Weishan Zhang

Emotion Recognition from Chinese Speech for Smart Affective Services Using a Combination of SVM and DBN

Accurate emotion recognition from speech is important for applications like smart health care, smart entertainment, and other smart services. High accuracy emotion recognition from Chinese speech is challenging due to the complexities of the Chinese language. In this paper, we explore how to improve the accuracy of speech emotion recognition, including speech signal feature extraction and emotion classification methods. Five types of features are extracted from a speech sample: mel frequency cepstrum coefficient (MFCC), pitch, formant, short-term zero-crossing rate and short-term energy. By comparing statistical features with deep features extracted by a Deep Belief Network (DBN), we attempt to find the best features to identify the emotion status for speech. We propose a novel classification method that combines DBN and SVM (support vector machine) instead of using only one of them. In addition, a conjugate gradient method is applied to train DBN in order to speed up the training process. Gender-dependent experiments are conducted using an emotional speech database created by the Chinese Academy of Sciences. The results show that DBN features can reflect emotion status better than artificial features, and our new classification approach achieves an accuracy of 95.8%, which is higher than using either DBN or SVM separately. Results also show that DBN can work very well for small training databases if it is properly designed.

A Realtime Framework for Video Object Detection with Storm

Real-time response is a challenging issue for video object detection, especially when the number of cameras is large and correspondingly the video data are big. The existing solutions for object detection fall short in addressing the real-time performance aspect, and can not handle fast response requirements such as fleeing vehicle tracking at run time. Therefore, in this paper we propose a Storm-based real-time framework for video object detection that can scale to handle large number of cameras. To evaluate its performance, we implement the framework in a Storm cluster environment where we test the detection rate and real-time performance of the framework. The results show that the detection rate is relatively acceptable and real-time response is achieved.

A genetic algorithm-based job scheduling model for big data analytics

Big data analytics (BDA) applications are a new category of software applications that process large amounts of data using scalable parallel processing infrastructure to obtain hidden value. Hadoop is the most mature open-source big data analytics framework, which implements the MapReduce programming model to process big data with MapReduce jobs. Big data analytics jobs are often continuous and not mutually separated. The existing work mainly focuses on executing jobs in sequence, which are often inefficient and consume high energy. In this paper, we propose a genetic algorithm-based job scheduling model for big data analytics applications to improve the efficiency of big data analytics. To implement the job scheduling model, we leverage an estimation module to predict the performance of clusters when executing analytics jobs. We have evaluated the proposed job scheduling model in terms of feasibility and accuracy.

Food Image Recognition with Convolutional Neural Networks

In this paper, we propose a food image recognition system with convolutional neural networks(CNN), which has been applied to image recognition successfully in the literature. A CNN which consists of five layers has been built and two group of controlled trials have been processed on it. Two datasets are prepared: one is UEC-FOOD100 dataset which is an open 100-class food image dataset including about 15000 images and the other is a fruit dataset that established by ourselves including over 40000 images. We have achieved the best accuracy of 80.8% on the fruit dataset and 60.9% on the multi-food dataset. In addition, we validate the method on two groups of controlled trials and discover the effect of color under various conditions that the color feature is not always helpful for improving the accuracy by comparing the results of two group of controlled trials. As future work, we will combine image segmentation with image recognition to get a better performance.

Emotion Recognition in Speech Using Multi-classification SVM

In order to improve the accuracy of emotion recognition in speech effectively, this paper proposes an emotion recognition algorithm based on SVM classification algorithm. Firstly, we use the SVM multi-class classification algorithm to optimize the parameters of penalty factor and kernel function. Then we use the optimized parameters to realize emotion recognition. Finally we obtain the accuracy of each kind of emotion using the Chinese emotional data set, using a variety of multi classification algorithm based on SVM. The emotion recognition can reach the highest rate of 96.00%.

