Identifying Interpretable Features in Convolutional Neural Networks
Keywords: interpretability, convolutional neural networks, psychophysics, monosemanticity
TL;DR: We propose an interpretability metric based on human psychophysics and quantify monosemantic directions in CNNs that do not align with individual neurons.
Abstract: Single neurons in neural networks are often 'interpretable' in that they represent individual, intuitively meaningful features. However, many neurons exhibit $\textit{mixed selectivity}$, i.e., they represent multiple unrelated features. A recent hypothesis proposes that features in deep networks may be represented on non-orthogonal axes by multiple neurons, since the number of possible interpretable features in natural data is generally larger than the number of neurons in a given network. Accordingly, we should be able to find meaningful directions in activation space that are not aligned with individual neurons. Here, we propose (1) an automated method for quantifying visual interpretability that is validated against a large database of human psychophysics judgments of neuron interpretability, and (2) an approach for finding meaningful directions in network activation space. We leverage these methods to discover directions in convolutional neural networks that are more intuitively meaningful than individual neurons. In a series of analyses to understand this phenomenon we find, for instance, examples of $\textit{feature synergy}$ where pairs of uninterpretable neurons work together to encode interpretable features. These results contribute to a larger effort to automate interpretability research, providing a foundation for scaling bespoke perceptual judgments to the analysis of complex neural network models.
Primary Area: visualization or interpretation of learned representations
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2024/AuthorGuide.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors' identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 5795
Loading