CANCELED: BME Seminar: Jeremias Sulam, Ph.D., Johns Hopkins University
Friday,
April 3, 2020
11:00 AM - 12:00 PM
All are welcome, (attendance required for graduate students). Lunch is provided.
Jeremias Sulam, PhD
Assistant Professor, Biomedical Engineering Department
Johns Hopkins University
Neural Networks as Sparse Pursuit Algorithms
Over the last few decades sparsity has become a driving force in the development of new and better algorithms in signal and image processing. In the context of the late deep learning zenith, a recent pivotal work showed that deep neural networks can be interpreted and analyzed as pursuit algorithms seeking for sparse representations of signals belonging to a multilayer synthesis sparse model. In this talk, we will first review these works and further show that this observation is correct but incomplete, in the sense that such a model provides a symbiotic mixture of coupled synthesis and analysis sparse priors. We will make this observation precise and use it to expand on uniqueness guarantees and stability bounds for the pursuit of multilayer sparse representations. We will then explore a convex relaxation of the resulting pursuit and derive efficient optimization algorithms to approximate its solution. Importantly, we will deploy these algorithms in a supervised learning formulation that generalizes feed-forward convolutional neural networks into recurrent ones, improving their performance without increasing the number of parameters the model.
Jeremias Sulam, PhD
Assistant Professor, Biomedical Engineering Department
Johns Hopkins University
Neural Networks as Sparse Pursuit Algorithms
Over the last few decades sparsity has become a driving force in the development of new and better algorithms in signal and image processing. In the context of the late deep learning zenith, a recent pivotal work showed that deep neural networks can be interpreted and analyzed as pursuit algorithms seeking for sparse representations of signals belonging to a multilayer synthesis sparse model. In this talk, we will first review these works and further show that this observation is correct but incomplete, in the sense that such a model provides a symbiotic mixture of coupled synthesis and analysis sparse priors. We will make this observation precise and use it to expand on uniqueness guarantees and stability bounds for the pursuit of multilayer sparse representations. We will then explore a convex relaxation of the resulting pursuit and derive efficient optimization algorithms to approximate its solution. Importantly, we will deploy these algorithms in a supervised learning formulation that generalizes feed-forward convolutional neural networks into recurrent ones, improving their performance without increasing the number of parameters the model.
Status: CANCELED
LOCATION:
← BACK TO EVENTS
- Morningside
- Seminar
- Engineering
- Alumni
- Faculty
- Graduate Students
- Family-friendly
- Postdocs
- Public
- Prospective Students
- Staff
- Students
- Trainees
Date Navigation Widget
Getting to Columbia
Other Calendars
- Alumni Events
- Barnard College
- Columbia Business School
- Columbia College
- Committee on Global Thought
- Heyman Center
- Jewish Theological Seminary
- Miller Theatre
- School of Engineering & Applied Science
- School of Social Work
- Teachers College
Guests With Disabilities
- Columbia University makes every effort to accommodate individuals with disabilities. Please notify us if you need any assistance by contacting the event’s point person. Alternatively, the Office of Disability Services can be reached at 212.854.2388 and [email protected]. Thank you.