The Truth about Saturn

• 

  Poster

Inspiration

When taking an Astronomy course, I noticed that I was being given a series of facts that I then had to make inferences upon. If an Astronomy dataset could be created, maybe my model would discover any unturned stones in the field!

What it does

It takes in a dataset of sentences and how true they are. Using that information, the model attempts to asses the validity of statements given to it.

How we built it

I built it using bert-base-uncased in order to run faster. I chose Adam to be my optimizer with a learning rate of 1e-5. For my loss, I chose MSELoss. I trained on 9000 statements over 2 epochs and tested against 2649 statements.

Challenges we ran into

I ran into issues when trying to combine both datasets. The strings differed and so did the truth values.

Accomplishments that we're proud of

I created a working final product!

What we learned

I learned that I got this! I also learned how to take my models and add some interactive elements.

What's next for The Truth about Saturn

I want the model to not output values for statements it does not know 100%. I also want the user to be able to correct the model.

Built With

• colab
• google-notebook
• numpy
• python
• pytorch