Deep Learning Labs #7 Warsaw

Deep Learning Labs #7 in Warsaw. An initiative by Nextgrid.ai to accelerate understanding & possibilities with deep-learning & reinforcement learning technologies.

Transcript

1. View Slide

2. View Slide

3. Video link: https://www.youtube.com/watch?v=JHX87iv8YJA
   

   View Slide

4. Who we are and what we do?
   

   View Slide

5. View Slide

6. View Slide

7. Deep Learning Labs / Warsaw
   Season #01 Episode #07
   

   View Slide

8. Reinforcement Learning
   (RL) Basics
   By Misha Zanka
   

   View Slide

9. What is RL?
   

   View Slide

10. Policy
    A policy is an agent's strategy.
    https://towardsdatascience.com/self-learning-ai-agents-iv-stochastic-policy-gradients-b53f088fce20
    

    View Slide

11. Stable-baselines
    ● Stable Baselines is a set of improved
    implementations of Reinforcement
    Learning (RL) algorithms based on OpenAI
    Baselines.
    ● Main feature is unified interface for all
    models.
    ● You are free to use other frameworks, but
    this one is the most user-friendly
    

    View Slide

12. Grading
    Today the ranking includes the following tasks :
    ● CartPole - 1 pt.
    ● LunarLander - 4 pt.
    ● Hopper - 6 pt.
    ● HalfCheetah - 12 pt.
    ● BipedalWalker - 24 pt.
    Extra 20% points to each task in case when team will make
    something special, like good presentation with insights or
    non-standard solution of the problem.
    We will maintain the leaderboard on our page.
    

    View Slide

13. Submission
    You will have the link to the google form where you will need to
    send:
    ● If you used stable-baselines .zip with a model and name
    of the algorithm used.
    ● If you used smth else, send trained model and instruction
    how to extract actions from your policy
    

    View Slide

14. Template for submitting results will be available
    

    View Slide