CNN Activation Functions — IB Extended Essay
Researched the influence of activation functions on handwritten digit classification accuracy in convolutional neural networks.
Overview
An independent research project completed for the International Baccalaureate (IB) Diploma Programme in 2019.
The extended essay investigated how activation function selection influences prediction accuracy in convolutional neural networks (CNNs).
The final paper consists of 25 pages and approximately 3,655 words.
Research Question
To what extent can certain activation functions produce more accurate predictions of handwritten digits in a convolutional neural network?
Research
The project explored the relationship between neural network activation functions and handwritten digit classification.
Through this work, I gained early experience with:
- Convolutional Neural Networks: Understanding their architecture and application to image classification.
- Activation Functions: Investigating how activation function selection affects model predictions.
- Python and TensorFlow: Working with tools for neural network experimentation.
- Technical Research: Developing a research question, conducting experiments, and documenting findings.
Project Materials
The original Python and TensorFlow experiment code is no longer available. Screenshots from the experiments are preserved in the essay’s appendix.