Ahmad Tawil

IBM · IBM AI Engineering Professional Certificate

Introduction to Deep Learning & Neural Networks with Keras

Introduction to Deep Learning & Neural Networks with Keras certificate

Neural network fundamentals through Keras — CNNs, RNNs, and transformers, capped by an aircraft-damage classification and captioning project.

Completed 2025-09-09

Verify credentialCoursework repository

🎓 Certificate #

Certificate

Provider: IBM (via Coursera)
Verified Certificate: View on Coursera

📘 Course Overview #

This course is part of the IBM AI Engineering Professional Certificate and provides a strong foundation in deep learning, focusing on the practical implementation of neural networks using Keras.

Topics covered include:

  • Fundamentals of artificial neural networks (ANNs)
  • Forward and backward propagation
  • Gradient descent and optimization
  • Activation functions (Sigmoid, ReLU, Softmax, etc.)
  • Dealing with vanishing gradients
  • Keras Sequential vs Functional API
  • Convolutional Neural Networks (CNNs)
  • Recurrent Neural Networks (RNNs)
  • Transformers
  • Autoencoders
  • End-to-end projects with image classification and captioning

🧪 Hands-On Labs & Projects #

Module 1: Foundations #

  • Implemented a simple neural network forward pass manually (NumPy).
  • Learned how weights, biases, and activation functions produce outputs.

Module 2: Training Neural Networks #

  • Backpropagation Lab: Solved the XOR problem step-by-step.
  • Activation Functions Lab: Explored Sigmoid vs ReLU, and vanishing gradient effects.

Module 3: Keras Basics #

  • Built models using Sequential and Functional API.
  • Applied classification (Car dataset) and regression (Concrete dataset).
  • Learned how .fit() automates forward/backward propagation.

Module 4: Advanced Architectures #

  • CNN Lab: Implemented image classification with Conv2D, MaxPooling, Flatten, Dense.
  • Transformers Lab: Implemented self-attention, cross-attention, and transformer blocks.
  • RNNs: Learned how hidden states retain sequential memory.

Module 5: Final Project #

  • Aircraft Damage Classification & Captioning
    • Used VGG16 as a pretrained feature extractor for binary classification (dent vs crack).
    • Built a custom Keras layer integrating BLIP (Transformer) for captioning & summarization.
    • Achieved ~84% test accuracy on classification + generated natural-language summaries.
    • Final Project Notebook

📊 Key Learnings #

  • Transfer learning with pretrained CNNs for efficient classification.
  • Building and training models with Keras (Sequential & Functional).
  • Handling overfitting with dropout, data augmentation, early stopping.
  • Integrating transformer-based models for image captioning.
  • Creating custom Keras layers with TensorFlow.
  • Combining structured outputs (labels) with natural language explanations (captions).

📌 Tools & Libraries #

  • TensorFlow & Keras
  • NumPy & Matplotlib
  • Hugging Face Transformers
  • PIL (Image Processing)

🔗 Next Steps #

This course set the stage for:

  • Advanced Deep Learning (CNNs, RNNs, Transformers)
  • Generative AI applications
  • Integration of deep learning in real-world projects like Trackr and NeuroFocus

🏆 Completion #

This repository serves as a portfolio-ready record of my work in Deep Learning with Keras, demonstrating both foundational understanding and practical implementations through labs and projects.