Deephys: Deep Electrophysiology, Debugging Neural Networks under Distribution Shifts
This provides a debugging tool for researchers and practitioners to diagnose neural network failures in out-of-distribution scenarios, though it is incremental as it builds on existing electrophysiology concepts.
The paper tackles the problem of deep neural networks failing under distribution shifts by introducing Deephys, a tool that visualizes and analyzes neural activity to understand these failures, demonstrating its validity through quantitative analyses on multiple datasets and architectures.
Deep Neural Networks (DNNs) often fail in out-of-distribution scenarios. In this paper, we introduce a tool to visualize and understand such failures. We draw inspiration from concepts from neural electrophysiology, which are based on inspecting the internal functioning of a neural networks by analyzing the feature tuning and invariances of individual units. Deep Electrophysiology, in short Deephys, provides insights of the DNN's failures in out-of-distribution scenarios by comparative visualization of the neural activity in in-distribution and out-of-distribution datasets. Deephys provides seamless analyses of individual neurons, individual images, and a set of set of images from a category, and it is capable of revealing failures due to the presence of spurious features and novel features. We substantiate the validity of the qualitative visualizations of Deephys thorough quantitative analyses using convolutional and transformers architectures, in several datasets and distribution shifts (namely, colored MNIST, CIFAR-10 and ImageNet).