Visual Interpretation of CNN Prediction Through Layerwise Sequential Selection of Discernible Neurons

In recent years, researchers have been working to interpret the insights of deep neural networks in pursuit overcoming their opaqueness and so-called ‘black-box’ tag. this work, we present a new posthoc visual interpretation technique that finds out discriminative image regions contributing highly towards networks’ prediction. We select most discernible set neurons per layer engineer forward pass operation gradually reach locations. While searching for neurons, existing approaches either end up with low-resolution visualization maps, or suffer lack neuron discriminativity way. Moreover, some methods concentrate only on current information overlooking meaningful from adjacent layers, limiting overall scope selection. address these issues by exploring layer-to-layer connected structure obtaining contributions its along e.g., succeeding preceding layer. introduce score function where such are assembled appropriate priorities. Layerwise then selected based top scores, ensuring reliable credible validate our proposal through objective subjective evaluations examining performance terms models’ faithfulness human-trust, visualize efficacy over other methods. also perform sets sanity check experiments method show reliability as map.