Advanced Transfer Learning with Deep Neural Networks
7,5 credits
The course covers the following topics:
Introduction: why deep learning with multiple tasks matters
Transfer learning via fine-tuning and domain adaptation
Multi-task learning
- with fixed neural network architectures
- with task-aware modulation
Meta-learning for few-shot classification and regression
- Black-box meta-learning methods
- Optimization-based meta-learning methods
- Non-parametric methods for few-shot learning
Advanced topics
- The problem of memorization in meta-learning
- Meta-learning without tasks provided: how to construct training tasks automatically
- Life-long learning: how to learn continuously from a sequence of tasks
