Title

Convolutional Neural Networks for Protein Image Classification

Author

Nick Littlefield, University of Southern MaineFollow

Date

Spring 2019

Document Type

Poster Session

Department

Computer Science

Advisor

Bruce MacLeod PhD

Keywords

protein patterns, Human Protein Atlas, convolutional neural networks

Abstract

A solution to the Kaggle competition: Human Protein Atlas Image Classification. Using microscopic images of cells provided by the Human Protein Atlas, convolutional neural networks, CNNs, were used to analyze and predict the location of protein patterns. Challenges included working with an unbalanced dataset, finding a correct learning rate, and choosing a correct architecture to solve the problem. To learn how to overcome these challenges and gain more understanding of the problem, various kernels and discussion posts for the competition, as well as papers on different CNN architectures were used.

Start Date

4-19-2019 1:15 PM

Recommended Citation

Littlefield, Nick, "Convolutional Neural Networks for Protein Image Classification" (2019). Thinking Matters Symposium Archive. 208.
https://digitalcommons.usm.maine.edu/thinking_matters/208