Rafał Bielech

Having a database of question-answers pairs from context, we can utilize a Universal Sentence Encoder from Google to quickly show top results to a newly asked questions using cosine similarity.

The text-to-SQL task falls under the semantic-parsing framework, a framework where natural language is converted to some logical form, like a symbolic logic, computer code, or in this case,SQL queries. The SQLNet model uses what is called a sketch to construct SQL queries. Sketch based approaches are popular among text-to-SQL models and, more generally, semantic parsing tasks. One reason to use sketches is to solve the so-called ”ordering issue”. Sketches impose a canonical structure on the output, whose dependencies allows us to intelligently populate the sketch.

Once the surveillance system was moved to run as an API, it was time to decouple the front-end and the back-end. In this project, I develop a light Django front-end application that interacts with two APIs through the use of Bearer tokens. This enables me to simplify the development of the front-end make it simpler to utilize in future projects.

Through iteration of changes to this application, this application currently is using a MobileNet SSD for person detection with a scalable API web interface. This project has seen multiple iterations, starting out from image background model to CNN to MobileNet SSD for person detection at home. Latest version is API-driven with authentication using web tokens.

This Interactive map visualization was a final project for a Data Visualization and Analysis class (COMPSCI590V). Heart disease is the leading cause of death in the United States, for most races. Alienating the causes of heart disease is very difficult, as different races and genders are affected differently. The map allows users to see how mortality affects various genders and races on the state and county level. Moreover, users can compare two different groups combinations.