An Integrated Approach for Vehicle Detection and Type Recognition

Vehicle detection and type recognition are important for intelligent transportation systems in smart cities. The real time high accuracy recognition with affordable hardware is a challenging issue due to the complexities of video data. In this paper, we propose an integrated approach that combining traditional three-frame difference and deep Convolutional Neural Networks (DCNNs) to detect vehicle and recognize vehicle type in traffic videos captured with fixed mounted cameras. This integrated approach can take advantage of the real-time motion detection ability of three-frame difference and capabilities of image recognition of DCNNs. We have evaluated the proposed approach using road traffic videos in terms of accuracy and performance, which show very promising results.

A MapReduce Cluster Deployment Optimization Framework with Geo-distributed Data

Big Data processing has become the common business needs in government and enterprise applications, e.g., Analysis or detection of climate change, economic development, or online customer behavior. Hadoop is the most mature open source big data processing framework, which implements the MapReduce programming paradigm. The mass source data are stored in HDFS supported by Hadoop and processed parallelly in computing nodes of a cluster. However, in many cases, the source data is simultaneously distributed across multiple data centers(Geo-distributed). Existing deployment research, merely focusing on moving all data to one data center to process, is often limited by the size of input data and the network transmission capacity between data centers, resulting in a lethal impact on the performance of big data processing. In this paper, we deal with Geo-distributed data sets, analyze possible cluster deployment way and then select the optimal one with the proposed cluster deployment optimization framework. We introduce decision making algorithm that the optimization framework relies on to determine an optimized cluster deployment way. In addition, we prove the benefit of our optimization framework by final experiment in Amazon EC2 over the common deployment for Geo-distributed data. The results show that the decision making algorithm is accurate and the optimization framework can significantly improve the Geo-distributed data processing performance by giving the optimized cluster deployment way.

A Tail-Tolerant Cloud API Wrapper

System operations (such as deployment, upgrade, and reconfiguration) for cloud applications are failure prone. These failures occur because these operations are performed through cloud APIs provided by cloud providers and because cloud APIs, in turn, are failure prone. Researchers have explored the characteristics of cloud APIs using Amazon EC2 (Elastic Compute Cloud) as a testbed and have devised mechanisms to improve cloud API performance. Specifically, mining the Amazon EC2 discussion forum revealed that 45 percent of complaints referred to cloud API timing failures. A series of experiments on cloud API timing behavior showed that cloud APIs have a long-tail distribution. A proposed cloud API wrapper implements mechanisms to avoid long tails. In experiments, this wrapper largely removed long tails, compared with the unwrapped APIs.

CF4BDA: A Conceptual Framework for Big Data Analytics Applications in the Cloud

Building big data analytics (BDA) applications in the cloud introduces inevitable challenges, such as loss of control and uncertainty. To address the existing challenges, numerous efforts have been made on BDA application engineering to optimize the quality of BDA applications in the cloud, such as performance and reliability. However, there is still a lack of systematic view on engineering BDA applications in the cloud. Therefore, in this paper, we present a conceptual framework named CF4BDA to analyze the existing work on BDA applications from two perspectives: 1) the lifecycle of BDA applications and 2) the objects involved in the context of BDA applications in the cloud. The framework can help researchers and practitioners identify the research opportunities in a structured way and guide implementing BDA applications in the cloud. We perform a preliminary evaluation of the usefulness of CF4BDA by applying it to analyze a set of representative studies.

PARMTRD: Parallel Association Rules Based Multiple-Topic Relationships Detection.

Lots of events happened everyday make social big data have plenty of topics. A topic usually comprises a series of stories. Clues of associations among stories are usually clear, but hidden associations among topics are not always intuitive. It is challenging to find topic associations due to intrinsic complexities of social big data, while analyzing relationships among topics is valuable to explore and reach to origination sources of specific events. Existing research rarely pay attention to analyze multiple-topic relationships. This paper proposes a mining approach for topic relationships detection based on parallel association rules, namely PARMTRD (Parallel Association Rules based Multiple-Topic Relationships Detection). PARMTRD obtains association keyword sets for each topic using parallel association rules based on large-scale frequent keyword sets, which mines association rules for multiple topics in parallel. PARMTRD detects the relevance among multiple topics by selecting and assembling association keywords from association keyword sets, which help to find sources of events. Experiments show that PARMTRD can detect the hidden relationships among multiple topics accurately and efficiently.