Project: Predictive Coding with CNN on MNIST Dataset

https://colab.research.google.com/drive/10yLJK2WheD733IbUA2oapxilS0fLVkBh?usp=sharing

Objective

To develop a Convolutional Neural Network (CNN) that classifies handwritten digits from the MNIST dataset. The project demonstrates a complete pipeline from data preprocessing to model evaluation and visualization of results.

Key Highlights

• Dataset: MNIST handwritten digit dataset, a benchmark dataset for image classification tasks.

• Technologies Used:

  • Python: Core programming language.

  • TensorFlow/Keras: Framework for building and training the CNN model.

  • Matplotlib: For visualizing training performance.

Steps in the Project

1. Data Preparation:

   • Reshaped and normalized the MNIST images to prepare them for the CNN.

   • Converted labels to one-hot encoded format for multiclass classification.

2. Model Architecture:

   • Designed a CNN with multiple convolutional, pooling, and dense layers:

     • 3 convolutional layers with ReLU activation.

     • Max-pooling layers for feature reduction.

     • Fully connected dense layers for classification.

     • Softmax activation for multiclass output.

1. 
   
   
   
   
   

   Training and Evaluation:

   • Optimized the model using the Adam optimizer and categorical cross-entropy loss function.

   • Achieved robust training performance across 10 epochs with validation on the test set.

2. Results:

   • Test Accuracy: Achieved a competitive accuracy on the MNIST test set.

   • Performance Visualizations:

     • Plotted training and validation accuracy and loss across epochs.

Learning Outcomes

• Gained hands-on experience in deep learning model development and evaluation.

• Learned to interpret training dynamics using visualizations and performance metrics.

• Successfully applied machine learning techniques to solve a classic image classification problem.