Conducted personal research with my partner Naoya Kumagai over the course of several months on Convolutional Neural Networks (CNNs). There are subclasses of neural networks that are architected for the task of identifying and analyzing visual imagery. Our goal was to demonstrate a solid understanding of this variety of Neural Networks both mathematically and practically using the MNIST dataset. Additionally, we created a final report to summarize our process which followed a machine learning research paper model.
There are three main elements of this project:
-
The Presentation (best if viewed first for an overview of the project)
a. Introducing the historical background of CNNs
b. Their utility and abundance in the real world
c. Forward propagation of CNNs
d. Backpropagation of CNNs
e. Experiment and Results -
The Experiment
a. Created a CNN model using keras on a colab notebook
b. Model trained 10 times using the the MNIST dataset
c. Tested error of recognition using a 10,000 image training set
d. Modeled results with graphs -
The Research Report
a. Abstract and Introduction of CNNs
b. Mathematical foundation of CNNs
c. Our experiment
d. Conclusion
** Please view the extension of this research paper and experiment under my Computer Vision Repo:
https://github.com/moiseyalaev/Computer-Vision-Projects
Thanks!