Deep Learning Exam Prep (80+ Questions) – CS 7643, CNNs, Transfer Learning, Loss Functions | AI/ML

This document provides over 80 thoroughly answered questions for CS 7643: Deep Learning, focusing on key exam material for the 2025/2026 academic year. Designed for graduate-level computer science and AI students, it covers fundamental and advanced deep learning concepts tested in Exam 3, including model architectures, optimization strategies, and interpretability tools. Topics include error types (modeling, estimation, and optimization errors), transfer learning workflows, and network architectures such as AlexNet, VGGNet, Inception, ResNet, and Mask R-CNN. The document also reviews concepts of equivariance and invariance, saliency maps, Class Activation Mapping (CAM, Grad-CAM), and adversarial examples. In-depth explanations are provided for: Loss functions: L1, L2, MSE, Binary Cross-Entropy, Focal Loss Object detection models: SSD, YOLO, R-CNN variants CNN operations: receptive fields, transpose convolutions, pooling layers Parameter and memory calculations for Conv, FC, and Pooling layers Style transfer and content loss (using Gram matrices) Visualization and interpretability techniques in CNNs This document is highly recommended for: – Master’s and PhD students in Computer Science, AI, or Machine Learning – Students preparing for CS 7643 exams or deep learning certification – Individuals enrolled in deep learning or neural network modeling courses at US universities It serves as a detailed and structured exam prep tool for both theoretical concepts and practical implementation insights in modern deep learning systems. Keywords: CS 7643, deep learning, CNN architectures, transfer learning, ResNet, VGGNet, AlexNet, Grad-CAM, saliency maps, error types, adversarial examples, convolutional layers, object detection, YOLO, SSD, R-CNN, style transfer, loss functions, MSE, Focal Loss, receptive field, transpose convolution