DhilipSanjay / ANN-with-Momentum

Artificial Neural Networks with Back Propagation & Momentum (No external libraries used)

ANN-with-Momentum

Artificial Neural Networks with Back Propagation and Momentum (Without using keras and tensorflow)

Models

1. Emergency Vehicle Classification - Ambulance
2. Classify image as cat or not cat

Implementation Steps

1. Import necessary libraries

   • numpy
   • matplotlib
   • glob
   • cv2
   • random
   • os

2. Download and preprocess the dataset

   • Load Training and Test Datasets
   • Shuffle Training and Test Datasets
   • Resize and Normalize the images

3. Initialize random weights and biases

   • Create dictionary to store weights and biases
   • Initialize weights and biases to zero for backpropagation

4. Fix all hyperparameters

   • Learning rate
   • Number of epochs
   • Number of layers
   • Number of units in each layer
   • Momentum (𝛂)

5. Until the termination condition is met, Do

   • For each training example, Do
     • Forward Propagation: Calculate the output of each unit and propagate through the layers.
     • Backward Propagation: Calculate the errors (for output and hidden units). Calculate the change in weights & biases. Add the momentum and update the weights & biases.

---

Emergency Vehicle Classification - Ambulance

Accuracy and Error - ANN without Momentum

Accuracy and Error - ANN with Momentum

Classification made by the model

---

References

• Building your Neural Network Step by Step
• Machine Learning - Tom Mitchell

Note

This is a project done for Machine Learning Course (Team of 4)