Discriminative optimisation of the figure of merit for phonetic spoken term detection

This work proposes to improve spoken term detection (STD) accuracy by optimising the Figure of Merit (FOM). In this article, the index takes the form of phonetic posterior-feature matrix. Accuracy is improved by formulating STD as a discriminative training problem and directly optimising the FOM, through its use as an objective function to train a transformation of the index. The outcome of indexing is then a matrix of enhanced posterior-features that are directly tailored for the STD task. The technique is shown to improve the FOM by up to 13% on held-out data. Additional analysis explores the effect of the technique on phone recognition accuracy, examines the actual values of the learned transform, and demonstrates that using an extended training data set results in further improvement in the FOM.

Natural balancing of flying capacitor voltages in multicell inverter under PD carrier-based PWM

The flying capacitor multicell inverter (FCMI) possesses natural balancing property. With the phase-shifted (PS) carrier-based scheme, natural balancing can be achieved in a straightforward manner. However, to achieve natural balancing with the harmonically optimal phase-disposition (PD) carrierbased scheme, the conventional approaches require (n-1) x (n-1) trapezoidal carrier signals for an n-level inverter, which is (n-1) x (n-2) times more than that in the standard PD scheme. This paper proposes two improved natural balancing strategies for FMI under PD scheme, which use the same (n-1) carrier signals as used in the standard PD scheme. In the first scheme, on-line detection is performed of the band in which the modulation signal is located, corresponding period number of the carrier, and rising or falling half cycle of the carrier waveform to generate the switching signals based on certain rules. In the second strategy, the output voltage level selection is first processed and the switching signals are then generated according to a rule based on preferential cell selection algorithm. These methods are easy to use and can be simply implemented as compared to the other available methods. Simulation and experimental results are presented for a five-level inverter to verify these proposed schemes.

Strategy in developing cell based therapy for large bone

Regeneration of osseous defects by tissue-engineering approach provides a novel means of treatment utilizing cell biology, materials science, and molecular biology. The concept of in vitro cultured osteoblasts having an ability to induce new bone formation has been demonstrated in the critical size defects using small animal models. The bone derived cells can be incorporated into bioengineered scaffolds and synthesize bone matrix, which on implantation can induce new bone formation. In search of optimal cell delivery materials, the extracellular matrix as cell carriers for the repair and regeneration of tissues is receiving increased attention. We have investigated extracellular matrix formed by osteoblasts in vitro as a scaffold for osteoblasts transplantation and found a mineralized matrix, formed by human osteoblasts in vitro, can initiate bone formation by activating endogenous mesenchymal cells. To repair the large bone defects, osteogenic or stem cells need to be prefabricated in a large three dimensional scaffold usually made of synthetic biomaterials, which have inadequate interaction with cells and lead to in vivo foreign body reactions. The interstitial extracellular matrix has been applied to modify biomaterials surface and identified vitronectin, which binds the heparin domain and RGD (Arg-Gly-Asp) sequence can modulate cell spreading, migration and matrix formation on biomaterials. We also synthesized a tri-block copolymer, methoxy-terminated poly(ethylene glycol)(MPEG)-polyL-lactide(PLLA)-polylysine(PLL) for human osteoblasts delivery. We identified osteogenic activity can be regulated by the molecular weight and composition of the triblock copolymers. Due to the sequential loss of lineage differentiation potential during the culture of bone marrow stromal cells that hinderers their potential clinical application, we have developed a clonal culture system and established several stem cell clones with fast growing and multi-differentiation properties. Using proteomics and subtractive immunization, several differential proteins have been identified and verified their potential application in stem cell characterization and tissue regeneration

Robust suspicious behaviour detection for smart surveillance systems

Video surveillance technology, based on Closed Circuit Television (CCTV) cameras, is one of the fastest growing markets in the field of security technologies. However, the existing video surveillance systems are still not at a stage where they can be used for crime prevention. The systems rely heavily on human observers and are therefore limited by factors such as fatigue and monitoring capabilities over long periods of time. To overcome this limitation, it is necessary to have “intelligent” processes which are able to highlight the salient data and filter out normal conditions that do not pose a threat to security. In order to create such intelligent systems, an understanding of human behaviour, specifically, suspicious behaviour is required. One of the challenges in achieving this is that human behaviour can only be understood correctly in the context in which it appears. Although context has been exploited in the general computer vision domain, it has not been widely used in the automatic suspicious behaviour detection domain. So, it is essential that context has to be formulated, stored and used by the system in order to understand human behaviour. Finally, since surveillance systems could be modeled as largescale data stream systems, it is difficult to have a complete knowledge base. In this case, the systems need to not only continuously update their knowledge but also be able to retrieve the extracted information which is related to the given context. To address these issues, a context-based approach for detecting suspicious behaviour is proposed. In this approach, contextual information is exploited in order to make a better detection. The proposed approach utilises a data stream clustering algorithm in order to discover the behaviour classes and their frequency of occurrences from the incoming behaviour instances. Contextual information is then used in addition to the above information to detect suspicious behaviour. The proposed approach is able to detect observed, unobserved and contextual suspicious behaviour. Two case studies using video feeds taken from CAVIAR dataset and Z-block building, Queensland University of Technology are presented in order to test the proposed approach. From these experiments, it is shown that by using information about context, the proposed system is able to make a more accurate detection, especially those behaviours which are only suspicious in some contexts while being normal in the others. Moreover, this information give critical feedback to the system designers to refine the system. Finally, the proposed modified Clustream algorithm enables the system to both continuously update the system’s knowledge and to effectively retrieve the information learned in a given context. The outcomes from this research are: (a) A context-based framework for automatic detecting suspicious behaviour which can be used by an intelligent video surveillance in making decisions; (b) A modified Clustream data stream clustering algorithm which continuously updates the system knowledge and is able to retrieve contextually related information effectively; and (c) An update-describe approach which extends the capability of the existing human local motion features called interest points based features to the data stream environment.

