10617 Final Project: Improving Multimodal Sentiment Analysis Using Multilogue-Net and Transformer Architecture
Final report linked here.
Our model weights can be found in their respective Jupyter Notebooks.
Folder containing Jupyter Notebook for our modified Multilogue-Net and transformer architecture; Jupyter Notebook contains instructions on how to run our model. Uses code from Multilogue-Net. We modified the model file, training script and requirements file. Our report explains in detail the theory behind our model.
Folder containing Jupyter Notebook for our baseline model. Baseline model was taken from MultiBench examples. We modified one of the supervised training scripts to produce an output that would evaluate the MAE metric for training and validation samples. Our report explains in detail the theory behind the baseline model.