Scribd
Upload
7 views3 pages

Class Notes Pyhon 2

The document outlines various advanced topics in deep learning, including the Transformer architecture, generative deep learning techniques, and best practices for model training. It discusses sequence-to-sequence learning, neural style transfer, and generative adversarial networks, along with their implementations in Keras. Additionally, it addresses the limitations of deep learning, the future of AI, and the importance of staying updated in the rapidly evolving field.

Uploaded by

Yavuz Taşoluk
Copyright
© All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF or read online on Scribd
7 views3 pages

Class Notes Pyhon 2

The document outlines various advanced topics in deep learning, including the Transformer architecture, generative deep learning techniques, and best practices for model training. It discusses sequence-to-sequence learning, neural style transfer, and generative adversarial networks, along with their implementations in Keras. Additionally, it addresses the limitations of deep learning, the future of AI, and the importance of staying updated in the rapidly evolving field.

Uploaded by

Yavuz Taşoluk
Copyright
© All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF or read online on Scribd
‘CONTENTS. 114. The Transformer architecture 336 Understanding self-attention 337 * Multihead attention 341 The Transformer encoder 342 = When to use sequence models over bagof-words models 349 11.5 Beyond text classification: Sequence-to-sequence learning 350 A machine translation example 351 * Sequence-to-sequence learning with RNNs 354 * Sequenceto-sequence learning with Transformer 358 1 2 Generative deep learning 364 12.1 Text generation 366 A brief history of generative deep learning for sequence generation 366 * How do you generate sequence data? 367 The importance of the sampling strategy 368 = Implementing text generation with Keras 369» A text-generation callback with variabletemperature sampling 372 * Wrapping up 376 122 DeepDream 376 Implementing DeepDream in Keras 377 * Wrapping up 383 12.3. Neural style transfer 383 The content loss 384 * Thestyleloss 384 = Neural style transfer in Keras. 385 * Wrapping up 391 12.4 Generating images with variational autoencoders 391 Sampling from latent spaces of images 391 * Concept vectors ‘for image editing 393 * Variational autoencoders 393 Implementing a VAE with Keras 396 = Wrapping up 401 12.5 Introduction to generative adversarial networks 401 A schematic GAN implementation 402 = A bag of tricks 403 * Getting our hands on the CelebA dataset 404 ‘The discriminator 405 * The generator 407 * The adversarial network 408 « Wrapping up 410 1 Best practices for the real world 412 13.1 Getting the most out of your models 413 Hyperparameter optimization 413 = Model ensembling 420 13.2 Scaling-up model taining 421 Speeding up training on GPU with mixed precision 422 ‘Multi-GPU training 425 © TPU training 428 ‘CONTENTS. 1 Conclusions 431 14.1 14.2 14.3 14.4 14.5 14.6 14.7 Key concepts in review 432 Various approaches toAL 432 © What makes deep learning special within the field of machine learning 432 = How to think about deep learning 433 * Key enabling technologies. 434° The ‘universal machine learning workflow 435 * Key network architectures 436 * The space of possibilities 440 The limitations of deep learning 442 ‘The risk of anthropomorphizing machine learning models 443 Automatons us, intelligent agents 445 * Local generalization vs extreme generalization 446 * The purpose of intelligence 448 Climbing the spectrum of generalization 449 Setting the course toward greater generality in AI 450 On the importance of setting the right objective: The shortcut rule 450 * Anew target 452 Implementing intelligence: The missing ingredients 454 Intelligence as sensitivity o abstract analogies 454 * The two poles of abstraction 455 * The missing half of the picture 458 The future of deep learning 459 ‘Models as programs 460 = Blending together deep learning and program synthesis 461 * Lifelong learning and modular subroutine reuse 463 © The long-term vision 465 Staying up to date in a fastmoving field 466 Practice on real-world problems using Kas the latest developments on arXiv 466 * ecosystem 467 Final words 467 index 469

You might also like