Deep Learning ( Professional Lvl )

60 hr

The program is structured into 20 sessions, each lasting 3 hours. Each session will blend theoretical explanations with practical hands-on labs. The following outline highlights the key topics covered:

1. Session 1: Introduction to Deep Learning and Neural Networks

• • Recap of Machine Learning fundamentals and motivation for Deep Learning

  • What is Deep Learning? History, Applications, Advantages, and Disadvantages

  • Introduction to Neural Networks: Neurons, Layers, Activation Functions

  • Forward Propagation and Backpropagation (Intuitive Understanding)

  • Introduction to Deep Learning Frameworks: TensorFlow, Keras, PyTorch (Choose one as the primary focus)

  • Setting up the Development Environment

  • Hands-on Project: Building a simple neural network from scratch (using NumPy or basic Python)

            2. Session 2: Deep Learning Framework Fundamentals (TensorFlow/Keras/PyTorch)

• • Introduction to Tensors and Operations

  • Building and Training a Simple Neural Network using the chosen framework

  • Defining Models: Sequential vs. Functional API

  • Optimizers: Gradient Descent, Adam, RMSprop

  • Loss Functions: Categorical Cross-Entropy, Mean Squared Error

  • Metrics: Accuracy, Precision, Recall, F1-score

  • Hands-on Project: Implementing a simple classification or regression task using the chosen framework.

           3. Session 3: Introduction to Convolutional Neural Networks

• • Motivation for CNNs: Image Recognition, Computer Vision

  • Convolutional Layers: Filters, Feature Maps, Strides, Padding

  • Pooling Layers: Max Pooling, Average Pooling

  • Activation Functions: ReLU, Sigmoid, Tanh

  • CNN Architecture: Convolutional Layers, Pooling Layers, Fully Connected Layers

  • Hands-on Project: Building a simple CNN to classify images from a small dataset (e.g., CIFAR-10).

           4. Session 4: CNN Architectures and Transfer Learning

• • Popular CNN Architectures: LeNet, AlexNet, VGGNet, ResNet, Inception

  • Transfer Learning: Using pre-trained models (e.g., ImageNet)

  • Fine-tuning Pre-trained Models

  • Data Augmentation Techniques

  • Hands-on Project: Implementing Transfer Learning using a pre-trained model on a custom image dataset.

           5. Session 5: Advanced CNN Techniques

• • Batch Normalization

  • Dropout

  • Regularization Techniques (L1, L2 Regularization)

  • Understanding and Mitigating Overfitting

  • Hands-on Project: Improving the performance of a CNN model by applying advanced techniques like batch normalization and dropout.

            6. Session 6: Object Detection with CNNs

• • Introduction to Object Detection

  • Object Detection Architectures: R-CNN, Fast R-CNN, Faster R-CNN, YOLO

  • Bounding Box Regression

  • Non-Maximum Suppression (NMS)

  • Hands-on Project: Implementing a simple object detection pipeline using a pre-trained object detection model.

           7. Session 7: Introduction to Recurrent Neural Networks

• • Motivation for RNNs: Natural Language Processing, Time Series Analysis

  • Recurrent Cells: Vanilla RNN, LSTM, GRU

  • Backpropagation Through Time (BPTT)

  • Vanishing and Exploding Gradient Problems

  • Hands-on Project: Building a simple RNN to predict the next character in a sequence.

          8. Session 8: Long Short-Term Memory (LSTM) Networks

• • LSTM Architecture: Input Gate, Forget Gate, Output Gate, Cell State

  • Advantages of LSTMs over Vanilla RNNs

  • Hands-on Project: Building an LSTM model for sentiment analysis or text classification.

         9. Session 9: Gated Recurrent Units (GRUs)

• • GRU Architecture: Update Gate, Reset Gate

  • Advantages and Disadvantages of GRUs compared to LSTMs

  • Hands-on Project: Building a GRU model for time series forecasting.

        10. Session 10: Sequence-to-Sequence Models

• Encoder-Decoder Architecture

• Attention Mechanism (Conceptual Introduction)

• Machine Translation

• Hands-on Project: Building a sequence-to-sequence model for a simple translation task.

        11. Session 11: Generative Adversarial Networks (GANs)

• Introduction to GANs: Generator and Discriminator Networks

• Training GANs

• Applications of GANs: Image Generation, Style Transfer

• Hands-on Project: Building a simple GAN to generate images.

        12. Session 12: Autoencoders

• Introduction to Autoencoders: Encoder and Decoder Networks

• Types of Autoencoders: Undercomplete Autoencoders, Sparse Autoencoders, Variational Autoencoders (VAEs)

• Applications of Autoencoders: Dimensionality Reduction, Anomaly Detection

• Hands-on Project: Building an autoencoder for anomaly detection.

        13. Session 13: Transformers

• • Introduction to Transformers: Attention Mechanism in Detail

  • Self-Attention

  • Multi-Head Attention

  • Transformer Architecture: Encoder and Decoder

  • Hands-on Project: Exploring and understanding pre-trained Transformer models (Hugging Face Transformers).

     14. Session 14: Graph Neural Networks (GNNs)

• Introduction to Graph Neural Networks

• Graph Convolutional Networks (GCNs)

• Applications of GNNs: Social Network Analysis, Drug Discovery

• Hands-on Project: Implementing a simple GNN for node classification.

( Check out the link for the other topics covered in Session 15...Session 20 )

• Total Credit Hours: 60
• Session Length: 3 hours per session
• Number of Sessions: 20
• Delivery Method: Remotely

• Upon completion of this course, trainees will be able to:

1. Describe the differences between Machine Learning and Deep Learning.

2. Explain the concept of neural networks and their building blocks (neurons, layers, activation functions).

3. Implement and train various deep learning models using TensorFlow, Keras, or PyTorch.

4. Preprocess data specifically for deep learning tasks.

5. Select appropriate deep learning architectures for different problems.

6. Tune hyperparameters to optimize model performance.

7. Diagnose and address common problems encountered during deep learning training (e.g., overfitting, vanishing gradients).

8. Deploy trained deep learning models using frameworks like TensorFlow Serving or Flask.

9. Contribute to deep learning projects and research.

10. Critically evaluate the performance and limitations of deep learning models.