Grad Seminar: A Representational Response Analysis Framework For Convolutional Neural Networks

Designing a CNN is not a straightforward process. Model architecture design, learning strategies, and data selection and processing must all be precisely tuned for a researcher to produce even a non-random performing model. When building a new model, researchers will rely on quantitative metrics to guide the development process. Typically, these metrics revolve around model performance characteristics constraints (e.g., accuracy, recall, precision, robustness) and computational (e.g., number of parameters, number of FLOPs), while the learned internal data processing behaviour of a CNN is ignored.

In this work, we propose a novel analytic framework that offers a broad range of complementary metrics that can be used by a researcher to study the internal behaviour of a CNN. We call the proposed framework Representational Response Analysis (RRA). The RRA framework is built around a common computational kNN based model of the latent embeddings of a dataset at each layer in a CNN.  Using RRA we study the impact of specific CNN design choices. Specifically, we use RRA to investigate the consequences on a CNN's latent representation when training with and without data augmentations, and to understand the latent embedding symmetries across different pooled spatial resolutions.  Using the insights from the pooled spatial resolution experiments we propose a novel CNN attention-based building block that is specifically designed to take advantage of key latent properties of a ResNet.