Bayesian Adaptive Inference and Adaptive Training

Large-vocabulary speech recognition systems are often built using found data, such as broadcast news. In contrast to carefully collected data, found data normally contains multiple acoustic conditions, such as speaker or environmental noise. Adaptive training is a powerful approach to build systems on such data. Here, transforms are used to represent the different acoustic conditions, and then a canonical model is trained given this set of transforms. This paper describes a Bayesian framework for adaptive training and inference. This framework addresses some limitations of standard maximum-likelihood approaches. In contrast to the standard approach, the adaptively trained system can be directly used in unsupervised inference, rather than having to rely on initial hypotheses being present. In addition, for limited adaptation data, robust recognition performance can be obtained. The limited data problem often occurs in testing as there is no control over the amount of the adaptation data available. In contrast, for adaptive training, it is possible to control the system complexity to reflect the available data. Thus, the standard point estimates may be used. As the integral associated with Bayesian adaptive inference is intractable, various marginalization approximations are described, including a variational Bayes approximation. Both batch and incremental modes of adaptive inference are discussed. These approaches are applied to adaptive training of maximum-likelihood linear regression and evaluated on a large-vocabulary speech recognition task. Bayesian adaptive inference is shown to significantly outperform standard approaches.     

Adaptive programming

An adaptive program is one that changes its behavior base on the current state of its environment. This notion of adaptivity is formalized, and a logic for reasoning about adaptive programs is presented. The logic includes several composition operators that can be used to define an adaptive program in terms of given constituent programs; programs resulting from these compositions retain the adaptive properties of their constituent programs. The authors begin by discussing adaptive sequential programs, then extend the discussion to adaptive distributed programs. The relationship between adaptivity and self-stabilization is discussed. A case study for constructing an adaptive distributed program where a token is circulated in a ring of processes is presented     

Adaptive generalisation

Adaptive generalisation is the ability to use prior knowledge in the performance of novel tasks. Thus, if we are to model intelligent behaviour with neural nets, they must be able to generalise across task domains. Our objective is to elucidate the aetiology of transfer of information between connectionist nets. First, a method is described that provides a standardised score for the quantification of how much task structure a net has extracted, and to what degree knowledge has been transferred between tasks. This method is then applied in three simulation studies to examine Input-to-Hidden (IH) and Hidden-to-Output (HO) decision hyperplanes as determinants of transfer effects. In the first study, positive transfer is demonstrated between functions that require the vertices of their spaces to be divided similarly, and negative transfer between functions that require decision regions of different shapes. In the other two studies, input and output similarity are varied independently in a series of paired associate learning tasks. Further explanation of transfer effects is provided through the use of a new technique that permits better visualisation of the entire computational space by showing both the relative position of inputs in Hidden Unit space, and the HO decision regions implemented by the set of weights.This research was supported by an award from the Economic and Social Research Council, Grant No R000233441. An earlier version of this paper appears in the Proceedings of the Second Irish Networks Conference, Belfast 1992.     

Adaptive Hypermedia

Adaptive hypermedia is a relatively new direction of research on the crossroads of hypermedia and user modeling. Adaptive hypermedia systems build a model of the goals, preferences and knowledge of each individual user, and use this model throughout the interaction with the user, in order to adapt to the needs of that user. The goal of this paper is to present the state of the art in adaptive hypermedia at the eve of the year 2000, and to highlight some prospects for the future. This paper attempts to serve both the newcomers and the experts in the area of adaptive hypermedia by building on an earlier comprehensive review (Brusilovsky, 1996; Brusilovsky, 1998).     

Adaptive graphics

This article presents the idea of a unifying framework that allows visual representations of information to be customized and mixed together into new ones. The net result is a fine-grained approach to representing data, better suited to accessing and rendering it over networks. Although the focus is on geometric models and 3D shape representations, many issues discussed are relevant to network-based visualization in general.     

Adaptive Actions

Adaptive Actions initiated in London in 2007 by Jean-Fran?ois Prost explores alterations in the workplace, the home, and public spaces in general. Identifying the variety of these personal and found alterations in the city as different forms of adaptation creates a vocabulary for the expression of the collective imagination, through the existing urban structures therein. These ‘actions’ modify and activate the intended use of architecture and enhance the character of urban environments. They create positive tensions that test the limits of tolerated appropriation. Can these simple actions, images, and ideas, such as the hybridization of conventional and unusual urban realities, infiltrate our collective imagination to promote feelings of identity and a sense of cultural belonging? Adaptive Actions points to how urban phenomena impact on residents’ perception of the environment and their relation to it. By offering a space to share experiences, ideas, forms of actions, and specific accomplishments, Adaptive Actions creates an inventory of alterations rarely visible to the public. Printed documents and organized events are being planned to increase visibility of the selected actions to the public eye and build affiliations and communal thinking.     

