Interpreting the Second-Order Effects of Neurons in CLIP
This work addresses the challenge of understanding neuron functions in vision-language models like CLIP, which is incremental as it builds on existing interpretation methods to uncover new insights and applications.
The authors tackled the problem of interpreting individual neurons in CLIP by introducing a 'second-order lens' to analyze their effects, revealing that neurons are polysemantic (e.g., representing unrelated concepts like ships and cars) and enabling applications such as generating adversarial examples and achieving zero-shot segmentation with improved performance.
We interpret the function of individual neurons in CLIP by automatically describing them using text. Analyzing the direct effects (i.e. the flow from a neuron through the residual stream to the output) or the indirect effects (overall contribution) fails to capture the neurons' function in CLIP. Therefore, we present the "second-order lens", analyzing the effect flowing from a neuron through the later attention heads, directly to the output. We find that these effects are highly selective: for each neuron, the effect is significant for <2% of the images. Moreover, each effect can be approximated by a single direction in the text-image space of CLIP. We describe neurons by decomposing these directions into sparse sets of text representations. The sets reveal polysemantic behavior - each neuron corresponds to multiple, often unrelated, concepts (e.g. ships and cars). Exploiting this neuron polysemy, we mass-produce "semantic" adversarial examples by generating images with concepts spuriously correlated to the incorrect class. Additionally, we use the second-order effects for zero-shot segmentation, outperforming previous methods. Our results indicate that an automated interpretation of neurons can be used for model deception and for introducing new model capabilities.