Timezone: »

Mechanistic Mode Connectivity
Ekdeep Singh Lubana · Eric Bigelow · Robert Dick · David Krueger · Hidenori Tanaka

Wed Jul 26 05:00 PM -- 06:30 PM (PDT) @ Exhibit Hall 1 #218

We study neural network loss landscapes through the lens of mode connectivity, the observation that minimizers of neural networks retrieved via training on a dataset are connected via simple paths of low loss. Specifically, we ask the following question: are minimizers that rely on different mechanisms for making their predictions connected via simple paths of low loss? We provide a definition of mechanistic similarity as shared invariances to input transformations and demonstrate that lack of linear connectivity between two models implies they use dissimilar mechanisms for making their predictions. Relevant to practice, this result helps us demonstrate that naive fine-tuning on a downstream dataset can fail to alter a model's mechanisms, e.g., fine-tuning can fail to eliminate a model's reliance on spurious attributes. Our analysis also motivates a method for targeted alteration of a model's mechanisms, named connectivity-based fine-tuning (CBFT), which we analyze using several synthetic datasets for the task of reducing a model's reliance on spurious attributes.

Author Information

Ekdeep Singh Lubana (University of Michigan; CBS, Harvard University)
Eric Bigelow (Harvard University)
Robert Dick (University of Michigan)
David Krueger (MILA (University of Montreal))
Hidenori Tanaka (Harvard University, Harvard University)

More from the Same Authors