Georgia Institute of Technology
Multi-Layer Dictionary Learning Using Low-Rank Updates
Abstract
dc:description.abstractThis dissertation presents a novel convolutional dictionary learning algorithm for signals with a large number of channels. This algorithm uses low-rank updates for the dictionary, so that a matrix decomposition necessary for pursuit can be updated efficiently. In later chapters, this algorithm is applied to multi-layer dictionary models with multi-channel dictionaries and single-channel coefficients, where the number of filters in one layer is the number of channels in the subsequent dictionary layer. This architecture is demonstrated on the task of JPEG compression artifact removal.
Degree
thesis:*- Level thesis:degree_level
- Doctoral
- Department dc:contributor.department
- Electrical and Computer Engineering
- Grantor dc:publisher
- Georgia Institute of Technology
- Year dc:date.issued
- 2022
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Richert, Lee
- Advisor dc:contributor.advisor
-
- Anderson, David V.
- Committee members dc:contributor.committeemember
-
- Davenport, Mark A.
- Romberg, Justin
- Abernethy, Jacob
- Pananjady, Ashwin
Subjects
dc:subject × 25- Dictionary learning
- Convolutional
- Low-rank
- Sparsity
- JPEG compression
- Multi-layer
- Cholesky updates
- Alternating Directions Method of Multipliers
- ADMM
- Sparse coding
- Pursuit
- LASSO
- Complex gradients
- Multi-channel
- Online learning
- Tensorflow
- Keras
- Keras Callbacks
- Inverse problem
- Backpropagation
- Convex optimization
- Constrained optimization
- Augmented Lagrangian
- Image restoration
- Convolutional neural network
Rights
- Language dc:language.iso
- en_US
Identifiers
dc:identifier.*- Handle dc:identifier.uri
- http://hdl.handle.net/1853/66503