Skip to main content

DeepForge: A Visual Development Environment for Deep Learning

Primary Investigators:
Peter Volgyesi and Akos Ledeczi
 
Brief Description of Project:
DeepForge (https://deepforge.org),  a powerful hybrid textual-visual web-based programming platform, is designed to lower the barrier to entry and facilitate the rapid development of deep learning models. Utilizing a cloud-based infrastructure, DeepForge promotes reproducibility, ease of access, and supports remote execution of machine learning pipelines. The project involves the development of adapters  1.)  to access training data from public scientific datasets (OpenML, Kaggle,  etc.), 2.)  to retrieve existing models from public sources (e.g., Model Zoo); and 3.) for common deployment targets, i.e., the platforms that execute the computationally expensive training jobs, such as NERSC and DOE supercomputers, public cloud providers (e.g.  Amazon EC2 GPU instances).

Desired Qualifications: 
Deep Learning Special Topics Course
Familiarity with JavaScript and Tensorflow/Keras

Nature of Supervision:
Work closely with PIs
 
A Brief Research Plan (period is for 10 weeks):
Week 1: Learn DeepForge. Develop an example machine learning pipeline.
Week 2: Identify specific project area based on student interest. Learn relevant technologies.
Weeks 3-8: Design and develop new component for DeepForge
Weeks 9-10: Develop example machine learning pipeline using new feature.

Number of Open Slots: 2
 
Contact Information:
Name: Akos Ledeczi
Department: Electrical Engineering and Computer Science
Email: akos.ledeczi@vanderbilt.edu