Title: Subspace Based Meta-learning
Abstract:
Meta-learning typically involves two phases. First, one learns a suitable representation from the previously seen tasks. Secondly, this representation is used for learning a new task using only a few samples (i.e., few-shot learning). In this talk I will discuss:
1. Linear meta learning: sample complexity of representation learning with general covariance
2. Linear meta learning: algorithm & analysis for overparameterized few-shot learning
3. Generalization to nonlinear meta-learning
Meta-learning typically involves two phases. First, one learns a suitable representation from the previously seen tasks. Secondly, this representation is used for learning a new task using only a few samples (i.e., few-shot learning). In this talk I will discuss:
1. Linear meta learning: sample complexity of representation learning with general covariance
2. Linear meta learning: algorithm & analysis for overparameterized few-shot learning
3. Generalization to nonlinear meta-learning
Bio:
Yue Sun is a 5th year PhD student from University of Washington, Seattle. He is interested in theoretical understanding of optimization, ML and control. His research works are about:
1. Nonconvex optimization on Riemannian manifolds (UW)
2. Low order linear system identification (UW)
3. Subspace based meta-learning (UW)
4. Nonconvex optimization applied in optimal control (UW)
Yue Sun is a 5th year PhD student from University of Washington, Seattle. He is interested in theoretical understanding of optimization, ML and control. His research works are about:
1. Nonconvex optimization on Riemannian manifolds (UW)
2. Low order linear system identification (UW)
3. Subspace based meta-learning (UW)
4. Nonconvex optimization applied in optimal control (UW)
5. Online optimization for video coding (Google, 2019)
6. Compressive sensing and phase retrieval (Ohio State U, 2015; Nokia Bell Labs, 2021)