Presentation is loading. Please wait.

Presentation is loading. Please wait.

Model-Free Episodic Control

Similar presentations


Presentation on theme: "Model-Free Episodic Control"— Presentation transcript:

1 Model-Free Episodic Control
Name Ze Liu Data

2 CONTENTS 01 02 03 04 05 INTRODUCTION INSPIRATION AND WHAT TO SOLVE
ALGORITHMS 04 EXPERIMENTAL RESULTS 05 REFERENTIAL VALUE

3 Model-Based or Model-Free Episodic Control INTRODUCTION PART 01

4 MDP S: a set of states A: a set of actions
PART ONE INTRODUCTION MDP S: a set of states A: a set of actions Ps′s,a: the probability that action a in state s will lead to state s' Rs,a: the immediate reward received after transitioning from state s to state s', due to action a γ: the discount factor

5 PART ONE INTRODUCTION Model-Free Model-based

6 PART ONE INTRODUCTION Episodic Control

7 INSPIRATION AND WHAT TO SOLVE PART 02

8 Traditional RL is data inefficient and too slow to train A
PART TWO INSPIRATION AND WHAT TO SOLVE Traditional RL is data inefficient and too slow to train A Traditional RL algorithms take many millions of interactions to attain human-level performance. Humans, on the other hand, can very quickly exploit highly rewarding nuances of an environment upon first discovery. B Research on the brain Study find that the hippocampal system may be used to guide sequential decision-making by co-representing environment states with the returns achieved from the various possible actions

9 ALGORITHMS PART 03

10 PART THREE ALGORITHMS

11 PART THREE ALGORITHMS Writing Look-up

12 PART THREE ALGORITHMS

13 PART THREE ALGORITHMS RP

14 PART THREE ALGORITHMS VAE

15 EXPERIMENTAL RESULTS PART 04

16 PART FOUR EXPERIMENTAL RESULTS

17 REFERENTIAL VALUE PART 05

18 THANK YOU FOR WATCHING


Download ppt "Model-Free Episodic Control"

Similar presentations


Ads by Google