Deep Learning with Keras : Building an AI that Talks like Shakespeare or Trump
Cheuk Ting Ho (~Cheukting) |
Computers are getting closer to speak like humans. But can we even make them speak in particular ways, like as Trump or Shakespeare? In this hands on talk we will see how this is possible with the latest deep learning techniques combined with word embeddings and other advanced NLP technique.
In this talk, first we would introduce two neural network and machine learning mechanisms which in popular and widely used in NLP (natural language processing): Word Embeddings and Recurrent Neural Network. Word Embeddings is a way to extract the context of a word by “learning” its presence in a paragraph; while Recurrent Neural Network, including LSTM (long short-term-memory), enable us to “train” sequential data. After that, we will showcase how to implement these mechanisms in a neutral network. With that, we can “build” a machine to generate articles, plays or speeches in the style of the training corpus and have lots of fun.
In the first half of the talk, concepts of how Word Embeddings and LSTM works will be explained. Audiences will understand why this is essential in the field of NLP and why we are using it. In the second half, a code demo will be used to showcase how to implement these mechanisms. Through an example, audiences will learn how Keras is used together with Tensorflow and Python to build a sequential neutral network. We will showcase generating a paragraph using Shakespeare’s play and another one using Trump’s speech.
This talk is for people who have some experience with data science and understand the concept of how a neural network works, but would like to go deeper into the details of how does it applied to NLP to solve more complex AI problems. We used very simple code but did a complex task like text generation, that opens the door for a lot of people who wants to experiment with deep learning.
Basic concepts of Neural Network like Stochastic Gradient Descent and back propagation, as it will not be covered in the talk due to time limit.
Source code available on Github: https://github.com/Cheukting/Style-mimicking-text-generator
Example slides: https://slides.com/cheukting_ho/pylondinium18
After spending 5 years doing research in theoretical physics at Hong Kong University of Science and Technology, Cheuk has transferred her analytical and logical skills in natural science and built a career in data science. Cheuk is now a Data Scientist in Hotelbeds Groups which is one of the biggest worldwide wholesaler in travel business.
Cheuk constantly contributes to the community by giving AI and deep learning workshops, volunteering at Datakind for charities. At the same time participate open source projects like Pandas, Gensim and Dateutil. Cheuk has also been a guest speaker at University of Oxford and Queen Mary University of London, and various conferences including PyData Amsterdam, PyCon Israel and PyLondinium. Believing in gender equality, Cheuk is currently a co-organizer of AI club for Gender Minorities to support Tech Diversity and Inclusion.
Public speaker in technology, example of my talk on YouTube: https://www.youtube.com/watch?v=bQ2Qu63SYHw
Slide Share: https://www.slideshare.net/CheukTingHo/presentations
Speaker Bio: https://www.papercall.io/speakers/26654/