Efficient duration and hierarchical modeling for human activity recognition

A challenge in building pervasive and smart spaces is to learn and recognize human activities of daily living (ADLs). In this paper, we address this problem and argue that in dealing with ADLs, it is beneficial to exploit both their typical duration patterns and inherent hierarchical structures. We exploit efficient duration modeling using the novel Coxian distribution to form the Coxian hidden semi-Markov model (CxHSMM) and apply it to the problem of learning and recognizing ADLs with complex temporal dependencies. The Coxian duration model has several advantages over existing duration parameterization using multinomial or exponential family distributions, including its denseness in the space of nonnegative distributions, low number of parameters, computational efficiency and the existence of closed-form estimation solutions. Further we combine both hierarchical and duration extensions of the hidden Markov model (HMM) to form the novel switching hidden semi-Markov model (SHSMM), and empirically compare its performance with existing models. The model can learn what an occupant normally does during the day from unsegmented training data and then perform online activity classification, segmentation and abnormality detection. Experimental results show that Coxian modeling outperforms a range of baseline models for the task of activity segmentation. We also achieve a recognition accuracy competitive to the current state-of-the-art multinomial duration model, while gaining a significant reduction in computation. Furthermore, cross-validation model selection on the number of phases K in the Coxian indicates that only a small K is required to achieve the optimal performance. Finally, our models are further tested in a more challenging setting in which the tracking is often lost and the activities considerably overlap. With a small amount of labels supplied during training in a partially supervised learning mode, our models are again able to deliver reliable performance, again with a small number of phases, making our proposed framework an attractive choice for activity modeling.     

Information in the nonstationary case

Information estimates such as the direct method of Strong, Koberle, de Ruyter van Steveninck, and Bialek (1998) sidestep the difficult problem of estimating the joint distribution of response and stimulus by instead estimating the difference between the marginal and conditional entropies of the response. While this is an effective estimation strategy, it tempts the practitioner to ignore the role of the stimulus and the meaning of mutual information. We show here that as the number of trials increases indefinitely, the direct (or plug-in) estimate of marginal entropy converges (with probability 1) to the entropy of the time-averaged conditional distribution of the response, and the direct estimate of the conditional entropy converges to the time-averaged entropy of the conditional distribution of the response. Under joint stationarity and ergodicity of the response and stimulus, the difference of these quantities converges to the mutual information. When the stimulus is deterministic or nonstationary the direct estimate of information no longer estimates mutual information, which is no longer meaningful, but it remains a measure of variability of the response distribution across time.     

SVM for English semantic classification in parallel environment

Semantic analysis is very important and very helpful for many researches and many applications for a long time. SVM is a famous algorithm which is used in the researches and applications in many different fields. In this study, we propose a new model using a SVM algorithm with Hadoop Map (M)/Reduce (R) for English document-level emotional classification in the Cloudera parallel network environment. Cloudera is also a distributed system. Our English testing data set has 25,000 English documents, including 12,500 English positive reviews and 12,500 English negative reviews. Our English training data set has 90,000 English sentences, including 45,000 English positive sentences and 45,000 English negative sentences. Our new model is tested on the English testing data set and we achieve 63.7% accuracy of sentiment classification on this English testing data set.     

A decision tree using ID3 algorithm for English semantic analysis

Natural language processing has been studied for many years, and it has been applied to many researches and commercial applications. A new model is proposed in this paper, and is used in the English document-level emotional classification. In this survey, we proposed a new model by using an ID3 algorithm of a decision tree to classify semantics (positive, negative, and neutral) for the English documents. The semantic classification of our model is based on many rules which are generated by applying the ID3 algorithm to 115,000 English sentences of our English training data set. We test our new model on the English testing data set including 25,000 English documents, and achieve 63.6% accuracy of sentiment classification results.     

