applications of deep learning 25-04-2019...deep learning methods employ multiple processing layers...

20
Deep Learning: Theory and Practice 25-04-2019 Applications of Deep Learning

Upload: others

Post on 17-Jul-2020

1 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Applications of Deep Learning 25-04-2019...Deep learning methods employ multiple processing layers to learn hierarchical representations of data and have produced state-of-the-art

Deep Learning: Theory and Practice

25-04-2019Applications of Deep Learning

Page 2: Applications of Deep Learning 25-04-2019...Deep learning methods employ multiple processing layers to learn hierarchical representations of data and have produced state-of-the-art

Natural Language Processing

https://arxiv.org/abs/1708.02709#

Page 3: Applications of Deep Learning 25-04-2019...Deep learning methods employ multiple processing layers to learn hierarchical representations of data and have produced state-of-the-art

Natural Language Processing

Page 4: Applications of Deep Learning 25-04-2019...Deep learning methods employ multiple processing layers to learn hierarchical representations of data and have produced state-of-the-art

Word Embeddings

Learn a vector representation for each word

Page 5: Applications of Deep Learning 25-04-2019...Deep learning methods employ multiple processing layers to learn hierarchical representations of data and have produced state-of-the-art

Language Modeling

Page 6: Applications of Deep Learning 25-04-2019...Deep learning methods employ multiple processing layers to learn hierarchical representations of data and have produced state-of-the-art

Sentiment Analysis/Summarization

Page 7: Applications of Deep Learning 25-04-2019...Deep learning methods employ multiple processing layers to learn hierarchical representations of data and have produced state-of-the-art

Machine Translation

Page 8: Applications of Deep Learning 25-04-2019...Deep learning methods employ multiple processing layers to learn hierarchical representations of data and have produced state-of-the-art

Architectures

Recurrent Neural Networks

Attention Based Models

Page 9: Applications of Deep Learning 25-04-2019...Deep learning methods employ multiple processing layers to learn hierarchical representations of data and have produced state-of-the-art

Speech Processing

https://www.microsoft.com/en-us/research/wp-content/uploads/2016/02/HintonDengYuEtAl-SPM2012.pdf

Page 10: Applications of Deep Learning 25-04-2019...Deep learning methods employ multiple processing layers to learn hierarchical representations of data and have produced state-of-the-art

Speech Recognition

2018

5.3%

Claims of human parity !!!

Page 11: Applications of Deep Learning 25-04-2019...Deep learning methods employ multiple processing layers to learn hierarchical representations of data and have produced state-of-the-art

Speaker Recognition Embedding

Embeddings used for verificationtask.

Page 12: Applications of Deep Learning 25-04-2019...Deep learning methods employ multiple processing layers to learn hierarchical representations of data and have produced state-of-the-art

Speech Enhancement

Removing Noise Using a Supervised Training Mechanism

Page 13: Applications of Deep Learning 25-04-2019...Deep learning methods employ multiple processing layers to learn hierarchical representations of data and have produced state-of-the-art

Architectures

Deep neural networks

Bidirectional LSTM Networks

Page 14: Applications of Deep Learning 25-04-2019...Deep learning methods employ multiple processing layers to learn hierarchical representations of data and have produced state-of-the-art

Image Processing

Page 15: Applications of Deep Learning 25-04-2019...Deep learning methods employ multiple processing layers to learn hierarchical representations of data and have produced state-of-the-art

Object Identification

Page 16: Applications of Deep Learning 25-04-2019...Deep learning methods employ multiple processing layers to learn hierarchical representations of data and have produced state-of-the-art

Object Segmentation

Page 17: Applications of Deep Learning 25-04-2019...Deep learning methods employ multiple processing layers to learn hierarchical representations of data and have produced state-of-the-art

Face Detection

Page 18: Applications of Deep Learning 25-04-2019...Deep learning methods employ multiple processing layers to learn hierarchical representations of data and have produced state-of-the-art

Architectures Commonly Used

Convolutional networks

Resnet Architectures

Page 19: Applications of Deep Learning 25-04-2019...Deep learning methods employ multiple processing layers to learn hierarchical representations of data and have produced state-of-the-art

Future DirectionsUnsupervised and Semi-supervised learning

• How little of labelled data is good enough ? • Transfer Learning Principles

Understanding And Insights Into Deep Networks • Understanding parameters and representations.

Deep Generative Models • Using deep learning for generative modeling

Page 20: Applications of Deep Learning 25-04-2019...Deep learning methods employ multiple processing layers to learn hierarchical representations of data and have produced state-of-the-art

Software ToolsTheano

http://deeplearning.net/software/theano/

PyTorch https://pytorch.org/

TensorFlow https://www.tensorflow.org/

Kaldi http://kaldi-asr.org/