Science

Circuit Probing Diagram

Uncovering Intermediate Variables in Transformers using Circuit Probing

Can we find subnetworks that compute high-level intermediate variables in transformers?

NeuroSurgeon Diagram

NeuroSurgeon: A Toolkit for Subnetwork Analysis

We present a python package that simplifies subnetwork analysis.

Break It Down Diagram

Break It Down: Evidence for Structural Compositionality in Neural Networks

Do neural networks self-organize into modular components when solving compositional tasks?

Same Different Diagram

Deep Neural Networks Can Learn Generalizable Same-Different Visual Relations

Can vision models learn same-different relations that generalize to different datasets?

Subtask Induction Diagram

Instilling Inductive Biases with Subnetworks

Can we influence the solutions that neural networks learn by transferring subnetworks from trained models to randomly intialized models?

Representations of Syntax Diagram

Representations of Syntax [MASK] Useful: Effects of Constituency and Dependency Structure in Recursive LSTMs

Which tree-structured neural network imparts the best inductive bias for syntactic agreement?

Can You Hear Me Now Diagram

Can You Hear Me Now?: Sensitive Comparisons of Human and Machine Perception

How might researchers make fair comparisons between human and machine perception systems?

Unequal Representations Diagram

Unequal Representations: Analyzing Intersectional Biases in Word Embeddings Using Representational Similarity Analysis

We know BERT encodes racial and gender biases. Does it also encode intersectional biases?

Can You Hear Me Now Diagram

Picking BERT’s Brain: Probing for Linguistic Dependencies in Contextualized Embeddings Using Representational Similarity Analysis

Can we use representational similarity analysis to uncover linguistic structure in word embeddings?