Dreamer v2: Mastering Atari with Discrete World Models (Machine Learning Research Paper Explained)

Yannic Kilcher · Advanced ·📄 Research Papers Explained ·5y ago
#dreamer #deeprl #reinforcementlearning Model-Based Reinforcement Learning has been lagging behind Model-Free RL on Atari, especially among single-GPU algorithms. This collaboration between Google AI, DeepMind, and the University of Toronto (UofT) pushes world models to the next level. The main contribution is a learned latent state consisting of one discrete part and one stochastic part, whereby the stochastic part is a set of 32 categorical variables, each with 32 possible values. The world model can freely decide how it wants to use these variables to represent the input, but is tasked wit…
Watch on YouTube ↗ (saves to browser)

Chapters (19)

Intro & Overview
4:50 Short Recap of Reinforcement Learning
6:05 Problems with Model-Free Reinforcement Learning
10:40 How World Models Help
12:05 World Model Learner Architecture
16:50 Deterministic & Stochastic Hidden States
18:50 Latent Categorical Variables
22:00 Categorical Variables and Multi-Modality
23:20 Sampling & Stochastic State Prediction
30:55 Actor-Critic Learning in Dream Space
32:05 The Incompleteness of Learned World Models
34:15 How General is this Algorithm?
37:25 World Model Loss Function
39:20 KL Balancing
40:35 Actor-Critic Loss Function
41:45 Straight-Through Estimators for Sampling Backpropagation
46:25 Experimental Results
52:00 Where Does It Fail?
54:25 Conclusion

