Machine Learning + Web Development
INK Lab
INK Lab is a group of reseachers working on next-generation machine intelligence techniques for label-efficient machine learning, knowledge-guided natural language processing and knowledge reasoning.
Over the summer, my goal was to create a web-based application to visualize and demonstrate the work that had been done to “explain” the algorithms behind the proposed neural network model. By doing so, other researchers and users of this model would be able to see the visual breakdown of sentence classification, using a hierarchical type of explanation.
The Challenge
The biggest challenge with this project was understanding the code. Many of the web applications required an understanding of the code in order to implement an interactive prototype.
1. Allow users to input text and see the sentiment analysis of their input.
2. Provide users with an explanation and background of the analysis.
1. We trained 2 language models: LSTM and BERT for sentiment analysis and relation extraction.
2. We then used the Sampling and Occlusion algorithm to explain and visualize trained results.
3. Finally, we built the web application using a Flask framework.
The Prototypes:
4. Inform: On the web application, we wanted to provide an easily understood explanation for the sentiment analysis.
The Prototypes: