Advanced Applied Deep Learning

Advanced Applied Deep Learning: Convolutional Neural Networks and Object Detection - Practice Course Structure

The course is structured to balance theory and hands-on practice, with a focus on building, fine-tuning, and optimizing CNN-based object detection models using Python and popular deep learning frameworks such as TensorFlow and PyTorch. New techniques are introduced each week, followed by practical sessions where students will implement and experiment with these concepts.

 Course Overview

• Duration: 16 weeks

• Tools & Libraries: Python, TensorFlow, Keras, PyTorch, OpenCV, YOLO, Faster R-CNN

• Goals:

  • Develop deep knowledge of CNNs for image classification and object detection.

  • Gain hands-on experience with state-of-the-art models and optimization techniques.

  • Implement models using Python and frameworks like TensorFlow and PyTorch.

Course Structure (Weeks 1-14)

Week 1: Introduction to Deep Learning and Neural Networks

• Practice: Build and train a simple neural network for image classification.

Week 2: Convolutional Neural Networks (CNNs) – Architecture and Layers

• Practice: Implement a CNN for image classification on CIFAR-10.

Week 3: Transfer Learning and Pre-trained Models

• Practice: Fine-tune a pre-trained VGG16 model on a custom dataset.

Week 4: Regularization Techniques and Preventing Overfitting

• Practice: Use dropout, data augmentation, and L2 regularization on a CNN model.

Week 5: Advanced Optimization Techniques

• Practice: Experiment with optimizers like Adam, RMSprop, and learning rate schedulers.

Week 6: Hyperparameter Tuning

• Practice: Perform hyperparameter tuning using grid search on a CNN model.

Week 7-9: Object Detection Fundamentals and Architectures

• Practice: Implement object detection models such as Faster R-CNN and SSD.

Week 10-12: YOLO and Real-Time Object Detection

• Practice: Implement YOLO for real-time object detection.

Week 13-14: Explainable AI and Model Optimization

• Practice: Apply Grad-CAM for visualization and optimize models using quantization and pruning.