. Once the designer has specified the links between the 2D garments and the figurine, he has just to take some simple actions on the figurine to complete the pre-positioning. Our method provides a mapping algorithm from the standard silhouette to a silhouette of a future client, which is the starting point of the pre-positioning. When finalized the automatic pre-positioning will help the designer to see directly the garment he invented on a “real” body.