Honor College Research Colloquium: James Smith
Speaker: James Smith, AU Honors College alum and AU Graduate Student
Topic: Discrete Cosine Transform Spectral Pooling Layers for Convolutional Neural Networks
Abstract: Pooling operations for convolutional neural networks provide the opportunity to greatly reduce network parameters, leading to faster training time and less data overfitting. Unfortunately, many of the common pooling methods such as max pooling and mean pooling lose information about the data (i.e., they are lossy methods). Recently, a new pooling method called spectral pooling has been utilized to pool data in the spectral domain. By doing so, greater information can be retained with the same network parameter reduction as spatial pooling. Additionally, the convolution step can be combined with spectral pooling to further reduce computational load compared to spatial pooling methods. Spectral pooling is currently implemented in the discrete Fourier domain, but it is found that implementing spectral pooling in the discrete cosine domain concentrates energy in even fewer spectra. An algorithm is presented that implements spectral pooling in the discrete cosine domain and compares results with other pooling methods on a large benchmark dataset. Although Discrete Cosine Transforms Spectral Pooling Layers (DCTSPL) require extra computation compared to normal spectral pooling, the overall time complexity does not change and, furthermore, greater information preservation is obtained, producing networks which converge faster and achieve a lower misclassification error.
Bio: James Smith completed his undergraduate degree at Auburn University in May 2017. He graduated as an University Honors Scholar with a Bachelor of Electrical Engineering and a duel minor in Computer Science and Political Science. While at Auburn, James completed several internships and held an undergraduate research fellowship, conducting research in antenna optimization by genetic algorithms. In addition to his academic pursuits, James held leadership positions in several organizations including Eta Kappa Nu, Spring Up Leadership Programs, and Auburn for Water, a philanthropic organization he co-founded with friends. Currently, James is a graduate student at Auburn University in the Department of Electrical and Computer Engineering, where he works on research under Dr.Bogdan Wilamowski and teaches undergraduate labs. His research area of interest is machine learning, specifically deep learning and optimization algorithms.
Last modified: February 5, 2018