Deep Learning Exam Prep (100+ Questions) – CS 7643, CNNs, Transfer Learning, Adversarial Images | AI/ML

This document features over 100 exam-focused questions and expertly explained answers for CS 7643 – Deep Learning, specifically tailored for Exam 2 (2025/2026 academic year). It is ideal for students in advanced machine learning or artificial intelligence courses looking to master foundational and applied concepts in deep learning architecture and model evaluation. The content thoroughly explores convolutional neural networks (CNNs), including core mechanisms like receptive fields, weight sharing, padding, stride, pooling, and kernel configurations. Key architecture models such as LeNet, AlexNet, VGG, Inception, ResNet, EfficientNet, and U-Net are broken down for understanding both their structural designs and use-case advantages. Advanced topics include: Loss and error types: modeling error, estimation error, optimization error Transfer learning strategies (finetune vs freeze) and their limitations Visual interpretability: saliency maps, Grad-CAM, texture vs shape bias Adversarial images, pixel attacks, and robustness methods Image segmentation and detection models (YOLO, R-CNN, Mask R-CNN) Fully convolutional networks, transpose convolution, upsampling layers Bounding box regression, multi-headed CNNs, mAP, and COCO dataset It also includes detailed parameter calculations, CNN layer operations, evaluation metrics, and design principles behind object recognition, detection, and segmentation models. This study resource is highly relevant for: – Graduate students in Computer Science or Artificial Intelligence – Deep learning, neural networks, or machine vision coursework – Individuals preparing for CS 7643 exams or AI/ML certification programs – Academic programs at US universities focusing on applied AI Structured for clarity and depth, this document supports concept reinforcement, problem-solving, and technical exam preparation. Keywords: CS 7643, deep learning, CNN architecture, AlexNet, VGGNet, ResNet, Inception, EfficientNet, LeNet, U-Net, transfer learning, receptive field, adversarial examples, Grad-CAM, texture bias, object detection, YOLO, SSD, RPN, image segmentation, loss functions, mAP, COCO dataset, neural networks