Group Name: Autoregression, Revisited
Group Members: Luke Cho, Rachel Wetherington
Project Description: Our idea is to replicate the results of the fourth module of the fourth ECoG tutorial (Training and evaluating encoding models) by comparing the results of a different but similar autoregressive model (XLNet). We will then compare the performance of XLNet to GPT-2 to see which one better predicts actual human data in regards to language prediction. In particular, we are interested in seeing if XLNet's use of incorporating context on both sides of a given word will improve its predictive ability relative to GPT-2.
Group Name: Autoregression, Revisited
Group Members: Luke Cho, Rachel Wetherington
Project Description: Our idea is to replicate the results of the fourth module of the fourth ECoG tutorial (Training and evaluating encoding models) by comparing the results of a different but similar autoregressive model (XLNet). We will then compare the performance of XLNet to GPT-2 to see which one better predicts actual human data in regards to language prediction. In particular, we are interested in seeing if XLNet's use of incorporating context on both sides of a given word will improve its predictive ability relative to GPT-2.