Канал: CIS 522 - Deep Learning

RL in Games Outro

RL in Games Outro

RL in Games Intro

RL in Games Intro

Lyle Ungar's Personal Meeting Room

Lyle Ungar's Personal Meeting Room

Guest Lecture 03/15

Guest Lecture 03/15

Invariances

Invariances

Correlation

Correlation

Self Supervision

Self Supervision

Conclusion

Conclusion

Discrete vs Continuous

Discrete vs Continuous

Intro

Intro

Successes and Limits

Successes and Limits

Model-Based RL Intro

Model-Based RL Intro

Model Learning and Usage

Model Learning and Usage

Actor-Critic Training

Actor-Critic Training

Policy Gradient Intro

Policy Gradient Intro

Model Based RL Examples

Model Based RL Examples

REINFORCE Algorithm

REINFORCE Algorithm

Model-Free RL Examples

Model-Free RL Examples

Imitation Learning

Imitation Learning

Direct Policy Search and Actor-Critic

Direct Policy Search and Actor-Critic

Model-Based RL

Model-Based RL

From Tabular Q Learning to Deep Q Learning

From Tabular Q Learning to Deep Q Learning

Experience Replay

Experience Replay

Target Networks

Target Networks

Double DQN

Double DQN

Deep Deterministic Policy Gradients

Deep Deterministic Policy Gradients

Transformers

Transformers

Markov Decision Processes

Markov Decision Processes

Intro to RL

Intro to RL

Solving MDPs

Solving MDPs

Temporal Difference and Q Learning

Temporal Difference and Q Learning

The Bellman Equation

The Bellman Equation

Policy and Value Iteration

Policy and Value Iteration

Wrap Up

Wrap Up

Postitional Encoding

Postitional Encoding

Masking

Masking

Transformer Architecture

Transformer Architecture

Explaining NLP

Explaining NLP

Language Processing

Language Processing

Attention

Attention

Multilingual Embeddings

Multilingual Embeddings

Bert Friends

Bert Friends

Transformers

Transformers

Bias

Bias

Using BERT

Using BERT

Sentence Embeddings

Sentence Embeddings

Distribution Similarity

Distribution Similarity

Intro

Intro

Embeddings

Embeddings

RNN Details

RNN Details

Hate Speech Detection

Hate Speech Detection

Backprop in RNNs

Backprop in RNNs

Language Models

Language Models

Intro

Intro

Basic RNNs

Basic RNNs

BiLSTMs

BiLSTMs

Embeddings

Embeddings

LSTMs

LSTMs

Seq2Seq

Seq2Seq

Tagging

Tagging

Recurrence

Recurrence

Summary

Summary

Time Series

Time Series

Memory

Memory

ELBO

ELBO

VAE Big Picture

VAE Big Picture

Training the Discriminator

Training the Discriminator

Controllable Gan

Controllable Gan

Tutorial 1 Recap

Tutorial 1 Recap

Convolutional GAN

Convolutional GAN

Conditional GANs

Conditional GANs

CycleGAN

CycleGAN

GAN Failure Modes

GAN Failure Modes

Full GAN

Full GAN

W8 Wrap Up

W8 Wrap Up

Intro to GANs

Intro to GANs

The GAN's Generator

The GAN's Generator

Big Picture of ELBO

Big Picture of ELBO

Latent Space

Latent Space

Applications of VAEs

Applications of VAEs

Linear Autoencoders

Linear Autoencoders

VAE Samples

VAE Samples

Week 7 Recap

Week 7 Recap

Datasets

Datasets

Generating with Gaussians

Generating with Gaussians

Convolutional Autoencoders

Convolutional Autoencoders

Density Networks

Density Networks

Continual Learning

Continual Learning

Week 6 Recap

Week 6 Recap

Convnets After AlexNet

Convnets After AlexNet

Large CNNs and Facial Recognition

Large CNNs and Facial Recognition

Week 7 Wrap Up

Week 7 Wrap Up

Making Convnets Scale

Making Convnets Scale

Domain Adaptation

Domain Adaptation

Historical Convnets

Historical Convnets

Convnets Outside of Images

Convnets Outside of Images

Hyperparameter Tuning

Hyperparameter Tuning

Data Augmentation

Data Augmentation

Maxpool and Convolution Gradients

Maxpool and Convolution Gradients

Getting a Feeling for the Dataset

Getting a Feeling for the Dataset

Writing Your Convnet

Writing Your Convnet

Training Loop

Training Loop

Dropout

Dropout

Tutorial 1 Recap

Tutorial 1 Recap

Visualizing Training Curves

Visualizing Training Curves

Convolutions Explained

Convolutions Explained

Electrophysiology Introduction

Electrophysiology Introduction

Max Pooling

Max Pooling

Edge Detection

Edge Detection

Convnet vs Fully Connected Parameters

Convnet vs Fully Connected Parameters

Putting it All Together

Putting it All Together

Good Representations

Good Representations

Week 5 Recap

Week 5 Recap

Regularization Is Shrinkage

Regularization Is Shrinkage

Adversarial Attacks

Adversarial Attacks

Network Pruning and the Lottery Ticket Hypothesis

Network Pruning and the Lottery Ticket Hypothesis

Stochastic Gradient Descent (SGD)

