Projective Latent Interventions for Understanding and Fine-tuning Classifiers
This work addresses the need for better interpretability and control in classifiers, particularly for domain experts in medical applications like fetal ultrasound imaging, though it appears incremental as it builds on existing embedding methods.
The authors tackled the problem of interpreting high-dimensional latent representations in neural network classifiers by introducing Projective Latent Interventions (PLIs), a technique that retrains classifiers by back-propagating manual changes from low-dimensional embeddings, and demonstrated its application in fetal ultrasound imaging to enhance performance for specific class pairs.
High-dimensional latent representations learned by neural network classifiers are notoriously hard to interpret. Especially in medical applications, model developers and domain experts desire a better understanding of how these latent representations relate to the resulting classification performance. We present Projective Latent Interventions (PLIs), a technique for retraining classifiers by back-propagating manual changes made to low-dimensional embeddings of the latent space. The back-propagation is based on parametric approximations of t-distributed stochastic neighbourhood embeddings. PLIs allow domain experts to control the latent decision space in an intuitive way in order to better match their expectations. For instance, the performance for specific pairs of classes can be enhanced by manually separating the class clusters in the embedding. We evaluate our technique on a real-world scenario in fetal ultrasound imaging.