Textures of optical flow for real-time anomaly detection in crowds

Automated visual surveillance of crowds is a rapidly growing area of research. In this paper we focus on motion representation for the purpose of abnormality detection in crowded scenes. We propose a novel visual representation called textures of optical flow. The proposed representation measures the uniformity of a flow field in order to detect anomalous objects such as bicycles, vehicles and skateboarders; and can be combined with spatial information to detect other forms of abnormality. We demonstrate that the proposed approach outperforms state-of-the-art anomaly detection algorithms on a large, publicly-available dataset.

Modulation of depth-dependent properties in tissue-engineered cartilage with a semi-permeable membrane and perfusion : a continuum model of matrix metabolism and transport

The functional properties of cartilaginous tissues are determined predominantly by the content, distribution, and organization of proteoglycan and collagen in the extracellular matrix. Extracellular matrix accumulates in tissue-engineered cartilage constructs by metabolism and transport of matrix molecules, processes that are modulated by physical and chemical factors. Constructs incubated under free-swelling conditions with freely permeable or highly permeable membranes exhibit symmetric surface regions of soft tissue. The variation in tissue properties with depth from the surfaces suggests the hypothesis that the transport processes mediated by the boundary conditions govern the distribution of proteoglycan in such constructs. A continuum model (DiMicco and Sah in Transport Porus Med 50:57-73, 2003) was extended to test the effects of membrane permeability and perfusion on proteoglycan accumulation in tissue-engineered cartilage. The concentrations of soluble, bound, and degraded proteoglycan were analyzed as functions of time, space, and non-dimensional parameters for several experimental configurations. The results of the model suggest that the boundary condition at the membrane surface and the rate of perfusion, described by non-dimensional parameters, are important determinants of the pattern of proteoglycan accumulation. With perfusion, the proteoglycan profile is skewed, and decreases or increases in magnitude depending on the level of flow-based stimulation. Utilization of a semi-permeable membrane with or without unidirectional flow may lead to tissues with depth-increasing proteoglycan content, resembling native articular cartilage.

Intrusion detection framework for encrypted networks

Network-based Intrusion Detection Systems (NIDSs) monitor network traffic for signs of malicious activities that have the potential to disrupt entire network infrastructures and services. NIDS can only operate when the network traffic is available and can be extracted for analysis. However, with the growing use of encrypted networks such as Virtual Private Networks (VPNs) that encrypt and conceal network traffic, a traditional NIDS can no longer access network traffic for analysis. The goal of this research is to address this problem by proposing a detection framework that allows a commercial off-the-shelf NIDS to function normally in a VPN without any modification. One of the features of the proposed framework is that it does not compromise on the confidentiality afforded by the VPN. Our work uses a combination of Shamir’s secret-sharing scheme and randomised network proxies to securely route network traffic to the NIDS for analysis. The detection framework is effective against two general classes of attacks – attacks targeted at the network hosts or attacks targeted at framework itself. We implement the detection framework as a prototype program and evaluate it. Our evaluation shows that the framework does indeed detect these classes of attacks and does not introduce any additional false positives. Despite the increase in network overhead in doing so, the proposed detection framework is able to consistently detect intrusions through encrypted networks.

History-aware, real-time risk detection in business processes (Extended Version)

This paper proposes a novel approach for identifying risks in executable business processes and detecting them at run time. The approach considers risks in all phases of the business process management lifecycle, and is realized via a distributed, sensor-based architecture. At design-time, sensors are defined to specify risk conditions which when fulfilled, are a likely indicator of faults to occur. Both historical and current execution data can be used to compose such conditions. At run-time, each sensor independently notifies a sensor manager when a risk is detected. In turn, the sensor manager interacts with the monitoring component of a process automation suite to prompt the results to the user who may take remedial actions. The proposed architecture has been implemented in the YAWL system and its performance has been evaluated in practice.

