ConceptAI & ML
Transfer learning
Using representations learned from one task to perform more learning on another task.
Know first
Named as a prerequisite by the lectures that define this. The number is how many say so. A grey one has no page yet.
Neural networks4Deep learning2Machine learning2Machine learning models2Reinforcement learning2Control systems1Convolutional neural networks1Distribution shift1
Opens up
Concepts that name this one as a prerequisite.
Lecturers
6Each in their own terms, from the easiest level.
Sara Beery
Fei-Fei Li, Justin Johnson & Serena Yeung
A technique where pre-trained big networks on some dataset are fine-tuned for a new, specific problem.Sergey Levine
Using experience from one set of tasks for faster learning and better performance on a new task.Phillip Isola
Using representations learned from previous tasks to accelerate and improve future learning.Barbara Kwiecinska
Transferring parameters from a well-trained model with large amounts of data to a model for a different task, typically by adjusting the final layers.Steve Brunton
You can try a controller in one environment, such as 2D or low flow velocities, and see how it works in a more complex setting like 3D or higher speeds.
Lectures that assume it
4Connections
23The lectures that define it or assume it, and the concepts next to it.
Concepts9
Show 3 moreShow fewer
Lectures14
- Training Neural Networks II
- Representation Learning: Similarity-Based
- Deep Learning Software
- Transfer Learning: Models
- Sequential Decision-Making: Policy Gradients vs. Value-Based Architectures
- HydroGym: A Reinforcement Learning Platform for Fluid Dynamics
Show 8 moreShow fewer
- Representation Learning: Reconstruction-Based
- Transfer Learning & Meta-Learning
- Transfer Learning: Data
- Current & Future Application of AI in Epilepsy by Dr. Barbara Kwiecinska, PGY4 Neurology Resident
- A Hacker's Guide to Language Models
- Transformers in Vision: Tackling problems in Computer Vision
- Emergent Behaviors in Foundational Models
- Transformers that Transform Well Enough to Support Near-Shallow Architectures