Adaptive marching cubes

The marching cubes algorithm (MC) is a powerful technique for surface rendering that can produce very high-quality images. However, it is not suitable for interactive manipulation of the 3D surfaces constructed from high-resolution volume datasets in terms of both space and time. In this paper, we present an adaptive version of MC called adaptive marching cubes (AMC). It significantly reduces the number of triangles representing the surface by adapting the size of the triangles to the shape of the surface. This improves the performance of the manipulation of the 3D surfaces. A typical example with the volume dataset of size 256×256×113 shows that the number of triangles is reduced by 55%. The quality of images produced by AMC is similar to that of MC. One of the fundamental problems encountered with adaptive algorithms is thecrack problem. Cracks may be created between two neighboring cubes processed with different levels of subdivision. We solve the crack problem by patching the cracks using polygons of the smae shape as those of the cracks. We propose a simple, but complete, method by first abstracting 22 basic configurations of arbitrarily sized cracks and then reducing the handling of these configurations to a simple rule. It requires onlyO(n ²) working memory for an×n×n volume data set.     

Adaptive sampling technique

An adaptive sampling technique utilizing the first three terms of a Taylor series expansion about thenth sampling instant for the error signal in an error-sampled feedback control system is presented. The necessary rules for determining the sign convention used in applying the technique are given along with an example illustrating the method. The procedure basically depends on the determination of a preselected constant difference between successive sampling instants.     

Adaptive geometry image

We present a novel post-processing utility called adaptive geometry image (AGIM) for global parameterization techniques that can embed a 3D surface onto a rectangular1 domain. This utility first converts a single rectangular parameterization into many different tessellations of square geometry images(GIMs) and then efficiently packs these GIMs into an image called AGIM. Therefore, undersampled regions of the input parameterization can be up-sampled accordingly until the local reconstruction error bound is met. The connectivity of AGIM can be quickly computed and dynamically changed at rendering time. AGIM does not have T-vertices, and therefore no crack is generated between two neighboring GIMs at different tessellations. Experimental results show that AGIM can achieve significant PSNR gain over the input parameterization, AGIM retains the advantages of the original GIM and reduces the reconstruction error present in the original GIM technique. The AGIM is also for global parameterization techniques based on quadrilateral complexes. Using the approximate sampling rates, the PolyCube-based quadrilateral complexes with AGIM can outperform state-of-the-art multichart GIM technique in terms of PSNR.     

Multiplierless Adaptive Filtering

Bose, T., Venkatachalam, A., and Thamvichai, R., Multiplierless Adaptive Filtering, Digital Signal Processing12 (2002) 107–118When digital filters are designed with power-of-2 coefficients, the multiplications can be implemented by simple shifting operations. For VLSI implementations, multiplierless filters are faster and more compact than filters with multipliers. In this paper, an algorithm for finding and updating the power-of-2 coefficients of an adaptive filter is designed. The new method uses the well-known Genetic Algorithm (GA) for this purpose. The GA is used in a unique way in order to reduce computations. Small blocks of data are used for the GA and only one new generation is produced per sample of data. This, coupled with the fact that the coefficients are power-of-2, yields a computational complexity of O(N) additions and no multiplications. The algorithm is investigated for applications in adaptive linear prediction and system identification. The results are very promising and illustrate the performance of the new algorithm.     

Adaptive sweeping techniques

This paper presents an adaptive approach to sweeping one-to-one and many-to-one geometry. The automatic decomposition of many-to-one geometry into one-to-one “blocks” and the selection of an appropriate node projection scheme are vital steps in the efficient generation of high-quality swept meshes. This paper identifies two node projection schemes which are used in tandem to robustly sweep each block of a one-to-one geometry. Methods are also presented for the characterization of one-to-one geometry and the automatic assignment of the most appropriate node projection scheme. These capabilities allow the sweeper to adapt to the requirements of the sweep block being processed. The identification of the two node projection schemes was made after an extensive analysis of existing schemes was completed. One of the node projection schemes implemented in this work, BoundaryError, was selected from traditional node placement algorithms. The second node projection scheme, SmartAffine, is an extension of simple affine transformations and is capable of efficiently sweeping geometry with source and/or target curvature while approximating the speed of a simple transform. These two schemes, when used in this adaptive setting, optimize mesh quality and the speed that swept meshes can be generated while minimizing required user interaction.     

