Susan Elias

Hierarchical activity recognition for dementia care using Markov Logic Network

Statistics reveal that globally, the aging population in different stages of dementia are struggling to cope with daily activities and are progressively becoming dependent on care takers thereby making dementia care a challenging social problem. Healthcare systems in smart environments that aim to address this growing social need require the support of technology to recognize and respond in an ubiquitous manner. To incorporate an efficient activity recognition and abnormality detection system in smart environments, the routine activities of the occupant are modeled and any deviation from the activity model is recognized as abnormality. Recognition systems are generally designed using machine learning strategies and in this paper a novel hybrid, data and knowledge-driven approach is introduced. Markov Logic Network (MLN) used in our design is a suitable approach for activity recognition as it integrates common sense knowledge with a probabilistic model that augments the recognition ability of the system. The proposed activity recognition system for dementia care uses a hierarchical approach to detect abnormality in occupant behavior using MLN. The abnormality in the context of dementia care is identified in terms of factors associated with the activity such as objects, location, time and duration. The task of recognition is done in a hierarchical manner based on the priority of the factor that is associated with each layer. The motivation for designing a hierarchical approach was to enable each layer to commence its computation after inferring from the lower layers that the ongoing activity was normal with regard to the associated factors. This hierarchical feature enables quick decisions, as factors that require immediate attention are processed first at the lowest layer. Experimental results indicate that the hierarchical approach has higher accuracy in recognition and efficient response time when compared to the existing approaches.

A REST based design for Web of Things in smart environments

Any physical Thing is a smart object, if it is embedded with sensor(s), actuator(s), a microcontroller and a low power radio. More often these components can be fabricated in chip, as small as the size of a human thumb finger. The main vision of “Web of Things” (WoTs) is to realize connectivity of the Web to billions of such smart objects. Applications can be built on top of standards like IPv6 over Low Power Wireless Personal Area Networks (6LoWPAN) at the L3 layer and IEEE 802.15.4 at the L2 layer of the network stack. Some of the popular applications include smart metering, building and home automation, healthcare, asset management, environmental monitoring and industrial automation. One of the main challenges for application developers is interoperability among the services provided by smart objects of disparate environments. This paper deals with interoperability at the application layer. We present an approach using the REST principles, for smart plant-watering application. The proposed approach can be adapted to any kind of applications involving smart objects. The paper also specifies an efficient syndication algorithm for event notifications from Things. We empirically show that our proposed approach achieves interoperability as well as performs efficiently.

Probabilistic ontology based activity recognition in smart homes using Markov Logic Network

Designing an activity recognition system that models various activities of an occupant is the fundamental task in creating a smart home. Activity Recognition (AR) modeling, has witnessed a comprehensive range of research, that focuses independently on probabilistic approaches and on ontology based models as well. The research presented in this paper introduces an innovative approach in AR system design that integrates probabilistic inference with the represented domain ontology. Data obtained from sensors are uncertain in nature and mapping uncertainty over ontology will not yield good accuracy in the context of AR. The proposed system augments ontology based activity recognition with probabilistic reasoning through Markov Logic Network (MLN) which is a statistical relational learning approach. The proposed system utilizes the model theoretic semantic property of description logic, to convert the represented ontology activity model to its corresponding first order rules. MLN is constructed by learning weighted first order rules that enable probabilistic reasoning within a knowledge representation framework. The experiments based on datasets obtained from smart home prototypes illustrate the effectiveness of integrating probabilistic reasoning over domain ontology and the result analysis shows enhanced recognition accuracy in comparison with existing approaches.

Dynamic consistency checking for temporal and spatial relations in multimedia presentations

Creating complex multimedia presentations involves the specification of temporal and spatial relations in the form of constraints. However, some of these constraints could contradict each other and hence lead to an inconsistency. The user may not be aware of this inconsistency while authoring. Hence this inconsistency has to be identified and removed by the presentation process prior to the play-out. In this paper, we examined an existing work based on graph theory for consistency checking. We propose a modification to this approach which simplifies the algorithm, reduces its total running time, and helps to make it dynamic. Another salient feature of our paper is the introduction of new temporal and spatial operators with higher expressive power than traditional ones. Thus, this paper presents a multimedia presentation mechanism, which dynamically maintains a consistent and complete set of constraints during authoring and play-out of the presentation.

An Ontology and Pattern Clustering Approach for Activity Recognition in Smart Environments

Activity recognition aims at modeling the occupants’ behavior by analyzing the sensor data collected from the smart environment. Though most of the activity recognition systems use supervised learning techniques for building such models there is a shift towards the unsupervised learning paradigm as the process of annotating and labeling the data is prone to errors. This paper proposes an Event Pattern Activity Modeling Framework (EPAM) to identify the occupant activity pattern from the sensor data by using an unsupervised machine learning approach and further analysis is done with a knowledge driven approach. In the context of smart environments, an activity is considered as a sequence of events that are generated continuously from the sensor data. The segmentation algorithm proposed in EPAM is used to identify appropriate event patterns for an activity that are then grouped together using a pattern clustering algorithm that presents a hierarchy of activities. The set of activities of the occupant, observed in a smart environment is not always sequential but is highly interleaved and discontinuous. The proposed algorithm accommodates this valid factor by an innovative use of the Jaro Winkler similarity measure. The hierarchy of activity generated by the pattern clustering approach is used for activity modeling. Ontology based activity modeling is preferred over other modeling techniques because of its unified modeling, representation and semantically clear reasoning. The experimental results show that the proposed EPAM framework of segmentation, pattern clustering and ontological modeling is efficient and more effective than the existing approach of activity modeling.

