Search

Epilepsy is one of the most common neurological disorders. The X-linked gene PCDH19 is associated with sporadic and familial epilepsy in humans, typically with early-onset clustering seizures and intellectual disability in females but not in so-called ‘carrier’ males, suggesting that mosaic PCDH19 expression is required to produce epilepsy. To characterize the role of loss of PCDH19 function in epilepsy, we generated zebrafish with truncating pcdh19 variants. Evaluating zebrafish larvae for electrophysiological abnormalities, we observed hyperexcitability phenotypes in both mosaic and non-mosaic pcdh19+/- and pcdh19-/- mutant larvae. Thus, we demonstrate that the key feature of epilepsy—network hyperexcitability—can be modeled effectively in zebrafish, even though overt spontaneous seizure-like swim patterns were not observed. Further, zebrafish with non-mosaic pcdh19 mutation displayed reduced numbers of inhibitory interneurons suggesting a potential cellular basis for the observed hyperexcitability. Our findings in both mosaic and non-mosaic pcdh19 mutant zebrafish challenge the prevailing theory that mosaicism governs all PCDH19-related phenotypes and point to interneuron-mediated mechanisms underlying these phenotypes.

In this paper, we propose an adaptive sampling technique that achieves considerable energy savings while maintaining the required intelligibility level for mobile sign language video communication. The adaptive sampling scheme adjusts the sampling rate of the camera sensors dynamically according to the properties of sign language video communication and available battery power of mobile devices. Experimental results show that this adaptive scheme performs better than continuous sensing scheme in balancing the energy- intelligibility tradeoffs.

In this paper, we study the Quay Crane (QC) assignment for incoming vessels by introducing hard time windows into the container terminal. Given a nonlinear programming model, a set of 0-1 variables is employed to describe the starting position of the operation for each QC in a certain vessel. We also employ another set of 0-1 variables to locate the ending operation position for each QC in the same vessel. The research objective is to minimize the maximum operation time of each QC. This is a typical NP-hard problem, so a genetic algorithm is applied to the searching of the optimum solution in real-time applications. We carried out a series of experiments in order to evaluate our model. The results show that the proposed model and algorithm can signiflcantly raise the operating e‐ciency of the container terminal.

The primary challenge to design a mobile sign language communication system is overcoming the limitation of battery power. A scheme which allocates the computational resource of the encoder adaptive to available battery power and deaf people’s visual system is proposed. In the scheme, encoding levels which determine number of reference frames and search range are adaptively selected according to the battery power and frame complexity at frame level. Then possible partition mode and quantization parameter are adaptively adjusted at the macro block (MB) level according to the relative priority of each MB. Experimental results show that the proposed algorithm obtains better peak-signal-noise-rate of face and hands that improves the intelligibility of sign language video. Encoding time can be saved by up to 89.4% compared to the standard H.264 encoder. Increased bit rate of the proposed algorithm is less than 5%, which is negligible.

The primary challenge to perform mobile sign language communication over cellular network is overcoming the data loss. We proposed a new multiple descriptions coding (MDC) for protecting regions of interest (ROI) in sign language video by combining the advantages of MDC and ROI coding. The set partitioning in hierarchical trees (SPIHT) algorithm is used to generate MDC streams, and these streams are divided into ROI streams and background (BG) streams. For protection of ROI, our method assigns different redundant trees to the two kinds streams, the fully coded trees are assigned as the redundant trees of the ROI streams, the partially coded trees are assigned as the redundant trees of the BG streams. Experiment results show that our method can further increase the reconstruction quality of the ROI over noisy channels compared with other regions of interest multiple description coding methods.

The high computational complexity of H.264 poses as a challenge for it's feasibility in power-constrained devices. In this paper, the key modules of the encoder which can be power-optimized are analyzed at first. Based on the analysis results, a novel computation resource allocation method is proposed. The algorithm can allocate the computation resource of the encoder adaptive to available battery power and video contents. Experimental results show the proposed algorithm can highly reduce the computation resource while maintaining video encoding quality.

A new multiple description (MD) coding for protecting regions of interest (ROI) in images is proposed by combining the advantages of MD coding and ROI coding. The set partitioning in hierarchical trees (SPIHT) algorithm is used to generate MD streams, and these streams are divided into ROI streams and background (BG) streams. For protection of ROI, our method assigns different redundant trees to the two kinds streams, the fully coded trees are assigned as the redundant trees of the ROI streams, the partially coded trees are assigned as the redundant trees of the BG streams. Experiment results show that our method can further increase the reconstruction quality of the ROI over noisy channels compared with other regions of interest multiple description coding methods.

This paper investigates an interesting question of solving incremental learning problems using ensemble algorithms. The motivation is to help classifiers learn additional information from new batches of data incrementally while preserving previously acquired knowledge. Experimental results show that the proposed dynamic weighting scheme can achieve better performance compared to the fixed weighting scheme on a variety of standard UCI benchmark datasets.

This paper considers the consensus problem of continuous-time sensor network in a noisy environment, in which both fixed topology case and switching topology case are considered. The main idea is based on the fact that the state of each agent has twofold identity, i.e., the state of itself and the control input of agents that take it as a neighbor. With this in mind, we first solely design unbiased Kalman filters for each agent such that the sensor network solves consensus problem when all information is contaminated by noises. Then, the convergence of the proposed algorithms is proved. All the proposed algorithms are fully distributed, unbiased, and use the least information. The validity of them is illustrated by the included numerical examples.

Managing customer credit is an important issue in the banking industry and should always be done in an automatic way, with credit scoring trusted. This paper presents our solution to PAKDD 2009 data mining competition as a case study of the credit scoring problem. Following a brief description of the data mining task, several challenges confronted in the task such as imbalanced dataset, missing values and data transformation are discussed. After series of preliminary experiments, logistic regression and AdaBoost were shown as the resulting classifiers on this particular problem. Furthermore, an ensemble of the two classifiers was created in order to achieve even better performance. The final result shows that our solution is effective and efficient with an AUC value of 0.6535, which was the fifth best result among more than 100 competitive teams.