Stochastic Gradient Descent (SGD)

Early Stopping

Early Stopping

Data Augmentation

Data Augmentation

L1 and L2 Regularization

L1 and L2 Regularization

Generalization and Overfitting

Generalization and Overfitting

Dropout

Dropout

Introduction to Week 5

Introduction to Week 5

Network Distillation

Network Distillation

The Difficulty of Training a Neural Network

The Difficulty of Training a Neural Network

Normalization

Normalization

Bias in ML

Bias in ML

Natural Gradients

Natural Gradients

The Importance of the Learning Rate

The Importance of the Learning Rate

Intro

Intro

Minibatches

Minibatches

Adaptive Gradient Descent

Adaptive Gradient Descent

Discussing Week 2 - Linear DL

Discussing Week 2 - Linear DL

Are Integrate-and-Fire Neurons Linear

Are Integrate-and-Fire Neurons Linear

Making Multi-Layer Perceptrons

Making Multi-Layer Perceptrons

Polynomial Features

Polynomial Features

Implementing ReLU

Implementing ReLU

Initialization

Initialization

Cross-Validation

Cross-Validation

Transfer Functions

Transfer Functions

Universal Approximation

Universal Approximation

Linear Learning in Big Multi-Layer Perceptrons

Linear Learning in Big Multi-Layer Perceptrons

Expressivity vs Learnability

Expressivity vs Learnability

Classifying Animal Faces

Classifying Animal Faces

Wrapping Up Week 3

Wrapping Up Week 3

How ReLU Networks Work: Locally Linear Functions

How ReLU Networks Work: Locally Linear Functions

Map Receptive Fields

Map Receptive Fields

Classification and Cross Entropy

Classification and Cross Entropy

Overview and Integrate-and-Fire Neurons

Overview and Integrate-and-Fire Neurons

The XOR Problem

The XOR Problem

Neural Tangent Kernels

Neural Tangent Kernels

Data Augmentation

Data Augmentation

Log(P) as a Cost Function

Log(P) as a Cost Function

Linear Decoding of Neural Data

Linear Decoding of Neural Data

Linear Regression and Matrix Inversion

Linear Regression and Matrix Inversion

How Multiple Dimensions Kick In

How Multiple Dimensions Kick In

Linear Regression with More Dimensions than Data

Linear Regression with More Dimensions than Data

Solving XOR with Linear Neural Networks

Solving XOR with Linear Neural Networks

Condition Numbers

Condition Numbers

Distance Between Gaussian Drawn High Dimensional Vectors

Distance Between Gaussian Drawn High Dimensional Vectors

Pytorch is Great

Pytorch is Great

Xavier vs Simpler Initialization

Xavier vs Simpler Initialization

Linear Regression and Neuroscience

Linear Regression and Neuroscience

Prove 90 Degrees

Prove 90 Degrees

Recap of Week 1

Recap of Week 1

Deep Linear ANNs

Deep Linear ANNs

Angle between Gaussian Drawn High Dimensional Vectors

Angle between Gaussian Drawn High Dimensional Vectors

Deep Learning, the Abstraction and the Implementation

Deep Learning, the Abstraction and the Implementation

Accelerating Growth

Accelerating Growth

Xavier Initialization

Xavier Initialization

Outliers in Neural Activities

Outliers in Neural Activities

Cosine Similarity

Cosine Similarity

Mean Absolute Error

Mean Absolute Error

Course Objectives and Short Term Plans

Course Objectives and Short Term Plans

Wrap up

Wrap up

What is Deep Learning

What is Deep Learning

Automatic Differentiation

Automatic Differentiation

Introduction to the Penn Deep Learning Academy

Introduction to the Penn Deep Learning Academy

Monte Carlo Tree Search (MCTS)

Monte Carlo Tree Search (MCTS)

Alphazero and Scaling

Alphazero and Scaling

Nontrivial Policies

Nontrivial Policies

Exploration Parameter

Exploration Parameter

Potential Policies

Potential Policies

How to Choose an Action

How to Choose an Action

Running MCTS in Parallel

Running MCTS in Parallel

AlphaZero - Putting it All Together

AlphaZero - Putting it All Together

Handmade Policies

Handmade Policies

Training AlphaZero

Training AlphaZero

How Q Is Calculated

How Q Is Calculated

Loading Pretrained Results

Loading Pretrained Results

Training a Policy Network

Training a Policy Network

Training Loop

Training Loop

RL and More Historical Ways of Thinking

RL and More Historical Ways of Thinking

Introduction to Lecture Two: Towards AlphaZero

Introduction to Lecture Two: Towards AlphaZero

Trees that Describe the Future

Trees that Describe the Future

Society and Wrap Up

Society and Wrap Up

From Heuristics to Neural Values

From Heuristics to Neural Values

Making Sense of the Value Function

Making Sense of the Value Function

Data Augmentation and Batching

Data Augmentation and Batching

How To Implement a DL Agent to Play Othello

How To Implement a DL Agent to Play Othello