Learning Forever, Backprop Is Insufficient

Explore a 26-minute video lecture on Continual Learning in Machine Learning, focusing on the limitations of traditional backpropagation and the proposed solution of Continual Backpropagation (CBP). Delve into the concept of plasticity decay and its impact on learning in non-stationary environments, particularly in Reinforcement Learning. Examine the structure and findings of a research paper that introduces CBP as a method to overcome the insufficiencies of normal backpropagation in lifelong learning scenarios. Follow the lecture's progression from an overview and paper introduction to detailed explanations of problems, environments, experiments, and the mechanics of Continual Backprop. Conclude with additional interesting experiments and a summary of the key takeaways, gaining insights into the future of adaptive learning algorithms for continual learning applications.

- Overview
- Paper Intro
- Problems & Environments
- Plasticity Decay Experiments
- Continual Backprop Explained
- Continual Backprop Experiments
- Extra Interesting Experiments
- Summary