Главная
Study mode:
on
1
Intro
2
What makes modern machine learning word
3
Predictive models are very powerful!
4
Automated decision making is very powerf
5
First setting: data-driven reinforcement lear
6
Second setting: data-driven model-based optimization
7
Off-policy RL: a quick primer
8
What's the problem?
9
Distribution shift in a nutshell
10
How do prior methods address this?
11
Learning with Q-function lower bounds Algorithm
12
Does the bound hold in practice?
13
How does CQL compare?
14
Predictive modeling and design
15
What's wrong with just doing prediction?
16
The model-based optimization problem
17
Uncertainty and extrapolation
18
What can we do?
19
Model inversion networks (MINS)
20
Putting it all together
21
Experimental results
22
Some takeaways
23
Some concluding remarks
Description:
Explore offline reinforcement learning and model-based optimization in this 34-minute lecture by Sergey Levine from UC Berkeley. Delve into the power of predictive models and automated decision-making, focusing on data-driven reinforcement learning and model-based optimization. Learn about off-policy RL, distribution shift challenges, and Q-function lower bounds. Examine the CQL algorithm and its performance. Investigate predictive modeling and design, addressing issues with simple prediction and exploring model-based optimization problems. Discover uncertainty and extrapolation concepts, and understand model inversion networks (MINS). Analyze experimental results and gain valuable insights into these cutting-edge machine learning techniques.

Offline Reinforcement Learning and Model-Based Optimization

Simons Institute
Add to list
#Computer Science #Machine Learning #Reinforcement Learning #Offline Reinforcement Learning #Predictive Modeling
1 of 23

Intro