IDL Lect 5A Why Deep Learning Architecture Matters for Explainable AI

In this lecture of INF-8605: Interpretability in Deep Learning, Prof. Dilip K. Prasad explains why deep learning architecture is important for both model performance and model interpretability. Different types of data need different neural network structures. Images, text, time series, graphs, tabular data, and multimodal data all have different patterns. This is why architectures such as MLPs, CNNs, RNNs/LSTMs, Transformers, and Graph Neural Networks are designed in different ways. This lecture explains how architecture affects: what features a model learns how information flows through the model where information is stored which explanation method is suitable how humans can inspect model behavior This lecture builds a foundation for later explainable AI methods such as feature attribution, saliency maps, Grad-CAM, attention analysis, probing, and representation analysis. Key topics covered What is deep learning architecture? Why different architectures exist Architecture and data type Architecture and explanation method Why interpretability depends on architecture MLP, CNN, RNN/LSTM, Transformer, and GNN comparison Architecture-aware interpretability Suggested audience This lecture is suitable for students, researchers, engineers, and AI learners who want to understand deep learning and explainable AI in a structured way. #DeepLearning #Interpretability #ExplainableAI #XAI #NeuralNetworks #MachineLearning #ArtificialIntelligence #CNN #Transformer #GNN