Robust Image Classification and Clustering via Distribution learning

• Author / Creator

  Tan, Zijie

• Distribution learning has long been a key area of research in computer vision.
  However, the potential of combining distribution learning with deep learning
  remains underexplored. To bridge this gap, this thesis discusses two proposed
  methods. The first, Differentiable Arithmetic Distribution Module (DADM),
  introduces differentiability to the construction of histograms, enabling deep
  learning models to leverage distributional information more effectively. By
  employing Kernel Density Estimate (KDE) within a deep learning framework,
  DADM captures distribution information that is nearly invariant to affine
  transformations, significantly enhancing the robustness of image classification
  models against such variations. The second method, Deep Clustering via Distribution Learning (DCDL), extends the application of distribution learning
  to clustering tasks, particularly in high-dimensional data spaces. DCDL integrates distribution learning into deep clustering frameworks through the introduction of Monte-Carlo Marginalization for Clustering (MCMarg-C), an algorithm that optimizes cluster formation by directly learning the underlying data
  distribution. This method improves clustering performance by maintaining
  data structure through dimensionality reduction and manifold approximation.
  Overall, this thesis aims to leverage the integration of distribution learning and
  deep learning to address the limitations of traditional deep learning methods,
  thereby developing more robust and scalable models for computer vision tasks
  such as image classification and clustering.

• Subjects / Keywords

  • disttribution learning
  • deep learning
  • machine learning
  • image classification
  • clustering

• Graduation date

  Fall 2024

• Type of Item

  Thesis

• Degree

  Master of Science

• DOI

  https://doi.org/10.7939/r3-7s7j-rt08

• License

  This thesis is made available by the University of Alberta Library with permission of the copyright owner solely for non-commercial purposes. This thesis, or any portion thereof, may not otherwise be copied or reproduced without the written consent of the copyright owner, except to the extent permitted by Canadian copyright law.