Extending noise robust structured support vector machines to larger vocabulary tasks

This paper describes a structured SVM framework suitable for noise-robust medium/large vocabulary speech recognition. Several theoretical and practical extensions to previous work on small vocabulary tasks are detailed. The joint feature space based on word models is extended to allow context-dependent triphone models to be used. By interpreting the structured SVM as a large margin log-linear model, illustrates that there is an implicit assumption that the prior of the discriminative parameter is a zero mean Gaussian. However, depending on the definition of likelihood feature space, a non-zero prior may be more appropriate. A general Gaussian prior is incorporated into the large margin training criterion in a form that allows the cutting plan algorithm to be directly applied. To further speed up the training process, 1-slack algorithm, caching competing hypothesis and parallelization strategies are also proposed. The performance of structured SVMs is evaluated on noise corrupted medium vocabulary speech recognition task: AURORA 4. © 2011 IEEE.

The retrieval of structured design rationale for the re-use of design knowledge with an integrated representation

Design knowledge can be acquired from various sources and generally requires an integrated representation for its effective and efficient re-use. Though knowledge about products and processes can illustrate the solutions created (know-what) and the courses of actions (know-how) involved in their creation, the reasoning process (know-why) underlying the solutions and actions is still needed for an integrated representation of design knowledge. Design rationale is an effective way of capturing that missing part, since it records the issues addressed, the options considered, and the arguments used when specific design solutions are created and evaluated. Apart from the need for an integrated representation, effective retrieval methods are also of great importance for the re-use of design knowledge, as the knowledge involved in designing complex products can be huge. Developing methods for the retrieval of design rationale is very useful as part of the effective management of design knowledge, for the following reasons. Firstly, design engineers tend to want to consider issues and solutions before looking at solid models or process specifications in detail. Secondly, design rationale is mainly described using text, which often embodies much relevant design knowledge. Last but not least, design rationale is generally captured by identifying elements and their dependencies, i.e. in a structured way which opens the opportunity for going beyond simple keyword-based searching. In this paper, the management of design rationale for the re-use of design knowledge is presented. The retrieval of design rationale records in particular is discussed in detail. As evidenced in the development and evaluation, the methods proposed are useful for the re-use of design knowledge and can be generalised to be used for the retrieval of other kinds of structured design knowledge. © 2012 Elsevier Ltd. All rights reserved.

The retrieval of structured design knowledge

Design rationale is an effective way of capturing knowledge, since it records the issues addressed, the options considered, and the arguments used when specific decisions are made during the design process. Design rationale is generally captured by identifying elements and their dependencies, i.e. in a structured way. Current retrieval methods focus mainly on either the classification of rationale or on keyword-based searches of records. Keyword-based retrieval is reasonably effective as the information in design rationale records is mainly described using text. However, most of the current keyword-based retrieval methods discard the implicit structures of these records, resulting either in poor precision of retrieval or in isolated pieces of information that are difficult to understand. This ongoing research aims to go beyond keyword-based retrieval by developing methods and tools to facilitate the provision of useful design knowledge in new design projects. Our first step is to understand the structured information derived from the relationship between lumps of text held in different nodes in the design rationale captured via a software tool currently used in industry, and study how this information can be utilised to improve retrieval performance. Specifically, methods for utilising various structured information are developed and implemented on a prototype keyword-based retrieval system developed in our earlier work. The implementation and evaluation of these methods shows that the structured information can be utilised in a number of ways, such as filtering the results and providing more complete information. This allows the retrieval system to present results that are easy to understand, and which closely match designers' queries. Like design rationale, other methods for representing design knowledge also in essence involve structured information and thus the methods proposed can be generalised to be adapted and applied for the retrieval of other kinds of design knowledge. Copyright © 2002-2012 The Design Society. All rights reserved.

Multiple classifier boosting and tree-structured classifiers

Visual recognition problems often involve classification of myriads of pixels, across scales, to locate objects of interest in an image or to segment images according to object classes. The requirement for high speed and accuracy makes the problems very challenging and has motivated studies on efficient classification algorithms. A novel multi-classifier boosting algorithm is proposed to tackle the multimodal problems by simultaneously clustering samples and boosting classifiers in Section 2. The method is extended into an online version for object tracking in Section 3. Section 4 presents a tree-structured classifier, called Super tree, to further speed up the classification time of a standard boosting classifier. The proposed methods are demonstrated for object detection, tracking and segmentation tasks. © 2013 Springer-Verlag Berlin Heidelberg.

A structured modelling approach to simulating dynamic behaviours in complex organisations

This article reports a case study application of a systematic approach to modelling complex organisations, centred on simulation modelling (SM). The approach leads to populated instances of complementary model types, in ways that systematically capture, validate and facilitate various uses of organisational understandings, knowledge and data normally distributed amongst multiple knowledge holders. The model-driven approach to decision making enables improved manufacturing responsiveness. Literature on modelling technologies relevant to manufacturing systems organisation design and change is presented, as is literature on production planning and control. This provides a rationale for the development of a new modelling methodology which combines the use of enterprise, causal loop and SM. Subsequently, this article describes how in the case of a specific manufacturing enterprise the combined modelling techniques have informed the choice of alternative production planning and control policies. An example enterprise model of a capacitor manufacturing company is illustrated as derivative causal-loop models that structure and enable the design and use of a general purpose simulation model.

A Structured Model of Video Reproduces Primary Visual Cortical Organisation

The visual system must learn to infer the presence of objects and features in the world from the images it encounters, and as such it must, either implicitly or explicitly, model the way these elements interact to create the image. Do the response properties of cells in the mammalian visual system reflect this constraint? To address this question, we constructed a probabilistic model in which the identity and attributes of simple visual elements were represented explicitly and learnt the parameters of this model from unparsed, natural video sequences. After learning, the behaviour and grouping of variables in the probabilistic model corresponded closely to functional and anatomical properties of simple and complex cells in the primary visual cortex (V1). In particular, feature identity variables were activated in a way that resembled the activity of complex cells, while feature attribute variables responded much like simple cells. Furthermore, the grouping of the attributes within the model closely parallelled the reported anatomical grouping of simple cells in cat V1. Thus, this generative model makes explicit an interpretation of complex and simple cells as elements in the segmentation of a visual scene into basic independent features, along with a parametrisation of their moment-by-moment appearances. We speculate that such a segmentation may form the initial stage of a hierarchical system that progressively separates the identity and appearance of more articulated visual elements, culminating in view-invariant object recognition.