Part 1 of 3 in a series exploring generative models.
The first in a series of paper replications that will collectively lay the foundation for a full replication of the stable diffusion model from scratch.
Using huggingface components to build a stable diffusion pipeline from scratch. Used fast.ai project as inspiration.
Given two premises that form a valid syllogism, this autoregressive model can accurately complete the syllogism by generating a conclusion.
Given two premises this validation model can classify validity with 85% accuracy on a 50/50 split dataset.
A PyTorch playing card multi-label classifier from scratch. Basic CNN, EfficientNet with & without augmentation. Achieves 98% accuracy.
Achieving 83% accuracy on the Titanic Kaggle competition dataset with a DNN built from scratch.
Project
Building and deploying an emotion classifying twitter bot that responds to users who prompt the bot with a # of interest. Bot uses a pretrained BERT encoder fine tuned on a tweet emotion dataset.