4 papers · 1 filter
Decomposing Representation Space into Interpretable Subspaces with Unsupervised Learning
Xinting Huang, Michael Hahn
Understanding internal representations of neural models is a core interest of mechanistic interpretability. Due to its large dimensionality, the representation space can encode var…
Understanding the Emergence of Seemingly Useless Features in Next-Token Predictors
Mark Rofin, Jalal Naghiyev, Michael Hahn
Trained Transformers have been shown to compute abstract features that appear redundant for predicting the immediate next token. We identify which components of the gradient signal…
Lower Bounds for Chain-of-Thought Reasoning in Hard-Attention Transformers
Alireza Amiri, Xinting Huang, Mark Rofin +1
Chain-of-thought reasoning and scratchpads have emerged as critical tools for enhancing the computational capabilities of transformers. While theoretical results show that polynomi…
InversionView: A General-Purpose Method for Reading Information from Neural Activations
Xinting Huang, Madhur Panwar, Navin Goyal +1
The inner workings of neural networks can be better understood if we can fully decipher the information encoded in neural activations. In this paper, we argue that this information…