Matching Items (2)
Filtering by
- All Subjects: Feature Learning
- Creators: Thornton, Trevor
Description
Feature representations for raw data is one of the most important component in a machine learning system. Traditionally, features are \textit{hand crafted} by domain experts which can often be a time consuming process. Furthermore, they do not generalize well to unseen data and novel tasks. Recently, there have been many efforts to generate data-driven representations using clustering and sparse models. This dissertation focuses on building data-driven unsupervised models for analyzing raw data and developing efficient feature representations.
Simultaneous segmentation and feature extraction approaches for silicon-pores sensor data are considered. Aggregating data into a matrix and performing low rank and sparse matrix decompositions with additional smoothness constraints are proposed to solve this problem. Comparison of several variants of the approaches and results for signal de-noising and translocation/trapping event extraction are presented. Algorithms to improve transform-domain features for ion-channel time-series signals based on matrix completion are presented. The improved features achieve better performance in classification tasks and in reducing the false alarm rates when applied to analyte detection.
Developing representations for multimedia is an important and challenging problem with applications ranging from scene recognition, multi-media retrieval and personal life-logging systems to field robot navigation. In this dissertation, we present a new framework for feature extraction for challenging natural environment sounds. Proposed features outperform traditional spectral features on challenging environmental sound datasets. Several algorithms are proposed that perform supervised tasks such as recognition and tag annotation. Ensemble methods are proposed to improve the tag annotation process.
To facilitate the use of large datasets, fast implementations are developed for sparse coding, the key component in our algorithms. Several strategies to speed-up Orthogonal Matching Pursuit algorithm using CUDA kernel on a GPU are proposed. Implementations are also developed for a large scale image retrieval system. Image-based "exact search" and "visually similar search" using the image patch sparse codes are performed. Results demonstrate large speed-up over CPU implementations and good retrieval performance is also achieved.
Simultaneous segmentation and feature extraction approaches for silicon-pores sensor data are considered. Aggregating data into a matrix and performing low rank and sparse matrix decompositions with additional smoothness constraints are proposed to solve this problem. Comparison of several variants of the approaches and results for signal de-noising and translocation/trapping event extraction are presented. Algorithms to improve transform-domain features for ion-channel time-series signals based on matrix completion are presented. The improved features achieve better performance in classification tasks and in reducing the false alarm rates when applied to analyte detection.
Developing representations for multimedia is an important and challenging problem with applications ranging from scene recognition, multi-media retrieval and personal life-logging systems to field robot navigation. In this dissertation, we present a new framework for feature extraction for challenging natural environment sounds. Proposed features outperform traditional spectral features on challenging environmental sound datasets. Several algorithms are proposed that perform supervised tasks such as recognition and tag annotation. Ensemble methods are proposed to improve the tag annotation process.
To facilitate the use of large datasets, fast implementations are developed for sparse coding, the key component in our algorithms. Several strategies to speed-up Orthogonal Matching Pursuit algorithm using CUDA kernel on a GPU are proposed. Implementations are also developed for a large scale image retrieval system. Image-based "exact search" and "visually similar search" using the image patch sparse codes are performed. Results demonstrate large speed-up over CPU implementations and good retrieval performance is also achieved.
ContributorsSattigeri, Prasanna S (Author) / Spanias, Andreas (Thesis advisor) / Thornton, Trevor (Committee member) / Goryll, Michael (Committee member) / Tsakalis, Konstantinos (Committee member) / Arizona State University (Publisher)
Created2014
Description
Every engineer is responsible for completing a capstone project as a culmination of accredited university learning to demonstrate technical knowledge and enhance interpersonal skills, like teamwork, communication, time management, and problem solving. This project, with three or four engineers working together in a group, emphasizes not only the importance of technical skills acquired through laboratory procedures and coursework, but the significance of soft skills as one transitions from a university to a professional workplace; it also enhances the understanding of an engineer's obligation to ethically improve society by harnessing technical knowledge to bring about change. The CC2541 Smart SensorTag is a device manufactured by Texas Instruments that focuses on the use of wireless sensors to create low energy applications, or apps; it is equipped with Bluetooth Smart, which enables it to communicate wirelessly with similar devices like smart phones and computers, assisting greatly in app development. The device contains six built-in sensors, which can be utilized to track and log personal data in real-time; these sensors include a gyroscope, accelerometer, humidifier, thermometer, barometer, and magnetometer. By combining the data obtained through the sensors with the ability to communicate wirelessly, the SensorTag can be used to develop apps in multiple fields, including fitness, recreation, health, safety, and more. Team SensorTag chose to focus on health and safety issues to complete its capstone project, creating applications intended for use by senior citizens who live alone or in assisted care homes. Using the SensorTag's ability to track multiple local variables, the team worked to collect data that verified the accuracy and quality of the sensors through repeated experimental trials. Once the sensors were tested, the team developed applications accessible via smart phones or computers to trigger an alarm and send an alert via vibration, e-mail, or Tweet if the SensorTag detects a fall. The fall detection service utilizes the accelerometer and gyroscope sensors with the hope that such a system will prevent severe injuries among the elderly, allow them to function more independently, and improve their quality of life, which is the obligation of engineers to better through their work.
ContributorsMartin, Katherine Julia (Author) / Thornton, Trevor (Thesis director) / Goryll, Michael (Committee member) / Electrical Engineering Program (Contributor) / School of Film, Dance and Theatre (Contributor) / Barrett, The Honors College (Contributor)
Created2015-12