View Position
| Project Title | A Science Gateway and Repository of Deep Learning Neural Network Models for Biomedical Image Processing |
| Summary | Built upon a successful BW Intern position 2015-16, interns will expand the work with more and larger datasets, and adding innovative models of deep learning neural networks, as well as constructing a science gateway to facilitate users from research community. In addition, we plan to design a workflow to enable containerize such deep learning models, and provide a repository service through Docker technology. |
| Job Description | Under guidance of the mentor, interns will: 1. In our last project, we successfully created a deep learning neural network model that has been proved effective in biomedical image segmentation. We have since discovered innovative applications in a wide range, such as satellite remote sensing, image registration and cancer classification, of this model. Interns will expand our work to make our model work with other data types, and create new models for new challenges in biomedical image processing area. 2. Based on a most recent study, most deep learning networks cannot take advantage of parallel computing on the CPU side. We plan to investigate this issue so that current technology such as the BlueWaters supercomputer can be more effectived used in big data analysis. 3. Evaluation of Deep Learning Neural Networks in Biomedical Image Processing. Identify and document DLNN best practices, parameter settings, use cases, etc. in the context of Biomedical Image Processing. 4. Assessment of new models of DLNN implementation including Docker and similar technologies to facilitate the deployment and maintenance of DLNN infrastructure. 5. Identify/Prepare/Create training data sets relevant to biomedical applications. 6. Benchmark DLNNs using well curated training sets and develop best practice recommendations for the Biomedical imaging community 7. Implement a science gateway for easy access to pre-built/pre-train DLNN models, training data sets, and best practice recommendations for Biomedical Image Processing |
| Use of Blue Waters | We plan to use BlueWaters in design and experiment with our deep learning neural networks models to perform training on biomedical data sets. 2. Based on a most recent study, most deep learning networks cannot take advantage of parallel computing on the CPU side. We plan to investigate this issue so that current technology such as the BlueWaters supercomputer can be more effectived used in big data analysis. |
| Conditions/Qualifications | In addition to the two students we have listed, we will be glad to open this opportunity to any students who are interested to study deep learning and parallel computing. The applicants must have: 1) basic math skills in statistics and linear algebra; 2) have taken introductory courses in artificial intelligence and machine learning; 3) can effectively communicate with researchers from other disciplinaries since we need to talk to experts from all fields. |
| Start Date | 06/01/2017 |
| End Date | 05/31/2018 |
| Location | Hood College of Frederick, Maryland. |
| Interns | Karen Canas
|