Text Generation using Recurrent Neural Networks
Shashwat Aggarwal (~imshashwataggarwal) |
Recurrent Neural Networks (RNNs) are the state of the art models that have shown great promise in many NLP tasks. There’s something magical about RNNs.
In the Talk, I will discuss the basic fundamentals of Recurrent Neural Networks, discuss the limitations of RNNs in generating texts and the problem of vanishing gradients. I will further discuss about LSTMs to overcome those problems and then train both character-level and word-level language models based on multi-layer LSTMs to generate text using Tensorflow and Numpy as dependancies.
The talk will be structured as below:
- What are Recurrent Neural Networks(RNNs)?
- Understanding the Vanishing Gradient Problem
- Discussing and Implementing LSTM in Tensorflow
- Training Character-level and Word-level language models on Harry Potter Corpus to generate some fun text following JK Rowling Style!
- Python programming fundamentals.
- Basics of Linear Algebra and Probability
- Basic Knowledge of Neural Networks
Content for the session can be found here.
Hi! I am Shashwat Aggarwal, third year CS student at Netaji Subhas Institute of Technology.
I am very passionate about Machine Learning and Deep learning. I have been working on ML and DL technologies from past 6 - 7 months and had worked on numerous projects involving Data Mining, Image Processing, Machine Learning, Deep Learning and Natural Language Processing.
I have a keen interest in research activities and I have worked as a Research Intern at National Informatics Centre New Delhi during Summer on Boosting Algorithms in supervised domain of machine learning. I am an avid python user and as called by the community Pythonista .