Transcutaneous immunization with novel lipid-based adjuvants induces protection against gastric Helicobacter pylori infection

The development of vaccines to combat pathogens that infect across mucosal surfaces has been a major goal of vaccine research. Successful mucosal vaccination requires the co-administration of adjuvants that can overcome the state of immune tolerance normally associated with mucosal application of proteins. In the case of oral immunization, delivery systems are also required to protect vaccine antigens against destruction by gastric pH and digestive enzymes. Furthermore, adjuvants used for mucosal delivery must be free of neurotoxic effects like those induced by the commonly used experimental mucosal adjuvant cholera toxin. Maintenance of the "cold chain" is also essential for the effectiveness of any vaccine and adjuvants/delivery systems that enhance the stability of a vaccine would offer a significant advantage. Needle-free methods of vaccination that induce protective immunity at multiple mucosal surfaces are also desirable for rapid vaccination of large populations. In the present study we show that transcutaneous immunization (TCI) using Lipid C, a novel lipid-based matrix originally developed for oral immunization, containing soluble Helicobacter sonicate significantly reduces the gastric bacterial burden in mice following gastric challenge with live Helicobacter pylori. Protection is associated with the production of splenic gamma interferon and gastric IgA and was achieved without the co-administration of potent and potentially toxic adjuvants, although protection was further enhanced by inclusion of CpG-ODN and cholera toxin in the lipid delivery system.

PCR bias Toward the wild-type k-ras and p53 sequences: Implications for PCR detection of mutations and cancer diagnosis

PCR-based cancer diagnosis requires detection of rare mutations in k- ras, p53 or other genes. The assumption has been that mutant and wild-type sequences amplify with near equal efficiency, so that they are eventually present in proportions representative of the starting material. Work on factor IX suggests that this assumption is invalid for one case of near- sequence identity. To test the generality of this phenomenon and its relevance to cancer diagnosis, primers distant from point mutations in p53 and k-ras were used to amplify wild-type and mutant sequences from these genes. A substantial bias against PCR amplification of mutants was observed for two regions of the p53 gene and one region of k-ras. For k-ras and p53, bias was observed when the wild-type and mutant sequences were amplified separately or when mixed in equal proportions before PCR. Bias was present with proofreading and non-proofreading polymerase. Mutant and wild-type segments of the factor V, cystic fibrosis transmembrane conductance regulator and prothrombin genes were amplified and did not exhibit PCR bias. Therefore, the assumption of equal PCR efficiency for point mutant and wild-type sequences is invalid in several systems. Quantitative or diagnostic PCR will require validation for each locus, and enrichment strategies may be needed to optimize detection of mutants.

Learning the kernel matrix with semidefinite programming

Kernel-based learning algorithms work by embedding the data into a Euclidean space, and then searching for linear relations among the embedded data points. The embedding is performed implicitly, by specifying the inner products between each pair of points in the embedding space. This information is contained in the so-called kernel matrix, a symmetric and positive semidefinite matrix that encodes the relative positions of all points. Specifying this matrix amounts to specifying the geometry of the embedding space and inducing a notion of similarity in the input space - classical model selection problems in machine learning. In this paper we show how the kernel matrix can be learned from data via semidefinite programming (SDP) techniques. When applied to a kernel matrix associated with both training and test data this gives a powerful transductive algorithm -using the labeled part of the data one can learn an embedding also for the unlabeled part. The similarity between test points is inferred from training points and their labels. Importantly, these learning problems are convex, so we obtain a method for learning both the model class and the function without local minima. Furthermore, this approach leads directly to a convex method for learning the 2-norm soft margin parameter in support vector machines, solving an important open problem.

Learning the Kernel Matrix with Semi-Definite Programming

Kernel-based learning algorithms work by embedding the data into a Euclidean space, and then searching for linear relations among the embedded data points. The embedding is performed implicitly, by specifying the inner products between each pair of points in the embedding space. This information is contained in the so-called kernel matrix, a symmetric and positive definite matrix that encodes the relative positions of all points. Specifying this matrix amounts to specifying the geometry of the embedding space and inducing a notion of similarity in the input space -- classical model selection problems in machine learning. In this paper we show how the kernel matrix can be learned from data via semi-definite programming (SDP) techniques. When applied to a kernel matrix associated with both training and test data this gives a powerful transductive algorithm -- using the labelled part of the data one can learn an embedding also for the unlabelled part. The similarity between test points is inferred from training points and their labels. Importantly, these learning problems are convex, so we obtain a method for learning both the model class and the function without local minima. Furthermore, this approach leads directly to a convex method to learn the 2-norm soft margin parameter in support vector machines, solving another important open problem. Finally, the novel approach presented in the paper is supported by positive empirical results.