NirViaje
commented
3 years ago A ROADMAP TO NEVER-ENDING REINFORCEMENT LEARNING
A never-ending learner has the following desiderata:
- It learns behaviors and skills while solving its tasks.
- It invents new subtasks that may later serve as stepping stones.
- It learns hierarchically, i.e. skills learned now can be built upon later.
- It learns without ergodic or resetting assumptions on the underlying (PO)MDP.
- It learns without episode boundaries.
- It learns in a single life without leveraging multiple episodes of experience.
Causal
- CausalVAE
- CausalCuriosity
- Causal World Models
- CausalWorld
- Disentangled Generative Causal Representation Learning
- Causal Effect Inference with Deep Latent-Variable Models
Visualization on RL
t-SNE for RL
Understanding RL Vision
Interactive Visualization for Debugging RL
Grounded language
Language Grounding to Vision and Control Fall 2017, CMU 10-808
Safety RL
DeepMind
openAI safety-gym
- https://openai.com/blog/safety-gym/
- http://github.com/openai/safety-gym
- https://d4mucfpksywv.cloudfront.net/safexp-short.pdf
overfitting in RL
Vectorize supervision signals
- reward and value
- curiosity
- alert
- distributional-Q RL
and daily notation.