Refine
Document Type
- Conference Proceeding (4)
- Article (1)
- Doctoral Thesis (1)
- Preprint (1)
Language
- English (7)
Keywords
- Machine learning (7) (remove)
Institute
Non-volatile NAND flash memories store information as an electrical charge. Different read reference voltages are applied to read the data. However, the threshold voltage distributions vary due to aging effects like program erase cycling and data retention time. It is necessary to adapt the read reference voltages for different life-cycle conditions to minimize the error probability during readout. In the past, methods based on pilot data or high-resolution threshold voltage histograms were proposed to estimate the changes in voltage distributions. In this work, we propose a machine learning approach with neural networks to estimate the read reference voltages. The proposed method utilizes sparse histogram data for the threshold voltage distributions. For reading the information from triple-level cell (TLC) memories, several read reference voltages are applied in sequence. We consider two histogram resolutions. The simplest histogram consists of the zero-and-one ratios for the hard decision read operation, whereas a higher resolution is obtained by considering the quantization levels for soft-input decoding. This approach does not require pilot data for the voltage adaptation. Furthermore, only a few measurements of extreme points of the threshold voltage distributions are required as training data. Measurements with different conditions verify the proposed approach. The resulting neural networks perform well under other life-cycle conditions.
The evaluation of the effectiveness of different machine learning algorithms on a publicly available database of signals derived from wearable devices is presented with the goal of optimizing human activity recognition and classification. Among the wide number of body signals we choose a couple of signals, namely photoplethysmographic (optically detected subcutaneous blood volume) and tri-axis acceleration signals that are easy to be simultaneously acquired using commercial widespread devices (e.g. smartwatches) as well as custom wearable wireless devices designed for sport, healthcare, or clinical purposes. To this end, two widely used algorithms (decision tree and k-nearest neighbor) were tested, and their performance were compared to two new recent algorithms (particle Bernstein and a Monte Carlo-based regression) both in terms of accuracy and processing time. A data preprocessing phase was also considered to improve the performance of the machine learning procedures, in order to reduce the problem size and a detailed analysis of the compression strategy and results is also presented.
The influence of sleep on human life, including physiological, psychological, and mental aspects, is remarkable. Therefore, it is essential to apply appropriate therapy in the case of sleep disorders. For this, however, the irregularities must first be recognised, preferably conveniently for the person concerned. This dissertation, structured as a composition of research articles, presents the development of mathematically based algorithmic principles for a sleep analysis system. The particular focus is on the classification of sleep stages with a minimal set of physiological parameters. In addition, the aspects of using the sleep analysis system as part of the more complex healthcare systems are explored. Design of hardware for non-obtrusive measurement of relevant physiological parameters and the use of such systems to detect other sleep disorders, such as sleep apnoea, are also referred to. Multinomial logistic regression was selected as the basis for development resulting from the investigations carried out. By following a methodical procedure, the number of physiological parameters necessary for the classification of sleep stages was successively reduced to two: Respiratory and Movement signals. These signals might be measured in a contactless way. A prototype implementation of the developed algorithms was performed to validate the proposed method, and the evaluation of 19324 sleep epochs was carried out. The results, with the achieved accuracy of 73% in the classification of Wake/NREM/REM stages and Cohen's kappa of 0.44, outperform the state of the art and demonstrate the appropriateness of the selected approach. In the future, this method could enable convenient, cost-effective, and accurate sleep analysis, leading to the detection of sleep disorders at an early stage so that therapy can be initiated as soon as possible, thus improving the general population's health status and quality of life.
Comparison of Data-Driven Modeling and Identification Approaches for a Self-Balancing Vehicle
(2023)
This paper gives a systematic comparison of different state–of–the–art modeling approaches and the corresponding parameter identification processes for a self–balancing vehicle. In detail, a nonlinear grey box model, its extension to consider friction effects, a parametric black box model based on regression neural networks, and a hybrid approach are presented. The parameters of the models are identified by solving a nonlinear least squares problem. The training, validation, and test datasets are collected in full–scale experiments using a self–balancing vehicle. The performance of the different models used for ego–motion prediction are compared in full–scale scenarios, as well. The investigated model architectures can be used to improve both, simulation environments and model–based controller design. This paper shows the upsides and downsides arising from using the different modeling approaches. Videos showing the self–balancing vehicle in action are available at: https://tinyurl.com/mvn8j7vf22nd
The main challenge in Bayesian models is to determine the posterior for the model parameters. Already, in models with only one or few parameters, the analytical posterior can only be determined in special settings. In Bayesian neural networks, variational inference is widely used to approximate difficult-to-compute posteriors by variational distributions. Usually, Gaussians are used as variational distributions (Gaussian-VI) which limits the quality of the approximation due to their limited flexibility. Transformation models on the other hand are flexible enough to fit any distribution. Here we present transformation model-based variational inference (TM-VI) and demonstrate that it allows to accurately approximate complex posteriors in models with one parameter and also works in a mean-field fashion for multi-parameter models like neural networks.
Development of an expert system to overpass citizens technological barriers on smart home and living
(2023)
Adopting new technologies can be overwhelming, even for people with experience in the field. For the general public, learning about new implementations, releases, brands, and enhancements can cause them to lose interest. There is a clear need to create point sources and platforms that provide helpful information about the novel and smart technologies, assisting users, technicians, and providers with products and technologies. The purpose of these platforms is twofold, as they can gather and share information on interests common to manufacturers and vendors. This paper presents the ”Finde-Dein-SmartHome” tool. Developed in association with the Smart Home & Living competence center [5] to help users learn about, understand, and purchase available technologies that meet their home automation needs. This tool aims to lower the usability barrier and guide potential customers to clear their doubts about privacy and pricing. Communities can use the information provided by this tool to identify market trends that could eventually lower costs for providers and incentivize access to innovative home technologies and devices supporting long-term care.
There have been substantial research efforts for algorithms to improve continuous and automated assessment of various health-related questions in recent years. This paper addresses the deployment gap between those improving algorithms and their usability in care and mobile health applications. In practice, most algorithms require significant and founded technical knowledge to be deployed at home or support healthcare professionals. Therefore, the digital participation of persons in need of health care professionals lacks a usable interface to use the current technological advances. In this paper, we propose applying algorithms taken from research as web-based microservices following the common approach of a RESTful service to bridge the gap and make algorithms accessible to caregivers and patients without technical knowledge and extended hardware capabilities. We address implementation details, interpretation and realization of guidelines, and privacy concerns using our self-implemented example. Also, we address further usability guidelines and our approach to those.