AutoCl : A Visual Interactive System for Automatic Deep Learning Classifier Recommendation Based on Models Performance
This addresses the challenge for non-experts in adopting deep learning models, though it is incremental as it builds on existing AutoML systems.
The paper tackles the problem of non-experts struggling to select appropriate deep learning classifiers by introducing AutoCl, a visual interactive recommender system that automatically recommends the best classifier with hyperparameters, and demonstrates its capability through use cases on image classification datasets.
Nowadays, deep learning (DL) models being increasingly applied to various fields, people without technical expertise and domain knowledge struggle to find an appropriate model for their task. In this paper, we introduce AutoCl a visual interactive recommender system aimed at helping non-experts to adopt an appropriate DL classifier. Our system enables users to compare the performance and behavior of multiple classifiers trained with various hyperparameter setups as well as automatically recommends a best classifier with appropriate hyperparameter. We compare features of AutoCl against several recent AutoML systems and show that it helps non-experts better in choosing DL classifier. Finally, we demonstrate use cases for image classification using publicly available dataset to show the capability of our system.