2020 fast.ai study 활동기록 저장소 이 저장소는 2020 fast.ai study의 활동기록을 저장하는 저장소입니다. fast.ai 스터디 발표자 시트 List Part Page 1 Deep Learning Is for Everyone 3 1 How to Learn Deep Learning 12 2 The Software: PyTorch, fastai, and Jupyter (And Why It Doesn’t Matter) 12 2 Running Your First Notebook 20 3 What Is Machine Learning? 20 3 Limitations Inherent to ML 26 4 How Our Image Recognizer Works 26 4 What Our Image Recognizer Learned 36 5 Image Recognizers Can Tackle Non-Image Tasks 36 5 Deep Learning Is Not Just for Image Classification 48 6 Validation Sets and Test Sets 48 6 A Choose Your Own Adventure Moment 54 7 The Preactice of Deep Learning 57 7 The Practice of Deep Learning : The Drivetrain Approach 65 8 Gathering Data 65 8 Gathering Data 70 9 From Data to DataLoaders 70 9 Training Your Model, and Using It to Clean Your Data 78 10 Tuning Your Model into an Online Application 78 10 Tuning Your Model into an Online Application:Deploying Your App 86 11 How to Avoid Disaster 86 11 Questionaire : Further Research 92 12 Key Example for Data Ethnics 93 12 Key Example for Data Ethnics 99 13 Integrating Machine Learning with Product Design 99 13 Topics in Data Ethnics : Feedback Loops 105 14 Topics in Data Ethnics : Bias 105 14 Topics in Data Ethnics : Bias 116 15 Topics in Data Ethnics : Disinformation 116 15 Identifying and Addressing Ethical Issues:Fairness, Accountability, and Transparency 123 16 Role of Policy 123 16 Deep Learning in Practice : That's a Wrap! 128 17 Pixels: The Foundations of Computer Vision 133 17 First Try: Pixel Similarity:NumPy Arrays and PyTorch Tensors 145 18 Computing Metrics Using Broadcastingn 145 18 Stochastic Gradient Descent:Calculating Gradients 156 19 Stochastic Gradient Descent:Stepping with a Learning Rate 157 19 Stochastic Gradient Descent:Summarizing Gradient Descent 163 20 The MNIST Loss Function:Sigmoid 163 20 The MNIST Loss Function:SGD and Mini-Batches 171 21 Putting It All Together:Creating an Optimizer 171 21 Questionnaire:Further Research 184 22 From Dogs and Cats to Pet Breeds 185 22 Presizing:Checking and Debugging a DataBlocks 194 23 Cross-Entropy Loss 194 23 Cross-Entropy Loss:Taking the log 203 24 Model Interpretation 203 24 Improving Our Model:Unfreezing and Transfer Learning 210 25 Improving Our Model:Discriminative Learning Rates 210 25 Questionnaire:Further Research 217 26 Other Computer Vision Problems: Multi-Label Classification 219 26 Other Computer Vision Problems: Binary Cross Entropy 231 27 Other Computer Vision Problems: Regression 231 27 Other Computer Vision Problems: Questionnaire 239 28 Training a State-of-the-Art Model: Imagenette 239 28 Training a State-of-the-Art Model: Questionnaire 252 29 Collaborative Filtering: A First Look at the Data 253 29 Collaborative Filtering: Creating Our Own Embedding Module 266 30 Collaborative Filtering: Interpreting Embeddings and Biases 267 30 Collaborative Filtering: Questionnaire 276 31 Categorical Embeddings 277 31 The Dataset: Look at the Data 286 32 Decision Trees 287 32 Decision Trees: Categorical Variables 297 33 Random Forests 298 33 Model Interpretation: Removing Redundant Features 306 34 Model Interpretation: Partial Dependence 308 34 Extrapolation and Neural Networks: Finding Out-of-Domain Data 318 35 Extrapolation and Neural Networks: Using a Neural Network 318 35 Conclusion 327 36 Text Preprocessing 329 36 Putting Our Texts into Batches for a Language Model 342 37 Training a Text Classifier 342 37 Fine-Tuning the Classifier 350 38 Going Deeper into fastai's Layered API 350 38 Pipeline 359 39 TfmdLists and Datasets: Transformed Collections 359 39 Understanding fastai's Applications: Wrap Up 373 40 The Data, Our First Language Model from Scratch 373 40 Our First Recurrent Neural Network 381 41 Improving the RNN 381 41 Creating More Signal 386 42 Multilayer RNNs 386 42 Exploding or Disappearing Activations 390 43 LSTM 390 43 Training a Language Model Using LSTMs 394 44 Regularizing an LSTM 394 44 Further Research 402