Binary is Unary

• 

  Prompt #1

• 

  Prompt #2

• 

  Prompt #3

• 

  Prompt #4

• 

  Prompt #5

Inspiration

• We first wanted to detect stereotypes in factors such as race and gender in ML generated images. Using a stable diffusion algorithm we were able to test many different prompts and gauge how the ML model reflects such biases through its training process.

What it does 🚗

• The model takes in a prompt (phrase or sentence) and generates an image based on what the sentence says.

How we built it 👷‍♂️

• Language: Python
• Framework: PyTorch

Challenges we ran into🕸

• We ran into issues trying to determine what specific factors to look for and how to quantify bias. By trying out many different types of prompts we were able to test a variety of inputs and get more comprehensive results.

Accomplishments that we're proud of 🏆

• Learned how to utilize the PyTorch framework
• Established knowledge of the basics of Machine Learning and Deep Learning
• Understood the prevalence of ML models and real-world applications

What we learned 🧠

• Learned how Deep Learning networks function
• Learned how to detect elements of bias in ML models
• Learned to download datasets using PyTorch

What's next? 🔮

Some questions we have for future analysis…
• Does the sentence structure inputted as the prompt change the way a model interprets it?
• Does the length and complexity of the sentence alter the way a model processes information?

Built With

• googlecolab
• python
• pytorch