The vector quantization (VQ) technique uses a codebook containing block patterns with corresponding index on each of them. In this thesis, we simple TSP (traveling salesperson) scheme in the VQ (vector quantization) index compression. The goal of this method is to improve bit ratio scheme with the same image quality. We apply the TSP (traveling salesperson) scheme to reorder the codewords in the codebook such that the di erence between the indexes in neighboring blocks of the image becomes small. Then, the block indexes in the image are remapped according to the reordered codebook. Thus, the variation between two neighboring block indexes is reduced. Finally, we compress the block indexes of the image with some lossless compression methods. Adding our TSP scheme as a step in VQ (vector quantization) index compression really achieves signi cant reducxtion of bit rates. Our experiment results show that the bpp (bits per pixel) in our method is less than the bpp of those without the TSP scheme.

The image comparison operation ??sessing how well one image matches another ??rms a critical component of many image analysis systems and models of human visual processing. Two norms used commonly for this purpose are L1 and L2, which are specific instances of the Minkowski metric. However, there is often not a principled reason for selecting one norm over the other. One way to address this problem is by examining whether one metric better captures the perceptual notion of image similarity than the other. With this goal, we examined perceptual preferences for images retrieved on the basis of the L1 versus the L2 norm. These images were either small fragments without recognizable content, or larger patterns with recognizable content created via vector quantization. In both conditions the subjects showed a consistent preference for images matched using the L1 metric. These results suggest that, in the domain of natural images of the kind we have used, the L1 metric may better capture human notions of image similarity.

SAS has become an increasingly important public-health problem in recent years. It can adversely affect neurocognitive, cardiovascular, respiratory diseases and can also cause behavior disorder. Since up to 90% of these cases are obstructive sleep apnea (OSA), therefore, the study of how to diagnose, detect and treat OSA is becoming a significant issue, academically and medically. Polysomnography (PSG) can monitor the OSA with relatively fewer invasive techniques. However, PSG-based sleep studies are expansive and time-consuming because they require overnight evaluation in sleep laboratories with dedicated systems and attending personnel. This work develops a flow rate based detection method for apneas. In particular, via signal processing, feature extraction and neural network, this thesis introduces a flow rate based detective system. The goal is to detect OSA with less time and reduced financial costs.

Designing the Nearest Neighbor Classifiers via the VQ Method

This research focuses on the development of a novel adaptive dynamical system approach to vector quantization or clustering based on only ordinary differential equations (ODEs) with potential for a real-time implementation. The ODE-based approach has an advantage in making it possible real-time implementation of the system with either electronic or photonic analog devices. This dynamical system consists of a set of energy functions which create valleys for representing clusters. Each valley represents a cluster of similar input patterns. The proposed system includes a dynamic parameter, called vigilance parameter. This parameter approximately reflects the radius of the generated valleys. Through several examples of different pattern clusters, it is shown that the model can successfully quantize/cluster these types of input patterns. Also, a hardware implementation by photonic and/or electronic analog devices is given In addition, we analyze and study stability of our dynamical system. By discovering the equilibrium points for certain input patterns and analyzing their stability, we have shown the quantizing behavior of the system with respect to its parameters. We also extend our model to include competition mechanism and vigilance dynamics. The competition mechanism causes only one label to be assigned to a group of patterns. The vigilance dynamics adjust vigilance parameter so that the cluster size or the quantizing resolution can be adaptive to the density and distribution of the input patterns. This reduces the burden of re-tuning the vigilance parameter for a given input pattern set and also better represents the input pattern space. The vigilance parameter approximately reflects the radius of the generated valley for each cluster. Making this parameter dynamic allows the bigger cluster to have a bigger radius and as a result a better cluster. Furthermore, an alternative dynamical system to our proposed system is also introduced. This system utilizes sigmoid and competitive functions. Although the results of this system are encouraging, the use of sigmoid function makes analyze and study stability of the system extremely difficult.

