Etash Guha

Overall, our paper provides a theoretical and empirical investigation on how scale impacts model forgetting. All code to reproduce our experiments is provided in our GitHub. Huge shout out to my coauthor @VihanL! 🧵8/8 https://t.co/nNh2yoQ1uE

We explore other connections between architecture and forgetting as predicted by our theoretical framework. In fact, our theoretical framework and experiments demonstrate that as the number of tasks seen increases, forgetting increases linearly. 🧵7/8 https://t.co/RkjmjJ7fz2