Adaptive hypergraph superpixels

Superpixel segmentation, which amounts to partitioning an image into a number of superpixels each of which is a set of pixels sharing common visual meanings, requires specific needs for different computer vision tasks. Graph based methods, as a kind of popular superpixel segmentation method, regard an image as a weighted graph whose nodes correspond to pixels of the image, and partition all pixels into superpixels according to the similarity between pixels over various feature spaces. Despite their improvement of the performance of segmentation, these methods ignore high-order relationship between them incurred from either locally neighboring pixels or structured layout of the image. Moreover, they measure the similarity of pairwise pixels using Gaussian kernel where a robust radius parameter is difficult to find for pixels which exhibit multiple features (e.g., texture, color, brightness). In this paper, we propose an adaptive hypergraph superpixel segmentation (AHS) of intensity images for solving both issues. AHS constructs a hypergraph by building the hyperedges with an adaptive neighborhood scheme, which explores an intrinsic relationship of pixels. Afterwards, AHS encodes the relationship between pairwise pixels using characteristics of current two pixels as well as their neighboring pixels defined by hyperedges. Essentially, AHS models the relationship of pairwise pixels in a high-order group fashion while graph based methods evaluate it in a one-vs-one fashion. Experiments on four datasets demonstrate that AHS achieves higher or comparable performance compared with state-of-the-art methods.     

Adaptive conductance control

Neuromodulation is central to the adaptation and robustness of animal nervous systems. This paper explores the classical paradigm of indirect adaptive control to design neuromodulatory controllers in conductance-based neuronal models. The adaptive control of maximal conductance parameters is shown to provide a methodology aligned with the central concepts of neuromodulation in physiology and of impedance control in robotics.     

自适应颜色迁移

该文提出一种基于统计学的自适应图像颜色迁移技术,把一幅图像的颜色迁移到另一幅图像中．对于给定两幅图像,首先把计算机表示的RGB颜色空间转换为LMS空间,并通过线性变换把LMS基变换到一组正交基来消除空间内各个分量之间的强相关性,从而把RGB颜色空间转换到一个正交空间．然后利用统计学方法进行局部和整体相结合的自适应采样分析,提取一幅图像的颜色信息和另一幅图像的形状信息,合成一幅崭新的图像,实现颜色图像到形状图像的颜色迁移过程．进一步,可以把这种迁移合成技术用来扩展现有的图像效果处理软件．     

Adaptive Fraud Detection

One method for detecting fraud is to check for suspicious changes in user behavior. This paper describes the automatic design of user profiling methods for the purpose of fraud detection, using a series of data mining techniques. Specifically, we use a rule-learning program to uncover indicators of fraudulent behavior from a large database of customer transactions. Then the indicators are used to create a set of monitors, which profile legitimate customer behavior and indicate anomalies. Finally, the outputs of the monitors are used as features in a system that learns to combine evidence to generate high-confidence alarms. The system has been applied to the problem of detecting cellular cloning fraud based on a database of call records. Experiments indicate that this automatic approach performs better than hand-crafted methods for detecting fraud. Furthermore, this approach can adapt to the changing conditions typical of fraud detection environments.     

Adaptive topology optimization

Topology optimization of continuum structures is often reduced to a material distribution problem. Up to now this optimization problem has been solved following a rigid scheme. A design space is parametrized by design patches, which are fixed during the optimization process and are identical to the finite element discretization. The structural layout is determined, whether or not there is material in the design patches. Since many design patches are necessary to describe approximately the structural layout, this procedure leads to a large number of optimization variables. Furthermore, due to a lack of clearness and smoothness, the results obtained can often only be used as a conceptual design idea.To overcome these shortcomings adaptive techniques, which decrease the number of optimization variables and generate smooth results, are introduced. First, the use of pure mesh refinement in topology optimization is discussed. Since this technique still leads to unsatisfactory results, a new method is proposed that adapts the effective design space of each design cycle to the present material distribution. Since the effective design space is approximated by cubic or Bézier splines, this procedure does not only decrease the number of design variables and lead to smooth results, but can be directly joined to conventional shape optimization. With examples for maximum stiffness problems of elastic structures the quality of the proposed techniques is demonstrated.