Examining the Effectiveness of Gamification in Human Computation

Within the human computation paradigm, gamification is increasingly gaining interest. This is because an enjoyable experience generated by game features can be a powerful approach to attract participants. Although potentially useful, little research has been conducted into understanding the effectiveness of gamification in human computation. In this experimental study, we operationalized effectiveness as perceived engagement and user acceptance and examined it by comparing the performance of a gamified human computation system against a non-gamified version. We also investigate the determinants of acceptance and how their effects differ between these two systems. Analysis of our data found that participants experienced more engagement and showed higher behavioral intentions toward the gamified system. Moreover, perceived output quality and perceived engagement were significant determinants of acceptance of the gamified system. In contrast, determinants for acceptance of the non-gamified system were perceived output quality and perceived usability.     

BSPNN: boosted subspace probabilistic neural network for email security

In the modern age of Internet connectivity, advanced information systems have accumulated huge volumes of data. Such fast growing, tremendous amount of data, collected and stored in large databases has far exceeded our human ability to comprehend without proper tools. There has been a great deal of research conducted to explore the potential applications of Machine Learning technologies in Security Informatics. This article studies the Network Security Detection problems in which predictive models are constructed to detect network security breaches such as spamming. Due to overwhelming volume of data, complexity and dynamics of computer networks and evolving cyber threats, current security systems suffer limited performance with low detection accuracy and high number of false alarms. To address such performance issues, a novel Machine Learning algorithm, namely Boosted Subspace Probabilistic Neural Network (BSPNN), has been proposed which combines a Radial Basis Function Neural Network with an innovative diversity-based ensemble learning framework. Extensive empirical analyses suggested that BSPNN achieved high detection accuracy with relatively small computational complexity compared with other conventional detection methods.     

Frequency multiplexed superconducting quantum interference device readout of large bolometer arrays for cosmic microwave background measurements

A technological milestone for experiments employing transition edge sensor bolometers operating at sub-Kelvin temperature is the deployment of detector arrays with 100s-1000s of bolometers. One key technology for such arrays is readout multiplexing: the ability to read out many sensors simultaneously on the same set of wires. This paper describes a frequency-domain multiplexed readout system which has been developed for and deployed on the APEX-SZ and South Pole Telescope millimeter wavelength receivers. In this system, the detector array is divided into modules of seven detectors, and each bolometer within the module is biased with a unique ～MHz sinusoidal carrier such that the individual bolometer signals are well separated in frequency space. The currents from all bolometers in a module are summed together and pre-amplified with superconducting quantum interference devices operating at 4 K. Room temperature electronics demodulate the carriers to recover the bolometer signals, which are digitized separately and stored to disk. This readout system contributes little noise relative to the detectors themselves, is remarkably insensitive to unwanted microphonic excitations, and provides a technology pathway to multiplexing larger numbers of sensors.     

Using retina modelling to characterize blinking: comparison between EOG and video analysis

A large number of car crashes are caused by drowsiness every year. The analysis of eye blinks provides reliable information about drowsiness. This paper proposes to study the relation between electrooculogram (EOG) and video analysis for blink detection and characterization. An original method to detect and characterize blinks from a video analysis is presented here. The method uses different filters based on the human retina modelling. A illumination robust filter is first used to normalize illumination variations of the video input. Then, Outer and an Inner Plexiform Layer filters are used to extract energy signals from eye area. The eye detection is processed mixing gradient and projection methods which makes it able to detect even closed eyes. The illumination robust filter makes it possible to detect the eyes even in night conditions, without using external lighting. The video analysis extracts two signals from the eye area measuring the quantity of static edges and moving edges. Blinks are then detected and characterized from these two signals. A comparison between the features extracted from the EOG and the same features extracted from the video analysis is then performed on a database of 14 different people. This study shows that some blink features extracted from the video are highly correlated with their EOG equivalent: the duration, the duration at 50%, the frequency, the percentage of eye closure at 80% and the amplitude velocity ratio. The frame rate influence on the accuracy of the different features extracted is also studied and enlightens on the need of a high frame rate camera to detect and characterize accurately blinks from a video analysis.