iccnn

simple convolution neural network for image classification

getting started

1. have python >=3.11.0 installed in your respective platform
2. clone the repository
3. create a python virtual environment python -m venv .venv
4. activate the python virtual environment source .venv/bin/activate for windows .venv/bin/activate
5. make sure the python in your path is from the virtual environment
6. install pip dependencies pip install -r requirements.txt
7. run python main.py

plan outline

week 1: environment setup and dataset preparation

thuva: environment setup

• install necessary libraries (tensorflow, keras, pytorch, etc.)
• set up a shared repository (e.g., github) for collaboration
• ensure all team members have the correct environment

person 2: dataset selection

• search for appropriate datasets from the uci machine learning repository (ensure it’s not cifar-10)
• download the dataset and prepare the data pipeline (e.g., loading, preprocessing, normalization)
• split the dataset into training (60%), validation (20%), and test (20%) sets

abithan: cnn architecture definition

• define the cnn architecture with the layers mentioned (convolutional, max-pooling, fully connected, dropout, softmax)
• decide the parameters (filter size, kernel size, activation functions, dropout rate) and justify them
• share initial architecture design with the team for feedback

person 4: literature review

• research activation functions (relu, softmax) and justify their use
• research optimizers (adam vs. sgd) and loss functions (sparse categorical crossentropy) for justification
• summarize findings for the team

deliverables by end of week 1:

• environment set up and shared repo
• get everyone upto speed with git and collaboration
• dataset selected, preprocessed, and split
• initial cnn architecture defined
• justification for activations, optimizers, and loss function prepared

week 2: cnn model implementation and training

person 1: cnn model implementation

• implement the cnn architecture in code (using tensorflow/keras or pytorch)
• ensure the model uses the adam optimizer and sparse categorical crossentropy loss function

person 2: training and validation

• train the cnn model for 20 epochs with a learning rate of 0.001
• plot training and validation loss for each epoch
• evaluate the initial performance on validation data

person 3: hyperparameter tuning

• run experiments with different learning rates (0.0001, 0.001, 0.01, 0.1)
• plot training and validation loss for different learning rates
• select the best learning rate with justification

person 4: testing and initial evaluation

• evaluate the trained cnn model on the test set
• record test accuracy, confusion matrix, precision, and recall
• analyze the results

deliverables by end of week 2:

• cnn model implemented and trained
• training and validation loss plots for different learning rates
• test accuracy, confusion matrix, precision, and recall recorded