digo: 생산성 향상을 위한 딥러닝 실험 관리 시스템

Abstract

Recently, advanced service using artificial intelligence has become a necessity, not an option. As a result, research on artificial intelligence has been accelerated, drawing attention to methods for efficient artificial intelligence research. A typical method is to use tools to effectively manage experiments in the course of the study. Existing deep learning studies have been inefficient due to collaboration based on fragmentary work methods and repetitive tasks for optimizing learning results. To improve these problems, this work designs and implements Digo (a combination of words that represent repetitive deep learning research as a compound word of dig and go), a collaborative-based deep learning experiment management tool that can provide a convenient and productive research environment, focusing on deep learning among artificial intelligence. Experiments and surveys were conducted on machine learning researchers to validate the performance of deep learning experimental management tools, and to confirm the convenience of hyperparameter automatic optimization and learning result visualization features.