Refine
Document Type
- Doctoral Thesis (2)
- Article (1)
Language
- English (3)
Has Fulltext
- yes (3)
Is part of the Bibliography
- no (3)
Keywords
- machine learning (3) (remove)
Institute
In the Internet of Things (IoT), Low-Power Wide-Area Networks (LPWANs) are designed to provide low energy consumption while maintaining a long communications’ range for End Devices (EDs). LoRa is a communication protocol that can cover a wide range with low energy consumption. To evaluate the efficiency of the LoRa Wide-Area Network (LoRaWAN), three criteria can be considered, namely, the Packet Delivery Rate (PDR), Energy Consumption (EC), and coverage area. A set of transmission parameters have to be configured to establish a communication link. These parameters can affect the data rate, noise resistance, receiver sensitivity, and EC. The Adaptive Data Rate (ADR) algorithm is a mechanism to configure the transmission parameters of EDs aiming to improve the PDR. Therefore, we introduce a new algorithm using the Multi-Armed Bandit (MAB) technique, to configure the EDs’ transmission parameters in a centralized manner on the Network Server (NS) side, while improving the EC, too. The performance of the proposed algorithm, the Low-Power Multi-Armed Bandit (LP-MAB), is evaluated through simulation results and is compared with other approaches in different scenarios. The simulation results indicate that the LP-MAB’s EC outperforms other algorithms while maintaining a relatively high PDR in various circumstances.
Computer vision aims at developing algorithms to extract high-level information from images and videos. In the industry, for instance, such algorithms are applied to guide manufacturing robots, to visually monitor plants, or to assist human operators in recognizing specific components. Recent progress in computer vision has been dominated by deep artificial neural network, i.e., machine learning methods simulating the way that information flows in our biological brains, and the way that our neural networks adapt and learn from experience. For these methods to learn how to accurately perform complex visual tasks, large amounts of annotated images are needed. Collecting and labeling such domain-relevant training datasets is, however, a tedious—sometimes impossible—task. Therefore, it has become common practice to leverage pre-available three-dimensional (3D) models instead, to generate synthetic images for the recognition algorithms to be trained on. However, methods optimized over synthetic data usually suffer a significant performance drop when applied to real target images. This is due to the realism gap, i.e., the discrepancies between synthetic and real images (in terms of noise, clutter, etc.). In my work, three main directions were explored to bridge this gap.
First, an innovative end-to-end framework is proposed to render realistic depth images from 3D models, as a growing number of solutions (especially in the industry) are utilizing low-cost depth cameras (e.g., Microsoft Kinect and Intel RealSense) for recognition tasks. Based on a thorough study of these devices and the different types of noise impairing them, the proposed framework simulates their inner mechanisms, comprehensively modeling vital factors such as sensor noise, material reflectance, surface geometry, etc. Able to simulate a wide panel of depth sensors and to quickly generate large datasets, this framework is used to train algorithms for various recognition tasks, consistently and significantly enhancing their performance compared to other state-of-the-art simulation tools.
In some cases, however, relevant 2D or 3D object representations to generate synthetic samples are not available. Considering this different case of data scarcity, a solution is then proposed to incrementally build a representation of visual scenes from partial observations. Provided observations are localized from one to another based on their content and registered in a global memory with spatial properties. Simultaneously, this memory can be queried to render novel views of the scene. Furthermore, unobserved regions can be hallucinated in memory, in consistence with previous observations, hallucinations, and global priors. The efficacy of the proposed mnemonic and generative system, trainable end-to-end, is demonstrated on various 2D and 3D use-cases.
Finally, an advanced convolutional neural network pipeline is introduced, tackling the realism gap from a novel angle. While most methods addressing this problem focus on bringing synthetic samples—or the knowledge acquired from them—closer to the real target domain, the proposed solution performs the opposite process, mapping unseen target images into controlled synthetic domains. The pre-processed samples can then be handed to downstream recognition methods, themselves purely trained on similar synthetic data, to greatly improve their accuracy.
For each approach, a variety of qualitative and quantitative studies are detailed, providing successful comparisons to state-of-the-art methods. By proposing solutions to bridge the realism gap from either side, as well as a pipeline to improve the acquisition and generation of new visual content, this thesis provides a unique perspective on the challenges of data scarcity when building robust recognition systems.
Our internal clock, the circadian clock, determines at which time we have our best cognitive abilities, are physically strongest, and when we are tired. Circadian clock phase is influenced primarily through exposure to light. A direct pathway from the eyes to the suprachiasmatic nucleus, where the circadian clock resides, is used to synchronise the circadian clock to external light-dark cycles.
In modern society, with the ability to work anywhere at anytime and a full social agenda, many struggle to keep internal and external clocks synchronised. Living against our circadian clock makes us less efficient and poses serious health impact, especially when exercised over a long period of time, e.g. in shift workers. Assessing circadian clock phase is a cumbersome and uncomfortable task. A common method, dim light melatonin onset testing, requires a series of eight saliva samples taken in hourly intervals while the subject stays in dim light condition from 5 hours before until 2 hours past their habitual bedtime.
At the same time, sensor-rich smartphones have become widely available and wearable computing is on the rise. The hypothesis of this thesis is that smartphones and wearables can be used to record sensor data to monitor human circadian rhythms in free-living. To test this hypothesis, we conducted research on specialised wearable hardware and smartphones to record relevant data, and developed algorithms to monitor circadian clock phase in free-living. We first introduce our smart eyeglasses concept, which can be personalised to the wearers head and 3D-printed. Furthermore, hardware was integrated into the eyewear to recognise typical activities of daily living (ADLs). A light sensor integrated into the eyeglasses bridge was used to detect screen use. In addition to wearables, we also investigate if sleep-wake patterns can be revealed from smartphone context information. We introduce novel methods to detect sleep opportunity, which incorporate expert knowledge to filter and fuse classifier outputs. Furthermore, we estimate light exposure from smartphone sensor and weather in- formation. We applied the Kronauer model to compare the phase shift resulting from head light measurements, wrist measurements, and smartphone estimations.
We found it was possible to monitor circadian phase shift from light estimation based on smartphone sensor and weather information with a weekly error of 32±17min, which outperformed wrist measurements in 11 out of 12 participants. Sleep could be detected from smartphone use with an onset error of 40±48 min and wake error of 42±57 min. Screen use could be detected smart eyeglasses with 0.9 ROC AUC for ambient light intensities below 200lux. Nine clusters of ADLs were distinguished using Gaussian mixture models with an average accuracy of 77%. In conclusion, a combination of the proposed smartphones and smart eyeglasses applications could support users in synchronising their circadian clock to the external clocks, thus living a healthier lifestyle.