Playlist

Uploads from Yannic Kilcher · Yannic Kilcher · 0 of 60

← Previous Next →
1 Yannic Kilcher Live Stream
Yannic Kilcher Live Stream
Yannic Kilcher
 2 Imagination-Augmented Agents for Deep Reinforcement Learning
Imagination-Augmented Agents for Deep Reinforcement Learning
Yannic Kilcher
 3 Learning model-based planning from scratch
Learning model-based planning from scratch
Yannic Kilcher
 4 Reinforcement Learning with Unsupervised Auxiliary Tasks
Reinforcement Learning with Unsupervised Auxiliary Tasks
Yannic Kilcher
 5 Attention Is All You Need
Attention Is All You Need
Yannic Kilcher
 6 git for research basics: fundamentals, commits, branches, merging
git for research basics: fundamentals, commits, branches, merging
Yannic Kilcher
 7 Curiosity-driven Exploration by Self-supervised Prediction
Curiosity-driven Exploration by Self-supervised Prediction
Yannic Kilcher
 8 World Models
World Models
Yannic Kilcher
 9 Challenging Common Assumptions in the Unsupervised Learning of Disentangled Representations
Challenging Common Assumptions in the Unsupervised Learning of Disentangled Representations
Yannic Kilcher
 10 Stochastic RNNs without Teacher-Forcing
Stochastic RNNs without Teacher-Forcing
Yannic Kilcher
 11 What’s in a name? The need to nip NIPS
What’s in a name? The need to nip NIPS
Yannic Kilcher
 12 BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding
BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding
Yannic Kilcher
 13 Batch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shift
Batch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shift
Yannic Kilcher
 14 GPT-2: Language Models are Unsupervised Multitask Learners
GPT-2: Language Models are Unsupervised Multitask Learners
Yannic Kilcher
 15 Neural Ordinary Differential Equations
Neural Ordinary Differential Equations
Yannic Kilcher
 16 The Odds are Odd: A Statistical Test for Detecting Adversarial Examples
The Odds are Odd: A Statistical Test for Detecting Adversarial Examples
Yannic Kilcher
 17 Discriminating Systems - Gender, Race, and Power in AI
Discriminating Systems - Gender, Race, and Power in AI
Yannic Kilcher
 18 Blockwise Parallel Decoding for Deep Autoregressive Models
Blockwise Parallel Decoding for Deep Autoregressive Models
Yannic Kilcher
 19 S.H.E. - Search. Human. Equalizer.
S.H.E. - Search. Human. Equalizer.
Yannic Kilcher
 20 Reinforcement Learning, Fast and Slow
Reinforcement Learning, Fast and Slow
Yannic Kilcher
 21 Adversarial Examples Are Not Bugs, They Are Features
Adversarial Examples Are Not Bugs, They Are Features
Yannic Kilcher
 22 I'm at ICML19 :)
I'm at ICML19 :)
Yannic Kilcher
 23 Population-Based Search and Open-Ended Algorithms
Population-Based Search and Open-Ended Algorithms
Yannic Kilcher
 24 Talking to companies at ICML19
Talking to companies at ICML19
Yannic Kilcher
 25 XLNet: Generalized Autoregressive Pretraining for Language Understanding
XLNet: Generalized Autoregressive Pretraining for Language Understanding
Yannic Kilcher
 26 Conversation about Population-Based Methods (Re-upload)
Conversation about Population-Based Methods (Re-upload)
Yannic Kilcher
 27 Reconciling modern machine learning and the bias-variance trade-off
Reconciling modern machine learning and the bias-variance trade-off
Yannic Kilcher
 28 Learning World Graphs to Accelerate Hierarchical Reinforcement Learning
Learning World Graphs to Accelerate Hierarchical Reinforcement Learning
Yannic Kilcher
 29 Manifold Mixup: Better Representations by Interpolating Hidden States
Manifold Mixup: Better Representations by Interpolating Hidden States
Yannic Kilcher
 30 Processing Megapixel Images with Deep Attention-Sampling Models
Processing Megapixel Images with Deep Attention-Sampling Models
Yannic Kilcher
 31 Gauge Equivariant Convolutional Networks and the Icosahedral CNN
Gauge Equivariant Convolutional Networks and the Icosahedral CNN
Yannic Kilcher
 32 Auditing Radicalization Pathways on YouTube
Auditing Radicalization Pathways on YouTube
Yannic Kilcher
 33 RoBERTa: A Robustly Optimized BERT Pretraining Approach
RoBERTa: A Robustly Optimized BERT Pretraining Approach
Yannic Kilcher
 34 Dynamic Routing Between Capsules
Dynamic Routing Between Capsules
Yannic Kilcher
 35 DEEP LEARNING MEME REVIEW - Episode 1
DEEP LEARNING MEME REVIEW - Episode 1
Yannic Kilcher
 36 Accelerating Deep Learning by Focusing on the Biggest Losers
Accelerating Deep Learning by Focusing on the Biggest Losers
Yannic Kilcher
 37 [News] The Siraj Raval Controversy
[News] The Siraj Raval Controversy
Yannic Kilcher
 38 LeDeepChef 👨‍🍳 Deep Reinforcement Learning Agent for Families of Text-Based Games
LeDeepChef 👨‍🍳 Deep Reinforcement Learning Agent for Families of Text-Based Games
Yannic Kilcher
 39 The Visual Task Adaptation Benchmark
The Visual Task Adaptation Benchmark
Yannic Kilcher
 40 IMPALA: Scalable Distributed Deep-RL with Importance Weighted Actor-Learner Architectures
IMPALA: Scalable Distributed Deep-RL with Importance Weighted Actor-Learner Architectures
Yannic Kilcher
 41 AlphaStar: Grandmaster level in StarCraft II using multi-agent reinforcement learning
AlphaStar: Grandmaster level in StarCraft II using multi-agent reinforcement learning
Yannic Kilcher
 42 SinGAN: Learning a Generative Model from a Single Natural Image
SinGAN: Learning a Generative Model from a Single Natural Image
Yannic Kilcher
 43 A neurally plausible model learns successor representations in partially observable environments
A neurally plausible model learns successor representations in partially observable environments
Yannic Kilcher
 44 MuZero: Mastering Atari, Go, Chess and Shogi by Planning with a Learned Model
MuZero: Mastering Atari, Go, Chess and Shogi by Planning with a Learned Model
Yannic Kilcher
 45 NeurIPS 2019
NeurIPS 2019
Yannic Kilcher
 46 Reinforcement Learning Upside Down: Don't Predict Rewards -- Just Map Them to Actions
Reinforcement Learning Upside Down: Don't Predict Rewards -- Just Map Them to Actions
Yannic Kilcher
 47 NeurIPS 19 Poster Session
NeurIPS 19 Poster Session
Yannic Kilcher
 48 Go-Explore: a New Approach for Hard-Exploration Problems
Go-Explore: a New Approach for Hard-Exploration Problems
Yannic Kilcher
 49 Reformer: The Efficient Transformer
Reformer: The Efficient Transformer
Yannic Kilcher
 50 [Interview] Mark Ledwich - Algorithmic Extremism: Examining YouTube's Rabbit Hole of Radicalization
[Interview] Mark Ledwich - Algorithmic Extremism: Examining YouTube's Rabbit Hole of Radicalization
Yannic Kilcher
 51 Turing-NLG, DeepSpeed and the ZeRO optimizer
Turing-NLG, DeepSpeed and the ZeRO optimizer
Yannic Kilcher
 52 Growing Neural Cellular Automata
Growing Neural Cellular Automata
Yannic Kilcher
 53 NeurIPS 2020 Changes to Paper Submission Process
NeurIPS 2020 Changes to Paper Submission Process
Yannic Kilcher
 54 Deep Learning for Symbolic Mathematics
Deep Learning for Symbolic Mathematics
Yannic Kilcher
 55 Online Education - How I Make My Videos
Online Education - How I Make My Videos
Yannic Kilcher
 56 [Rant] coronavirus
[Rant] coronavirus
Yannic Kilcher
 57 Axial Attention & MetNet: A Neural Weather Model for Precipitation Forecasting
Axial Attention & MetNet: A Neural Weather Model for Precipitation Forecasting
Yannic Kilcher
 58 Agent57: Outperforming the Atari Human Benchmark
Agent57: Outperforming the Atari Human Benchmark
Yannic Kilcher
 59 State-of-Art-Reviewing: A Radical Proposal to Improve Scientific Publication
State-of-Art-Reviewing: A Radical Proposal to Improve Scientific Publication
Yannic Kilcher
 60 Can we Contain Covid-19 without Locking-down the Economy?
Can we Contain Covid-19 without Locking-down the Economy?
Yannic Kilcher
The Secret Spy Tech Inside Every Credit Card
Next Up
The Secret Spy Tech Inside Every Credit Card
Veritasium