Analyzing the Efficiency of Subsequent Convolutional Layers with Small-Scale Images
CSEF · 2018 Computational Systems & Analysis
Overview
Objectives/Goals My project attempts to find the optimal number of convolutional layers (a conv. layer teaches filters to recognize details) to place next to each other in order to improve the training efficiency of a neural network. Methods/Materials Using Python, four convolutional neural networks were trained on the CIFAR-10 dataset. Each model n had n conv. layers placed subsequently (otherwise, their architectures were the same). Each model was trained 5 times, running 100 loops over the training data, then assessed on its accuracy. The libraries used in this project were Keras (with TensorFlow as its backend), SciPy, Pandas, and Matplotlib. Results There was no significant improvement between the model that used blocks of three subsequent convolutional layers and blocks of four conv. layers. Conclusions/Discussion Neural networks using blocks of three convolutional layers trained the most efficiently. These results could have applications in feature detection with low resolution images.
Summary statement
I found the optimal number of convolutional layers (filters in a neural network that can be trained to detect features like edges) to place subsequently in order to improve training efficiency.
Help received
None. I designed and conducted the experiment myself.
Competition history
- CSEF 2018
Resources
Related projects
ISEF · 2018
Analyzing the Efficiency of Subsequent Convolutional Layers with Small-Scale Images
CSEF · 2016
Application of a Deep Learning Architecture Using Convolutional and Recurrent LSTM Networks to Video Classification
JSHS · 2023
Implementing Quantum-Classical Machine Learning Architectures to Optimize Convolutional Neural Networks
ISEF · 2024
Impacts of a Hidden Layer Design on CNN for Traffic Sign Recognition
CSEF · 2018
Analyzing 2D-Convolutional Neural Network's Ability to Learn Obstacle Avoidance in Self-Driving Applications
CSEF · 2004
Neurotic About Hidden Neurons
ISEF · 2020
Testing a Variety of Methods for Training a Neural Network to Determine Which One Achieves the Lowest Error Rate Fastest?
ISEF · 2019
Optimizing Cell Quantification in Biological Assays Using a Convolutional Neural Network
Closest projects by meaning, across every fair and year in the corpus.
Browse more like this
Source: California Science & Engineering Fair public projects