Leveraging deep learning for computer vision in algorithm configuration-based AutoML
Automated machine learning; algorithm configuration; computer vision; deep learning; transfer learning.
Automated machine learning (AutoML) has been a field of great interest to both industry and academia. It has allowed developers working on increasingly common machine learning (ML) powered aplications with little to no ML expertise to achieve satisfactory results with little manpower investment. While AutoML tools have already proven themselves valuable in this way, their overall performance still often falls short of the state-of-the-art, especially in application domains where domain-specific algorithms are predominant, as is the case for the use of deep learning in computer vision (CV) tasks. In this work, we attempt to extend an existing AutoML tool, iSklearn, with transfer learning, leveraging the potential of deep learning for CV datasets. Preliminary results show much improved performance on these datasets, while also confirming insights previously observed in the deep learning literature in a new context.