3 papers
cs.GR2025
Uncovering Conceptual Blindspots in Generative Image Models Using Sparse Autoencoders
Matyas Bohacek, Thomas Fel, Maneesh Agrawala +1
Despite their impressive performance, generative image models trained on large-scale datasets frequently fail to produce images with seemingly simple concepts -- e.g., human hands…
cs.LG2024
Abrupt Learning in Transformers: A Case Study on Matrix Completion
Pulkit Gopalani, Ekdeep Singh Lubana, Wei Hu
Recent analysis on the training dynamics of Transformers has unveiled an interesting characteristic: the training loss plateaus for a significant number of training steps, and then…
cs.LG2024
Towards an Understanding of Stepwise Inference in Transformers: A Synthetic Graph Navigation Model
Mikail Khona, Maya Okawa, Jan Hula +5
Stepwise inference protocols, such as scratchpads and chain-of-thought, help language models solve complex problems by decomposing them into a sequence of simpler subproblems. Desp…