Deep neural networks (DNNs), particularly convolutional neural networks (CNNs), have been widely applied in the many fields, such as computer vision, natural language processing, speech recognition and etc. Although DNNs achieve dramatic accuracy improvements in these real-world tasks, they require significant amounts of resources (e.g., memory, energy, storage, bandwidth and computation resources). This limits the application of these networks on resource-constrained systems, such as mobile and edge devices. A large body of literature has been proposed to addresses this problem from the perspective of compressing DNNs while preserving their performance. In this thesis, we focus on compressing deep CNNs based on vector quantization techniques. The first part of this thesis summarizes some basic concepts in machine learning and popular techniques on model compression, including pruning, quantization, low-rank factorization and knowledge distillation approaches. Our main interest is quantization techniques, which compress networks by reducing the precision of parameters. Full-precision weights, activations and even gradients in networks can be quantized to 16-bit floating point numbers, 8-bit integers, or even binary numbers. Despite a possible performance degradation, quantization can greatly reduce the model size while maintaining model accuracy. In the second part of this thesis, we propose a novel vector quantization approach, which we refer to as Vector Quantization with Learned Codebook, or VQLC, for CNNs. Rather than performing scalar quantization, we choose vector quantization that can simultaneously quantize multiple weights at once. Instead of taking a pretraining/clustering approach as in most works, in VQLC, the codebook for quantization are learned together with neural network training from scratch. For the forward pass, the traditional convolutional filters are replaced by the convex combinations of a set of learnable codewords. During inference, the compressed model will be represented by a small-sized codebook and a set of indices, resulting in a significant reduction of model size while preserving the network's performance. Lastly, we validate our approach by quantizing multiple modern CNNs on several popular image classification benchmarks and compare with state-of-the-art quantization techniques. Our experimental results show that VQLC demonstrates at least comparable and often superior performance to the existing schemes. In particular, VQLC demonstrates significant advantages over the existing approaches on wide networks at the high rate of compression.

Real-time rendering of static volumetric data is generally known to be a memory and computationally intensive process. With the advance of graphic hardware, especially GPU, it is now possible to do this using desktop computers. However, with the evolution of real-time CT and MRI technologies, volumetric rendering is an even bigger challenge. The first one is how to reduce the data transmission between the main memory and the graphic memory. The second one is how to efficiently take advantage of the time redundancy which exists in the time-varying volumetric data. Most previous researches either focus on one problem or the other. In this thesis, we implemented a system which efficiently deals with both of the challenges. We proposed an optimized compression scheme that explores the time redundancy as well as space redundancy of time-varying volumetric data. The compressed data is then transmitted to graphic memory and directly rendered by GPU, so the data transfer between main memory and graphic memory is significantly reduced. With our implemented system, we successfully reduce more than half of the time of transferring the whole data directly. We also compare our proposed compression scheme with the one without exploiting time redundancy. The optimized compression scheme shows a reduce compression distortion over time. With usability, portability and extensibility in mind, the implemented system is also quite flexible.

Given the recent abundance of brand choice data from scanner panels market researchers have neglected the measurement and analysis of perceptions. Heterogeneity of perceptions is still a largely unexplored issue in market structure and segmentation studies. Over the last decade various parametric approaches toward modelling segmented perception-preference structures such as combined MDS and Latent Class procedures have been introduced. These methods, however, are not taylored for qualitative data describing consumers' redundant and fuzzy perceptions of brand images. A completely different method is based on topology-sensitive vector quantization (VQ) for consumers-by-brands-by-attributes data. It maps the segment-specific perceptual structures into bubble-pie-bar charts with multiple brand positions demonstrating perceptual distinctiveness or similarity. Though the analysis proceeds without any distributional assumptions it allows for significance testing. The application of exploratory and inferential data processing steps to the same data base is statistically sound and particularly attractive for market structure analysts. A brief outline of the VQ method is followed by a sample study with travel market data which proved to be particularly troublesome for conventional processing tools. (author's abstract) / Series: Report Series SFB "Adaptive Information Systems and Modelling in Economics and Management Science"

Recent years have witnessed increasing interest in using metaheuristics to solve the codebook generation problem (CGP) of vector quantization as well as increasing interest in reducing the computation time of metaheuristics. One of the recently proposed methods aimed at reducing the computation time of metaheuristics is based on the notion of pattern reduction (PR). The problem with PR is in that it may compress and remove patterns that are not supposed to be compressed and removed, thus decreasing the quality of the solution. In this thesis, we proposed a fuzzy version of PR called fuzzy pattern reduction (FPR) to reduce the possibility of compressing and removing patterns that are not supposed to be compressed and removed. To evaluate the performance of the proposed algorithm, we apply it to the following four metaheuristics: generalized Lloyd algorithm, code displacement, genetic k-means algorithm, and particle swarm optimization and use them to solve the CGP. Our experimental results show that the proposed algorithm can not only significantly reduce the computation time but also improve the quality of all the metaheuristics evaluated.

In this paper, we propose a novel approach of vector quantization using a merge-based hierarchical neural network. Vector quantization¡]VQ¡^is known as a very useful technique for lossy data compression. Recently, Neural network¡]NN¡^algorithms have been used for VQ. Vlajic and Card proposed a modified adaptive resonance theory (modified ART2¡^[1] which is a constructing tree structure clustering method. However, modified ART2 has disadvantages of slow construction rate and constructing many redundant levels. Therefore, we propose a more efficient approach for constructing the tree in this paper. Our method establishes only those required levels without losing the fidelity of a compressed image.