Composite Activity Recognition in Smart Homes Using Markov Logic Network

Smart environments have progressed and evolved into a significant research area with development of sensor technology, wireless communication and machine learning strategies. Ambient Intelligence incorporated into smart environment assists in resolving many social related applications to facilitate the future society. The initiative of modeling Activity of Daily Living (ADL) and Ambient Assisted Living (AAL) in smart homes have helped in the deployment of applications to various domains like elderly care, health care etc. Activity recognition is the task involved in reasoning within smart homes with the aim of recognizing the ongoing activity of the occupant. Constructing an activity model is essential to carry out recognition and is achieved through various machine learning and artificial intelligence techniques. Data driven approach constructs activity model through statistical machine learning mechanisms while knowledge driven approach constructs activity model through knowledge representation and modeling strategies. Uncertainty and temporal data are better handled by data driven approach while re-usability and context based analysis is handled better by knowledge driven approach of activity modeling. To combine the features of data driven and knowledge driven approaches, a hybrid activity modeling technique is required. The proposed system performs activity modeling via Markov Logic Network, a machine learning strategy that combines probabilistic reasoning and logical reasoning with a single framework. Activities in a smart home are categorized as simple and composite activities, wherein composite activities are defined as related simple activities within a given time interval. The proposed system models both simple and composite activity using soft and hard rules of MLN. Experiments carried over the proposed system shows the effectiveness of the proposed work for recognizing simple and composite activity.

Personalized e-course composition approach using digital pheromones in improved particle swarm optimization

One of the uphill tasks associated with the authoring of e-courses, for e-learning systems, is that the current composition techniques do not support ‘personalized-learning’ or in other words, the current composition methods fail to take into consideration the difference in individual learning capabilities and the background knowledge of the individual learners, which do not provide materials that exactly meet the demands of the individual learners. In order to provide solution for this problem, in the past, various e-course composition approaches had been proposed to use various methods of computational optimization techniques like genetic algorithm and particle swarm optimization. This paper proposes an improved personalized e-course composition approach based on modified particle swarm optimization algorithm along with digital pheromones. The final results of our ongoing research in this area, is furnished in this paper. Results of the various simulation-based experiments that have been conducted are furnished at the end of this paper. These results demonstrate that our proposed approach is an effective solution to the problem of ‘personalized learning’. In addition, our proposed approach is compared with the existing approaches, which uses Basic particle swarm optimization algorithm (BPSO) and modified PSO algorithm. These comparisons demonstrate that our proposed model is more efficient than others.

Referral based expertise search system in a time evolving social network

To solve some difficult problems that requires procedural knowledge, people often seek the advice of experts who have got competence in that problem domain. This paper focuses on locating and determining an expert in a particular knowledge domain. In most cases, social network of a user is explored through referrals to locate human experts. Past work in searching for experts through referrals focused primarily on static social network. However, static social network fail to accurately represent the set of experts, as in a knowledge domain as time evolves experts continuously keep changing. This paper focuses on the problem of finding experts through referrals in a time evolving co-author social network. Authors and co-authors of research publication for instance are domain experts. In this paper, we propose a solution where the network is expanded incrementally and the information on domain experts is suitably modified. This will avoid periodic global expertise recomputation and would help to effectively retrieve relevant information on domain experts. A novel data structure is also introduced in our study to effectively track the change in expertise of an author with time.

View-Invariant Gait Recognition Through Genetic Template Segmentation

Template-based model-free approach provides by far the most successful solution to the gait recognition problem in literature. Recent work discusses how isolating the head and leg portion of the template increase the performance of a gait recognition system making it robust against covariates like clothing and carrying conditions. However, most involve a manual definition of the boundaries. The method we propose, the genetic template segmentation, employs the genetic algorithm to automate the boundary selection process. This method was tested on the gait energy image (GEI), gait entropy image, and active energy image templates. GEI seems to exhibit the best result when segmented with our approach. Experimental results depict that our approach significantly outperforms the existing implementations of view-invariant gait recognition.

Real time cross layer design using particle swarm optimization

Cross Layer Optimization (CLO) strategies are currently being incorporated in network operating system for efficient utilization of resources to enable effective information management. In wireless adhoc networks real time optimizations need to be performed and hence CLO strategies that have faster response time are required. In this paper we propose a Cross Layer Optimization strategy that uses a variant of the Particle Swarm Optimization (PSO) for real time cross layer design of the network. The variant of the PSO used in this research work uses digital pheromones for improved performance. The proposed PSO-CLO strategy can be used for delay sensitive, bandwidth intensive and loss-tolerant wireless multimedia transmissions that have an ever demanding need for better Quality of Service. Our experimental results show that the proposed PSO-CLO strategy has significantly faster response time in comparison with the classical CLO solutions.